{ "edition": "7 November 2026", "method": "https://alexissantos.dev/ai-dev-tools#method", "categories": [ { "id": "terminal", "title": "Terminal agents", "tools": [ { "rank": 1, "name": "Claude Code", "url": "https://code.claude.com", "score": 100.0, "what": "Terminal coding agent from Anthropic", "verdict": "The default harness in HN setup threads, often run through Desktop, Web or T3 Code. Praised for UX, skills and remote control; complaints are usage burn, ignored memory and unwanted files.", "tags": [], "signals": { "hn_devs": 463, "hn_comments_12m": 32941, "hn_title_points": 78331, "lobsters": 6, "devto_posts": 1913, "github_stars": 154893, "downloads_month": 47376016, "tone": 0.46 } }, { "rank": 2, "name": "Codex CLI", "url": "https://developers.openai.com/codex", "score": 88.1, "what": "OpenAI's terminal and desktop coding agent", "verdict": "Usually the second subscription beside Claude, kept for review and second opinions. Credited with solid one-shots and fast-moving desktop features; dinged for hitting limits, bugs and loud default design taste.", "tags": [ "Open source", "Rising" ], "signals": { "hn_devs": 277, "hn_comments_12m": 15265, "hn_title_points": 26411, "lobsters": 0, "devto_posts": 231, "github_stars": 133573, "downloads_month": 95594449, "tone": 0.27 } }, { "rank": 3, "name": "OpenCode", "url": "https://opencode.ai", "score": 84.5, "what": "Open-source terminal coding agent, any model", "verdict": "The go-to open harness for DeepSeek, GLM or local models, often on the $10 OpenCode Go plan. Liked for per-agent model choice; a few moved on to Pi or T3 Code.", "tags": [ "Open source", "Rising" ], "signals": { "hn_devs": 136, "hn_comments_12m": 5264, "hn_title_points": 6953, "lobsters": 2, "devto_posts": 41, "github_stars": 225234, "downloads_month": 14195259, "tone": 0.46 } }, { "rank": 4, "name": "Pi", "url": "https://pi.dev", "score": 67.5, "what": "Minimal, hackable open-source coding agent harness", "verdict": "A lean harness people bend to their own workflows, often with local or open models and the oh-my-pi fork. Liked for small prompts that fit tight context windows.", "tags": [ "Open source", "Rising" ], "signals": { "hn_devs": 67, "hn_comments_12m": 375, "hn_title_points": 166, "lobsters": 0, "devto_posts": 6, "github_stars": 120619, "downloads_month": 1509211, "tone": 0.73 } }, { "rank": 5, "name": "DeepSeek Harness", "url": "https://deepseek.com/harness/en/", "score": 60.1, "what": "DeepSeek's own agent harness, desktop and TUI", "verdict": "A cheap home setup for many commenters, paired with DeepSeek Flash for a few dollars a month. Praised for speed, value and plugins; web search is the recurring weak spot.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 10, "hn_comments_12m": 499, "hn_title_points": 1265, "lobsters": 1, "devto_posts": 6, "github_stars": 275634, "downloads_month": null, "tone": 0.7 } }, { "rank": 6, "name": "Aider", "url": "https://aider.chat", "score": 57.6, "what": "Open-source terminal pair programmer, manual context", "verdict": "The pre-agent pair programmer, still liked by a few for manual file control and accuracy. Most commenters moved on, calling the manual context loading busywork next to agentic tools.", "tags": [ "Open source", "Fading" ], "signals": { "hn_devs": 64, "hn_comments_12m": 324, "hn_title_points": 512, "lobsters": 0, "devto_posts": 2, "github_stars": 50007, "downloads_month": 260509, "tone": 0.12 } }, { "rank": 7, "name": "Gemini CLI", "url": "https://github.com/google-gemini/gemini-cli", "score": 52.6, "what": "Google's open-source terminal agent for Gemini", "verdict": "Older threads used it as the free option and liked its web search and code explanations. By late 2026 commenters call it deprecated for subscribers and replaced by Antigravity CLI.", "tags": [ "Open source" ], "signals": { "hn_devs": 14, "hn_comments_12m": 1953, "hn_title_points": 4160, "lobsters": 0, "devto_posts": 97, "github_stars": 108766, "downloads_month": 1627950, "tone": -0.05 } }, { "rank": 8, "name": "Copilot CLI", "url": "https://github.com/features/copilot/cli", "score": 48.0, "what": "GitHub Copilot's terminal coding agent", "verdict": "Used where work already pays for GitHub Copilot; users say it runs close to Claude Code on the same models. Complaints target Autopilot mode, token burn and a thin plugin system.", "tags": [], "signals": { "hn_devs": 5, "hn_comments_12m": 657, "hn_title_points": 316, "lobsters": 0, "devto_posts": 18, "github_stars": 11375, "downloads_month": 6685193, "tone": 0.44 } }, { "rank": 9, "name": "Goose", "url": "https://block.github.io/goose", "score": 42.7, "what": "Block's open-source extensible agent, Linux Foundation", "verdict": "Block's open agent, now under the Linux Foundation, used for everyday tasks and custom bots. A few like its TUI; others found it too agentic or no better than prompting directly.", "tags": [ "Open source", "Rising" ], "signals": { "hn_devs": 10, "hn_comments_12m": 59, "hn_title_points": 485, "lobsters": 1, "devto_posts": 34, "github_stars": 56330, "downloads_month": null, "tone": -0.07 } }, { "rank": 10, "name": "Mistral Vibe", "url": "https://mistral.ai/news/devstral-2-vibe-cli", "score": 35.4, "what": "Mistral's open-source terminal coding agent", "verdict": "Mostly used as a cheap way to run hosted GLM-5.3. Fans call it fast and good enough; critics say the harness is sluggish, times out or ignores instructions.", "tags": [ "Open source" ], "signals": { "hn_devs": 2, "hn_comments_12m": 83, "hn_title_points": 779, "lobsters": 0, "devto_posts": 1, "github_stars": 5331, "downloads_month": null, "tone": 0.43 } }, { "rank": 11, "name": "Muse Code", "url": "https://ai.meta.com/muse/", "score": 35.0, "what": "Meta's terminal agent for Muse models", "verdict": "Usually a cheap extra subscription next to Claude or Codex. One user relies on it for code review; others note tight limits on the top model or say they barely touch it.", "tags": [], "signals": { "hn_devs": 4, "hn_comments_12m": 66, "hn_title_points": 365, "lobsters": 1, "devto_posts": 1, "github_stars": null, "downloads_month": null, "tone": 0.19 } }, { "rank": 12, "name": "Kimi CLI", "url": "https://github.com/MoonshotAI/kimi-cli", "score": 34.8, "what": "Moonshot's open-source terminal agent for Kimi", "verdict": "Moonshot's harness for Kimi models, used by Kimi subscribers. Commenters credit its checkpoint rollback for taming Kimi's tool-call loops; one critic called it a copy of basic harnesses.", "tags": [ "Open source" ], "signals": { "hn_devs": 3, "hn_comments_12m": 73, "hn_title_points": 8, "lobsters": 0, "devto_posts": 1, "github_stars": 11420, "downloads_month": null, "tone": 0.47 } }, { "rank": 13, "name": "Qwen Code", "url": "https://github.com/QwenLM/qwen-code", "score": 30.9, "what": "Alibaba's terminal agent for Qwen models", "verdict": "Mostly paired with local Qwen models on home hardware. One user built a 3D game with it in two days; others found local runs slow and error-prone next to Claude Code.", "tags": [ "Open source" ], "signals": { "hn_devs": 3, "hn_comments_12m": 675, "hn_title_points": 23, "lobsters": 0, "devto_posts": 2, "github_stars": 29072, "downloads_month": 385446, "tone": 0.2 } }, { "rank": 14, "name": "Zerostack", "url": "https://github.com/gi-dellav/zerostack", "score": 27.4, "what": "", "verdict": "", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 1, "hn_comments_12m": 26, "hn_title_points": 624, "lobsters": 0, "devto_posts": 0, "github_stars": 1745, "downloads_month": null, "tone": 0 } }, { "rank": 15, "name": "Codewhale", "url": "https://github.com/codewhale-hq/Codewhale", "score": 23.4, "what": "", "verdict": "", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 1, "hn_comments_12m": 13, "hn_title_points": 0, "lobsters": 0, "devto_posts": 0, "github_stars": 41161, "downloads_month": null, "tone": 0 } }, { "rank": 16, "name": "Amp", "url": "https://ampcode.com", "score": 22.6, "what": "", "verdict": "", "tags": [], "signals": { "hn_devs": 5, "hn_comments_12m": 43, "hn_title_points": 3, "lobsters": 0, "devto_posts": 1, "github_stars": null, "downloads_month": null, "tone": 0 } }, { "rank": 17, "name": "jcode", "url": "https://github.com/1jehuang/jcode", "score": 21.0, "what": "", "verdict": "", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 0, "hn_comments_12m": 17, "hn_title_points": 12, "lobsters": 0, "devto_posts": 0, "github_stars": 21591, "downloads_month": null, "tone": 0 } }, { "rank": 18, "name": "Plandex", "url": "https://plandex.ai", "score": 19.6, "what": "", "verdict": "", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 1, "hn_title_points": 261, "lobsters": 0, "devto_posts": 0, "github_stars": 15899, "downloads_month": null, "tone": 0 } }, { "rank": 19, "name": "Droid", "url": "https://factory.ai", "score": 17.7, "what": "", "verdict": "", "tags": [], "signals": { "hn_devs": 1, "hn_comments_12m": 20, "hn_title_points": 2, "lobsters": 1, "devto_posts": 1, "github_stars": null, "downloads_month": null, "tone": 0 } }, { "rank": 20, "name": "Crush", "url": "https://github.com/charmbracelet/crush", "score": 17.1, "what": "", "verdict": "", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 7, "hn_title_points": 0, "lobsters": 0, "devto_posts": 0, "github_stars": 28989, "downloads_month": null, "tone": 0 } }, { "rank": 21, "name": "Open Interpreter", "url": "https://github.com/openinterpreter/openinterpreter", "score": 15.6, "what": "", "verdict": "", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 1, "hn_title_points": 0, "lobsters": 0, "devto_posts": 0, "github_stars": 68825, "downloads_month": null, "tone": 0 } }, { "rank": 22, "name": "Kiro CLI", "url": "https://kiro.dev/cli/", "score": 13.8, "what": "", "verdict": "", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 18, "hn_title_points": 27, "lobsters": 0, "devto_posts": 7, "github_stars": null, "downloads_month": null, "tone": 0 } } ] }, { "id": "editors", "title": "Editors & IDEs", "tools": [ { "rank": 1, "name": "VS Code + Copilot", "url": "https://code.visualstudio.com/docs/copilot/overview", "score": 87.7, "what": "GitHub Copilot agent inside VS Code", "verdict": "Mostly the option work pays for. Praised for PR reviews, model choice and keep/undo diffs; others call it meh and dry next to Claude.", "tags": [ "Fading" ], "signals": { "hn_devs": 130, "hn_comments_12m": 1923, "hn_title_points": 5969, "lobsters": 7, "devto_posts": 183, "github_stars": null, "downloads_month": null, "tone": 0.04 } }, { "rank": 2, "name": "Cursor", "url": "https://cursor.com", "score": 86.9, "what": "AI-first VS Code fork with agents", "verdict": "Often described as where commenters used to be. Usage limits and API-priced bills keep pushing people to a Claude subscription; a few still like its bundled plans.", "tags": [ "Fading" ], "signals": { "hn_devs": 244, "hn_comments_12m": 7647, "hn_title_points": 25602, "lobsters": 3, "devto_posts": 247, "github_stars": null, "downloads_month": null, "tone": -0.2 } }, { "rank": 3, "name": "Zed", "url": "https://zed.dev", "score": 83.7, "what": "Fast Rust editor with ACP agent support", "verdict": "The editor many keep beside their agents, driving Claude or Codex over ACP or just reviewing code. Liked for speed and agent support; some found it overhyped or unusable without a GPU.", "tags": [ "Open source" ], "signals": { "hn_devs": 70, "hn_comments_12m": 2405, "hn_title_points": 11616, "lobsters": 1, "devto_posts": 8, "github_stars": 94166, "downloads_month": null, "tone": 0.31 } }, { "rank": 4, "name": "Antigravity", "url": "https://antigravity.google", "score": 69.0, "what": "Google's agent IDE and CLI for Gemini", "verdict": "The only subscription route to Gemini. Some use it daily for search and planning; complaints center on billing and account friction and a weak harness.", "tags": [], "signals": { "hn_devs": 29, "hn_comments_12m": 1584, "hn_title_points": 5200, "lobsters": 0, "devto_posts": 90, "github_stars": null, "downloads_month": null, "tone": 0.23 } }, { "rank": 5, "name": "Windsurf", "url": "https://windsurf.com", "score": 64.0, "what": "AI VS Code fork, team now at Google", "verdict": "Mostly mentioned in the past tense now that its team went to Google. Older threads liked its codebase Q&A, autocomplete and parallel flows, but said it struggled on large files.", "tags": [ "Fading" ], "signals": { "hn_devs": 62, "hn_comments_12m": 379, "hn_title_points": 5115, "lobsters": 0, "devto_posts": 16, "github_stars": null, "downloads_month": null, "tone": 0.08 } }, { "rank": 6, "name": "Cline", "url": "https://cline.bot", "score": 63.9, "what": "Open-source agent extension for VS Code", "verdict": "Popular for pairing cheap API models like DeepSeek Flash, or local Ollama models, with VS Code. Liked for permission prompts and provider choice; several left after finding it slow and unstable.", "tags": [ "Open source", "Fading" ], "signals": { "hn_devs": 35, "hn_comments_12m": 325, "hn_title_points": 879, "lobsters": 0, "devto_posts": 10, "github_stars": 71447, "downloads_month": null, "tone": 0.12 } }, { "rank": 7, "name": "Kiro", "url": "https://kiro.dev", "score": 56.7, "what": "Amazon's spec-driven agentic VS Code fork", "verdict": "Mostly a work-mandated tool at AWS shops. Its spec-driven workflow gets borrowed elsewhere; Amazon engineers say the harness trails Claude Code and degrades over long context.", "tags": [ "Rising" ], "signals": { "hn_devs": 13, "hn_comments_12m": 433, "hn_title_points": 1797, "lobsters": 0, "devto_posts": 98, "github_stars": null, "downloads_month": null, "tone": 0.16 } }, { "rank": 8, "name": "Roo Code", "url": "https://roocode.com", "score": 48.8, "what": "Cline fork for VS Code, now discontinued", "verdict": "Ran on your own API key or local models inside VS Code. Users liked its control, visible costs and architect and orchestrator modes; it has been discontinued, with forks carrying on.", "tags": [ "Open source" ], "signals": { "hn_devs": 13, "hn_comments_12m": 72, "hn_title_points": 76, "lobsters": 0, "devto_posts": 2, "github_stars": 24240, "downloads_month": null, "tone": 0.46 } }, { "rank": 9, "name": "Kilo Code", "url": "https://kilocode.ai", "score": 46.6, "what": "Roo Code fork for VS Code, JetBrains", "verdict": "Run on your own keys or Kilo credits across many models. Users say it gets the job done; gripes are weak compaction and a recent version that feels blander.", "tags": [ "Open source" ], "signals": { "hn_devs": 6, "hn_comments_12m": 69, "hn_title_points": 169, "lobsters": 0, "devto_posts": 0, "github_stars": 27898, "downloads_month": null, "tone": 0.4 } }, { "rank": 10, "name": "Neovim (CodeCompanion / avante)", "url": "https://github.com/olimorris/codecompanion.nvim", "score": 46.2, "what": "AI chat and edit plugins for Neovim", "verdict": "Neovim users reach for CodeCompanion for in-editor chat and quick edits, handing bigger tasks to OpenCode or Claude Code. Local-model setups can loop, and completion trails Cursor Tab.", "tags": [ "Open source" ], "signals": { "hn_devs": 38, "hn_comments_12m": 12, "hn_title_points": 1, "lobsters": 1, "devto_posts": 13, "github_stars": 6927, "downloads_month": null, "tone": 0.44 } }, { "rank": 11, "name": "JetBrains Junie", "url": "https://www.jetbrains.com/junie/", "score": 44.8, "what": "JetBrains' coding agent inside its IDEs", "verdict": "Popular with IntelliJ users, often paired with cheap Gemini Flash. Good for small, well-defined tasks and tests; quotas are tight and some rate the harness below Claude Code.", "tags": [], "signals": { "hn_devs": 12, "hn_comments_12m": 65, "hn_title_points": 101, "lobsters": 1, "devto_posts": 1, "github_stars": null, "downloads_month": null, "tone": 0.62 } }, { "rank": 12, "name": "Emacs gptel", "url": "https://github.com/karthink/gptel", "score": 40.9, "what": "LLM client and agent toolkit for Emacs", "verdict": "Built for human-in-the-loop work: chat, code review and questions inside Emacs. Praised for staying out of the way and tiny token bills; agentic features are mostly do-it-yourself.", "tags": [ "Open source" ], "signals": { "hn_devs": 6, "hn_comments_12m": 57, "hn_title_points": 19, "lobsters": 0, "devto_posts": 0, "github_stars": 3563, "downloads_month": null, "tone": 0.53 } }, { "rank": 13, "name": "Continue", "url": "https://continue.dev", "score": 39.6, "what": "Open-source assistant for VS Code and JetBrains", "verdict": "Was the usual way to plug Ollama or LM Studio models into VS Code. Commenters report it was acquired by Cursor in 2026, and recent mentions are mostly about its future.", "tags": [ "Open source", "Fading" ], "signals": { "hn_devs": 10, "hn_comments_12m": 27, "hn_title_points": 0, "lobsters": 0, "devto_posts": 1, "github_stars": 36427, "downloads_month": null, "tone": 0.18 } }, { "rank": 14, "name": "Whiteboard", "url": "https://github.com/devdotfast/whiteboard", "score": 24.8, "what": "", "verdict": "", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 0, "hn_comments_12m": 6, "hn_title_points": 452, "lobsters": 0, "devto_posts": 1, "github_stars": 3463, "downloads_month": null, "tone": 0 } }, { "rank": 15, "name": "Void", "url": "https://voideditor.com", "score": 21.7, "what": "", "verdict": "", "tags": [ "Open source" ], "signals": { "hn_devs": 1, "hn_comments_12m": 0, "hn_title_points": 32, "lobsters": 0, "devto_posts": 0, "github_stars": 28713, "downloads_month": null, "tone": 0 } } ] }, { "id": "orchestration", "title": "Running many agents", "tools": [ { "rank": 1, "name": "Herdr", "url": "https://herdr.dev", "score": 83.6, "what": "Terminal workspace for many agent sessions", "verdict": "Runs several Claude Code, Codex or Pi sessions side by side. Loved for agent-done pings and a pane overview; a few never clicked with it or switched to Rex.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 41, "hn_comments_12m": 552, "hn_title_points": 1346, "lobsters": 0, "devto_posts": 3, "github_stars": 49030, "downloads_month": null, "tone": 0.42 } }, { "rank": 2, "name": "tmux + git worktrees", "url": "https://git-scm.com/docs/git-worktree", "score": 72.9, "what": "DIY parallel agents with tmux, worktrees", "verdict": "The do-it-yourself baseline: a worktree per task, a tmux pane per agent, send-keys for messages. Fans say the new agent IDEs only repackage it; many build their own wrappers on top.", "tags": [ "Rising" ], "signals": { "hn_devs": 109, "hn_comments_12m": 10, "hn_title_points": 0, "lobsters": 2, "devto_posts": 53, "github_stars": null, "downloads_month": null, "tone": 0.42 } }, { "rank": 3, "name": "Paseo", "url": "https://github.com/getpaseo/paseo", "score": 65.2, "what": "Open-source desktop and mobile agent interface", "verdict": "A front end for Pi, Claude Code, Codex and OpenCode agents. Liked for its phone app and client-server setup; reports of memory strain and hung sub-agent status updates.", "tags": [ "Rising" ], "signals": { "hn_devs": 11, "hn_comments_12m": 175, "hn_title_points": 127, "lobsters": 0, "devto_posts": 1, "github_stars": 23394, "downloads_month": null, "tone": 0.43 } }, { "rank": 4, "name": "Gas Town", "url": "https://github.com/gastownhall/gastown", "score": 61.8, "what": "Steve Yegge's multi-agent coding orchestrator", "verdict": "Steve Yegge's many-agent orchestrator. On HN it is mostly a punchline: too complex, too opinionated and a token sink, with nobody showing evals against a plain harness.", "tags": [ "Open source" ], "signals": { "hn_devs": 11, "hn_comments_12m": 609, "hn_title_points": 1532, "lobsters": 0, "devto_posts": 1, "github_stars": 18725, "downloads_month": null, "tone": -0.52 } }, { "rank": 5, "name": "T3 Code", "url": "https://t3.codes", "score": 60.5, "what": "Open-source GUI driving multiple agent harnesses", "verdict": "Drives Claude Code, Codex and Pi from one UI across machines. Fans praise mid-conversation agent handoff and remote sessions; others found it buggy or heavy on resources.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 10, "hn_comments_12m": 64, "hn_title_points": 32, "lobsters": 0, "devto_posts": 3, "github_stars": 50869, "downloads_month": null, "tone": 0.65 } }, { "rank": 6, "name": "Beads", "url": "https://github.com/steveyegge/beads", "score": 60.3, "what": "Git-backed issue tracker for coding agents", "verdict": "Gives agents epics, tasks and dependencies to pull from, and some teams run their whole backlog on it. Critics call it bloated and question why it needs a daemon and database.", "tags": [ "Open source" ], "signals": { "hn_devs": 10, "hn_comments_12m": 459, "hn_title_points": 338, "lobsters": 0, "devto_posts": 4, "github_stars": 28572, "downloads_month": null, "tone": 0.48 } }, { "rank": 7, "name": "Orca", "url": "https://github.com/stablyai/orca", "score": 51.2, "what": "Open-source agent IDE for worktrees, remote", "verdict": "Often self-hosted, with a mobile app for steering agents away from the desk. Praised for remote SSH and multi-worktree views; one user hit install trouble on a headless server.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 12, "hn_comments_12m": 9, "hn_title_points": 0, "lobsters": 0, "devto_posts": 2, "github_stars": 118432, "downloads_month": null, "tone": 0.52 } }, { "rank": 8, "name": "Conductor", "url": "https://conductor.build", "score": 49.5, "what": "Mac app for parallel worktree agents, closing", "verdict": "Ran Claude and Codex in parallel git worktrees, and recent threads say it is winding down. Users liked it for mid-size tasks and review; gripes were weak multi-repo support and Mac-only builds.", "tags": [ "Rising" ], "signals": { "hn_devs": 14, "hn_comments_12m": 42, "hn_title_points": 5, "lobsters": 3, "devto_posts": 2, "github_stars": null, "downloads_month": null, "tone": 0.09 } }, { "rank": 9, "name": "Juggler", "url": "https://github.com/juggler-ai/juggler", "score": 39.7, "what": "GUI coding agent from JUCE's creator", "verdict": "Treats local and remote sessions the same, which suits cross-platform work. Users like the control and visibility; one found its nested output stressful, another wished it were less vibecoded.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 2, "hn_comments_12m": 98, "hn_title_points": 330, "lobsters": 0, "devto_posts": 0, "github_stars": 853, "downloads_month": null, "tone": 0.27 } }, { "rank": 10, "name": "Ralph loop", "url": "https://ghuntley.com/ralph/", "score": 39.2, "what": "Loop-until-done agent prompting pattern", "verdict": "A pattern, not a product: rerun the agent until the task list says done. Used for overnight PR polishing and test coverage; skeptics call it token burning that hides broken tests.", "tags": [], "signals": { "hn_devs": 3, "hn_comments_12m": 51, "hn_title_points": 72, "lobsters": 0, "devto_posts": 8, "github_stars": null, "downloads_month": null, "tone": 0.07 } }, { "rank": 11, "name": "Vibe Kanban", "url": "https://www.vibekanban.com", "score": 36.8, "what": "Kanban board that dispatches coding agents", "verdict": "Users liked managing agents across projects from one board, but its developers stopped investing in it, and some felt it added little to their productivity.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 29, "hn_title_points": 199, "lobsters": 0, "devto_posts": 0, "github_stars": 28613, "downloads_month": 14853, "tone": 0.25 } }, { "rank": 12, "name": "Omnara", "url": "https://github.com/omnara-ai/omnara", "score": 33.7, "what": "Remote and API layer for coding agents", "verdict": "Began as phone and web remote control for Claude Code and Codex, now pitched as a self-hostable agent API. One user relies on it in transit; another found alternatives more reliable.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 49, "hn_title_points": 571, "lobsters": 0, "devto_posts": 0, "github_stars": 3365, "downloads_month": null, "tone": 0.0 } }, { "rank": 13, "name": "Munder Difflin", "url": "https://munderdiffl.in", "score": 30.7, "what": "", "verdict": "", "tags": [], "signals": { "hn_devs": 2, "hn_comments_12m": 8, "hn_title_points": 313, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": 0 } }, { "rank": 14, "name": "workmux", "url": "https://github.com/raine/workmux", "score": 30.0, "what": "", "verdict": "", "tags": [ "Open source" ], "signals": { "hn_devs": 1, "hn_comments_12m": 16, "hn_title_points": 12, "lobsters": 0, "devto_posts": 0, "github_stars": 3017, "downloads_month": null, "tone": 0 } }, { "rank": 15, "name": "qm", "url": "https://github.com/yc-software/qm", "score": 28.9, "what": "", "verdict": "", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 0, "hn_comments_12m": 4, "hn_title_points": 684, "lobsters": 0, "devto_posts": 0, "github_stars": 15650, "downloads_month": null, "tone": 0 } }, { "rank": 16, "name": "Ruflo", "url": "https://github.com/ruvnet/ruflo", "score": 25.0, "what": "", "verdict": "", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 6, "hn_title_points": 7, "lobsters": 0, "devto_posts": 1, "github_stars": 81151, "downloads_month": null, "tone": 0 } }, { "rank": 17, "name": "Superset", "url": "https://superset.sh", "score": 22.6, "what": "", "verdict": "", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 19, "hn_title_points": 0, "lobsters": 0, "devto_posts": 0, "github_stars": 16066, "downloads_month": null, "tone": 0 } }, { "rank": 18, "name": "Claude Squad", "url": "https://github.com/smtg-ai/claude-squad", "score": 21.8, "what": "", "verdict": "", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 5, "hn_title_points": 9, "lobsters": 0, "devto_posts": 0, "github_stars": 8745, "downloads_month": null, "tone": 0 } }, { "rank": 19, "name": "oh-my-claudecode", "url": "https://github.com/Yeachan-Heo/oh-my-claudecode", "score": 19.8, "what": "", "verdict": "", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 0, "hn_comments_12m": 2, "hn_title_points": 2, "lobsters": 0, "devto_posts": 0, "github_stars": 40664, "downloads_month": null, "tone": 0 } }, { "rank": 20, "name": "Sculptor", "url": "https://imbue.com/sculptor/", "score": 19.2, "what": "Imbue's desktop UI for Claude Code", "verdict": "A desktop UI for running Claude Code agents. The scraped comments are almost all about actual sculptors, so there are no recent user reports to summarize.", "tags": [], "signals": { "hn_devs": 1, "hn_comments_12m": 99, "hn_title_points": 341, "lobsters": 0, "devto_posts": 1, "github_stars": null, "downloads_month": null, "tone": 0 } }, { "rank": 21, "name": "Crystal", "url": "https://github.com/stravu/crystal", "score": 11.0, "what": "", "verdict": "", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 0, "hn_title_points": 0, "lobsters": 0, "devto_posts": 0, "github_stars": 3129, "downloads_month": null, "tone": 0 } }, { "rank": 22, "name": "AX (Agent Executor)", "url": "https://agentexecutor.io", "score": 2.6, "what": "", "verdict": "", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 1, "hn_title_points": 0, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": 0 } } ] }, { "id": "review", "title": "Code review & PR agents", "tools": [ { "rank": 1, "name": "Claude Code review", "url": "https://code.claude.com/docs/en/code-review", "score": 57.6, "what": "Anthropic's managed PR review for Claude Code", "verdict": "Used as a PR pass alongside Codex or as a local /review skill. The recurring complaint is price, quoted at $15 to $25 per review, plus slowness and reviews that need re-checking.", "tags": [ "Open source", "Rising" ], "signals": { "hn_devs": 13, "hn_comments_12m": 552, "hn_title_points": 29, "lobsters": 0, "devto_posts": 12, "github_stars": 9827, "downloads_month": null, "tone": -0.31 } }, { "rank": 2, "name": "Copilot code review", "url": "https://docs.github.com/en/copilot/concepts/agents/code-review", "score": 55.1, "what": "GitHub's built-in AI pull request reviewer", "verdict": "The low-friction pick for teams already on GitHub; several call its PR comments surprisingly insightful. Complaints center on noisy, overeager comments and the 2026 change billing reviews against Actions minutes.", "tags": [], "signals": { "hn_devs": 3, "hn_comments_12m": 75, "hn_title_points": 353, "lobsters": 0, "devto_posts": 1, "github_stars": null, "downloads_month": null, "tone": -0.13 } }, { "rank": 3, "name": "CodeRabbit", "url": "https://coderabbit.ai", "score": 45.9, "what": "AI pull request review bot", "verdict": "The name people reach for when describing an AI PR reviewer, usually as the reference point. Few first-hand endorsements here; several teams say they had gripes or built their own instead.", "tags": [], "signals": { "hn_devs": 1, "hn_comments_12m": 149, "hn_title_points": 773, "lobsters": 0, "devto_posts": 3, "github_stars": null, "downloads_month": null, "tone": -0.08 } }, { "rank": 4, "name": "Cursor Bugbot", "url": "https://cursor.com/bugbot", "score": 42.4, "what": "Cursor's AI bug-finding PR reviewer", "verdict": "The most praised review bot in these threads; users say it finds subtle bugs Claude and Codex miss. Downsides: inconsistent across rounds, and a move to per-run pricing drove some away.", "tags": [], "signals": { "hn_devs": 4, "hn_comments_12m": 25, "hn_title_points": 3, "lobsters": 0, "devto_posts": 2, "github_stars": null, "downloads_month": null, "tone": 0.44 } }, { "rank": 5, "name": "Codex code review", "url": "https://developers.openai.com/codex/integrations/github", "score": 39.3, "what": "OpenAI Codex reviewing PRs and diffs", "verdict": "Popular as the cross-vendor second opinion: write with Claude, review with Codex. Praised for catching real bugs; criticized for endless nitpick loops and an opaque separate quota.", "tags": [], "signals": { "hn_devs": 1, "hn_comments_12m": 560, "hn_title_points": 14, "lobsters": 0, "devto_posts": 1, "github_stars": null, "downloads_month": null, "tone": 0.18 } }, { "rank": 6, "name": "Codex Security", "url": "https://openai.com/index/codex-security/", "score": 34.6, "what": "OpenAI's open-source security review CLI", "verdict": "OpenAI's open-source security scanning CLI. The launch thread, answered by the Promptfoo team, is dominated by model refusals on defensive security work and slow access for open-source maintainers.", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 112, "hn_title_points": 662, "lobsters": 1, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": -0.17 } }, { "rank": 7, "name": "Qodo", "url": "https://www.qodo.ai", "score": 30.1, "what": "AI code review and PR agent", "verdict": "An AI PR reviewer that teams try and often drop over noise. One user likes it for speeding up human review rather than replacing it, and Bugbot gives more quota per dollar.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 14, "hn_title_points": 204, "lobsters": 0, "devto_posts": 0, "github_stars": 15406, "downloads_month": null, "tone": -0.33 } }, { "rank": 8, "name": "Graphite", "url": "https://graphite.com", "score": 28.8, "what": "Stacked PR tool with AI code review", "verdict": "Mostly known for stacked PRs, with an AI review agent on top, and now part of Cursor. The one real user here credits stacks with keeping AI-era PRs small.", "tags": [], "signals": { "hn_devs": 2, "hn_comments_12m": 359, "hn_title_points": 981, "lobsters": 0, "devto_posts": 2, "github_stars": null, "downloads_month": null, "tone": 0 } }, { "rank": 9, "name": "Greptile", "url": "https://greptile.com", "score": 26.6, "what": "Codebase-aware AI pull request reviewer", "verdict": "A PR review bot that shows up in several stacks next to Codex and Claude Code. One user found few false positives; others found it noisy or pricey like its rivals.", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 64, "hn_title_points": 90, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": 0.25 } }, { "rank": 10, "name": "Open Code Review (Alibaba)", "url": "https://github.com/alibaba/open-code-review", "score": 24.6, "what": "Alibaba's open-source CLI code review tool", "verdict": "Alibaba's open-sourced internal code review CLI. Commenters see the value in its curated rule files, which originally shipped in Chinese and prompted community translations; first-hand usage reports are scarce.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 0, "hn_comments_12m": 9, "hn_title_points": 0, "lobsters": 0, "devto_posts": 0, "github_stars": 49613, "downloads_month": null, "tone": 0.22 } }, { "rank": 11, "name": "Sourcery", "url": "https://sourcery.ai", "score": 19.1, "what": "AI code review bot for PRs", "verdict": "A PR review bot recommended alongside CodeRabbit for catching contract violations and antipatterns. Most support comes from one repeat commenter; a team user calls it chatty but improving.", "tags": [], "signals": { "hn_devs": 1, "hn_comments_12m": 15, "hn_title_points": 10, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": 0.6 } }, { "rank": 12, "name": "Sphinx", "url": "https://sphinx-six-theta.vercel.app/", "score": 11.5, "what": "Hook that quizzes you on AI code", "verdict": "A Show HN hook that quizzes you on your AI's code before you ship it. The launch drew 253 points, but no user comments about it turned up.", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 0, "hn_title_points": 253, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": 0 } } ] }, { "id": "context", "title": "Context, MCP & skills", "tools": [ { "rank": 1, "name": "AGENTS.md", "url": "https://agents.md", "score": 94.1, "what": "Open instruction file format for coding agents", "verdict": "The default home for project rules; nearly every setup described here has one, sometimes as CLAUDE.md. Pain points are files that keep growing and instructions agents still ignore.", "tags": [ "Fading" ], "signals": { "hn_devs": 90, "hn_comments_12m": 2004, "hn_title_points": 4010, "lobsters": 0, "devto_posts": 161, "github_stars": null, "downloads_month": null, "tone": 0.42 } }, { "rank": 2, "name": "Agent Skills", "url": "https://agentskills.io", "score": 81.0, "what": "Open format for reusable agent instruction packs", "verdict": "Widely used for project-specific workflows and teaching agents in-house tools. A vocal minority says too many skills pollute context and that stronger models increasingly do fine without them.", "tags": [ "Rising" ], "signals": { "hn_devs": 26, "hn_comments_12m": 1741, "hn_title_points": 233, "lobsters": 1, "devto_posts": 122, "github_stars": null, "downloads_month": null, "tone": 0.24 } }, { "rank": 3, "name": "Superpowers", "url": "https://github.com/obra/superpowers", "score": 56.5, "what": "Skills library for structured Claude Code workflows", "verdict": "A Claude Code skills pack that walks the agent through brainstorm, design doc, plan, then build. Fans like the structure; skeptics say current models do fine run raw.", "tags": [ "Open source" ], "signals": { "hn_devs": 10, "hn_comments_12m": 909, "hn_title_points": 1746, "lobsters": 2, "devto_posts": 15, "github_stars": 313991, "downloads_month": null, "tone": 0.2 } }, { "rank": 4, "name": "Spec Kit", "url": "https://github.com/github/spec-kit", "score": 52.4, "what": "GitHub's spec-driven development toolkit", "verdict": "GitHub's spec-driven development kit, used to force requirements, plan and tasks before code. The main complaint is volume: reams of repetitive markdown that are tedious to review.", "tags": [ "Open source" ], "signals": { "hn_devs": 4, "hn_comments_12m": 138, "hn_title_points": 209, "lobsters": 0, "devto_posts": 6, "github_stars": 148358, "downloads_month": null, "tone": -0.06 } }, { "rank": 5, "name": "Context7", "url": "https://context7.com", "score": 52.3, "what": "MCP server for up-to-date library docs", "verdict": "One of the standard MCP servers people install, used to pull current library docs so the agent stops guessing APIs. No complaints turned up in these threads.", "tags": [ "Open source" ], "signals": { "hn_devs": 6, "hn_comments_12m": 92, "hn_title_points": 28, "lobsters": 0, "devto_posts": 7, "github_stars": 64316, "downloads_month": 3176334, "tone": 0.72 } }, { "rank": 6, "name": "GitHub MCP server", "url": "https://github.com/github/github-mcp-server", "score": 47.4, "what": "Official MCP server for GitHub APIs", "verdict": "Used to let agents read issues, search dependency code and open PRs, even from ChatGPT. The open questions are whether it beats the gh CLI and how it ties into Copilot billing.", "tags": [ "Open source" ], "signals": { "hn_devs": 3, "hn_comments_12m": 56, "hn_title_points": 28, "lobsters": 0, "devto_posts": 7, "github_stars": 34301, "downloads_month": null, "tone": 0.36 } }, { "rank": 7, "name": "Semble", "url": "https://github.com/MinishLab/semble", "score": 37.4, "what": "Token-efficient code search for agents", "verdict": "A code search tool for agents that claims far fewer tokens than grep. Two users keep it beside ripgrep and Serena; a skeptic says models aren't trained to use such tools.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 9, "hn_comments_12m": 58, "hn_title_points": 464, "lobsters": 0, "devto_posts": 15, "github_stars": 6377, "downloads_month": null, "tone": 0.33 } }, { "rank": 8, "name": "rtk", "url": "https://github.com/rtk-ai/rtk", "score": 36.0, "what": "CLI proxy that compresses agent command output", "verdict": "A shell wrapper that trims command output before it reaches the agent's context. Its own stats claim big savings, but independent benchmarks found agents take more turns and save little.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 0, "hn_comments_12m": 64, "hn_title_points": 339, "lobsters": 0, "devto_posts": 3, "github_stars": 85816, "downloads_month": null, "tone": -0.08 } }, { "rank": 9, "name": "Impeccable", "url": "https://github.com/pbakaus/impeccable", "score": 35.9, "what": "Design skill to de-slop AI frontends", "verdict": "A design skill for stripping the generic AI look from frontends. Users say it removes slop tells more aggressively than rivals, though output is a starting point and token use is high.", "tags": [ "Open source" ], "signals": { "hn_devs": 1, "hn_comments_12m": 190, "hn_title_points": 455, "lobsters": 0, "devto_posts": 5, "github_stars": 105331, "downloads_month": null, "tone": 0.33 } }, { "rank": 10, "name": "Graphify", "url": "https://github.com/Graphify-Labs/graphify", "score": 31.8, "what": "Codebase knowledge graph for coding agents", "verdict": "Maps a codebase into a knowledge graph so agents spend fewer tokens. Mostly cited as the baseline rivals claim to beat; one tester found models can't use its output well.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 0, "hn_comments_12m": 31, "hn_title_points": 134, "lobsters": 0, "devto_posts": 4, "github_stars": 143885, "downloads_month": null, "tone": 0.0 } }, { "rank": 11, "name": "Figma MCP", "url": "https://www.figma.com/developers/mcp", "score": 31.5, "what": "Figma's MCP server for design-to-code", "verdict": "Turns Figma designs into code; two users credit it with faster, pixel-accurate page builds. Recent threads call it close to unusable and fault Figma for gating access to approved clients.", "tags": [], "signals": { "hn_devs": 2, "hn_comments_12m": 49, "hn_title_points": 26, "lobsters": 0, "devto_posts": 5, "github_stars": null, "downloads_month": null, "tone": -0.29 } }, { "rank": 12, "name": "Repomix", "url": "https://repomix.com", "score": 31.4, "what": "", "verdict": "", "tags": [ "Open source" ], "signals": { "hn_devs": 2, "hn_comments_12m": 16, "hn_title_points": 9, "lobsters": 0, "devto_posts": 1, "github_stars": 29177, "downloads_month": 401454, "tone": 0 } }, { "rank": 13, "name": "claude-mem", "url": "https://github.com/thedotmack/claude-mem", "score": 29.4, "what": "", "verdict": "", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 22, "hn_title_points": 10, "lobsters": 0, "devto_posts": 7, "github_stars": 114348, "downloads_month": null, "tone": 0 } }, { "rank": 14, "name": "Linear MCP", "url": "https://linear.app/docs/mcp", "score": 29.0, "what": "", "verdict": "", "tags": [], "signals": { "hn_devs": 4, "hn_comments_12m": 13, "hn_title_points": 5, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": 0 } }, { "rank": 15, "name": "Serena", "url": "https://github.com/oraios/serena", "score": 28.7, "what": "MCP server exposing LSP code tools", "verdict": "An MCP server that gives agents language-server tools for symbol lookup, indexing and refactors. The few real users say it works well, including inside OpenCode and JetBrains IDEs.", "tags": [], "signals": { "hn_devs": 1, "hn_comments_12m": 53, "hn_title_points": 11, "lobsters": 0, "devto_posts": 2, "github_stars": 31180, "downloads_month": 147731, "tone": 0.71 } }, { "rank": 16, "name": "Sentry MCP", "url": "https://docs.sentry.io/product/sentry-mcp/", "score": 23.0, "what": "", "verdict": "", "tags": [], "signals": { "hn_devs": 1, "hn_comments_12m": 5, "hn_title_points": 6, "lobsters": 0, "devto_posts": 1, "github_stars": 958, "downloads_month": null, "tone": 0 } }, { "rank": 17, "name": "codebase-memory-mcp", "url": "https://github.com/DeusData/codebase-memory-mcp", "score": 22.6, "what": "", "verdict": "", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 0, "hn_comments_12m": 11, "hn_title_points": 0, "lobsters": 0, "devto_posts": 4, "github_stars": 48287, "downloads_month": null, "tone": 0 } }, { "rank": 18, "name": "headroom", "url": "https://github.com/headroomlabs-ai/headroom", "score": 22.6, "what": "", "verdict": "", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 0, "hn_comments_12m": 4, "hn_title_points": 16, "lobsters": 0, "devto_posts": 1, "github_stars": 79068, "downloads_month": null, "tone": 0 } }, { "rank": 19, "name": "context-mode", "url": "https://github.com/mksglu/context-mode", "score": 20.5, "what": "Plugin that shrinks MCP output in context", "verdict": "A plugin that shrinks bulky MCP tool output before it hits the agent's context window. Only three real mentions, each from someone who keeps it in their standard setup.", "tags": [ "New in 2026" ], "signals": { "hn_devs": 0, "hn_comments_12m": 106, "hn_title_points": 104, "lobsters": 0, "devto_posts": 2, "github_stars": 33644, "downloads_month": null, "tone": 0 } } ] }, { "id": "browser", "title": "Browser & web for agents", "tools": [ { "rank": 1, "name": "Playwright MCP", "url": "https://github.com/microsoft/playwright-mcp", "score": 90.8, "what": "Microsoft's MCP server for browser automation", "verdict": "The default way to give an agent a browser, used daily for UI checks and scraping. Works out of the box; the complaints are slowness and heavy context use.", "tags": [ "Open source" ], "signals": { "hn_devs": 13, "hn_comments_12m": 270, "hn_title_points": 495, "lobsters": 1, "devto_posts": 84, "github_stars": 39303, "downloads_month": 30936316, "tone": 0.23 } }, { "rank": 2, "name": "Chrome DevTools MCP", "url": "https://github.com/ChromeDevTools/chrome-devtools-mcp", "score": 58.2, "what": "Official MCP server for Chrome DevTools", "verdict": "Chosen over Playwright when the agent needs real DevTools: network requests, performance traces, Lighthouse audits. Users run it for automated QA and debugging; one says it clutters context.", "tags": [ "Open source" ], "signals": { "hn_devs": 2, "hn_comments_12m": 44, "hn_title_points": 659, "lobsters": 0, "devto_posts": 15, "github_stars": 54930, "downloads_month": 8604227, "tone": 0.53 } }, { "rank": 3, "name": "agent-browser", "url": "https://github.com/vercel-labs/agent-browser", "score": 48.8, "what": "Vercel's browser automation CLI for agents", "verdict": "Vercel's browser automation CLI for coding agents, often benchmarked against Playwright MCP. Users say it works fine, and its bundled skill turns up issues they would not think to test.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 1, "hn_comments_12m": 94, "hn_title_points": 99, "lobsters": 0, "devto_posts": 5, "github_stars": 46631, "downloads_month": 8262982, "tone": 0.36 } }, { "rank": 4, "name": "Firecrawl", "url": "https://firecrawl.dev", "score": 40.5, "what": "Web scraping and search API for LLMs", "verdict": "A common scraping and search API for agents, rated well and self-hostable. Critics cite request logging, a skill that nudges agents to send feedback, and site owners paying for its customers' traffic.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 73, "hn_title_points": 53, "lobsters": 0, "devto_posts": 3, "github_stars": 199456, "downloads_month": null, "tone": 0.17 } }, { "rank": 5, "name": "Lightpanda", "url": "https://lightpanda.io", "score": 38.7, "what": "Lightweight headless browser written in Zig", "verdict": "A Zig headless browser built for agents and scraping. One team got big CPU and memory wins over Chrome; others hit pages it can't render and note it's easy to detect.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 48, "hn_title_points": 957, "lobsters": 0, "devto_posts": 1, "github_stars": 41556, "downloads_month": null, "tone": -0.25 } }, { "rank": 6, "name": "Stagehand", "url": "https://www.stagehand.dev", "score": 38.1, "what": "AI browser automation framework by Browserbase", "verdict": "Browserbase's open-source AI browser automation framework, with act, extract and observe calls and a cached replay layer. Threads are mostly vendor posts; one skeptic asks why not let agents write Playwright directly.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 44, "hn_title_points": 342, "lobsters": 0, "devto_posts": 1, "github_stars": 25978, "downloads_month": 8473095, "tone": 0.0 } }, { "rank": 7, "name": "Claude in Chrome", "url": "https://claude.com/chrome", "score": 35.7, "what": "Anthropic's Chrome extension browsing agent", "verdict": "Used for web chores in your logged-in browser: pulling Strava GPX files, downloading Datadog logs, driving ChatGPT, validating UX. Weak spots include missed obvious errors, multi-account setups and click-hijack vulnerabilities.", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 112, "hn_title_points": 322, "lobsters": 0, "devto_posts": 4, "github_stars": null, "downloads_month": null, "tone": 0.5 } }, { "rank": 8, "name": "Vibium", "url": "https://github.com/VibiumDev/vibium", "score": 35.2, "what": "Browser automation for agents from Selenium's creator", "verdict": "Browser automation for agents from Selenium's creator, with a CLI and agent skill. Most mentions come from the creator or people watching its Claude-coauthored repo; one user wants it leaner on tokens.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 50, "hn_title_points": 448, "lobsters": 0, "devto_posts": 0, "github_stars": 2955, "downloads_month": null, "tone": 0.08 } }, { "rank": 9, "name": "browser-use", "url": "https://browser-use.com", "score": 29.5, "what": "Open-source framework for LLM web agents", "verdict": "An open-source framework for LLM-driven web agents from YC W25, with 120k GitHub stars. Recent HN threads barely discuss it directly; it mostly appears as a benchmark comparison target.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 337, "hn_title_points": 413, "lobsters": 0, "devto_posts": 4, "github_stars": 120256, "downloads_month": 8061974, "tone": 0 } }, { "rank": 10, "name": "Scrapling", "url": "https://github.com/D4Vinci/Scrapling", "score": 27.0, "what": "Stealthy Python web scraping library", "verdict": "A stealthy Python scraping library that agents call for fetching pages, often behind an MCP. One frequent user says it loads nearly any site, even Google results, though output can be context-heavy.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 6, "hn_title_points": 1, "lobsters": 0, "devto_posts": 1, "github_stars": 99200, "downloads_month": 919855, "tone": 0.67 } }, { "rank": 11, "name": "Browser MCP", "url": "https://browsermcp.io", "score": 26.1, "what": "MCP server that drives your browser", "verdict": "Lets Cursor, Claude or VS Code drive your own browser. The Show HN drew 616 points, but later comments use 'browser MCP' generically, so there's little signal on this tool itself.", "tags": [ "Open source" ], "signals": { "hn_devs": 1, "hn_comments_12m": 130, "hn_title_points": 691, "lobsters": 0, "devto_posts": 5, "github_stars": 7307, "downloads_month": null, "tone": 0 } }, { "rank": 12, "name": "Jina Reader", "url": "https://jina.ai/reader", "score": 25.4, "what": "URL-to-markdown API for LLMs", "verdict": "An API that turns any URL into clean markdown for LLMs. Used as a first-pass fetcher that cuts page tokens sharply; the limit is it's read-only, with no clicking or form filling.", "tags": [], "signals": { "hn_devs": 1, "hn_comments_12m": 4, "hn_title_points": 4, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": 0.6 } }, { "rank": 13, "name": "Browserbase", "url": "https://www.browserbase.com", "score": 21.7, "what": "Hosted headless browser infrastructure for agents", "verdict": "Hosted browsers for agents, from the company behind Stagehand. One startup builds on it for session persistence; site operators blame it and its peers for abusive bot traffic and captcha bypass.", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 37, "hn_title_points": 43, "lobsters": 0, "devto_posts": 1, "github_stars": null, "downloads_month": null, "tone": -0.17 } } ] }, { "id": "sandbox", "title": "Sandboxes & safety", "tools": [ { "rank": 1, "name": "bubblewrap", "url": "https://github.com/containers/bubblewrap", "score": 76.7, "what": "Linux namespace sandbox for agent processes", "verdict": "The DIY pick on Linux: people wrap Claude Code, Codex and Pi in bwrap scripts the harness cannot disable. Main knock: Linux only, and weaker than a microVM.", "tags": [ "Open source" ], "signals": { "hn_devs": 7, "hn_comments_12m": 372, "hn_title_points": 326, "lobsters": 0, "devto_posts": 1, "github_stars": 9405, "downloads_month": null, "tone": 0.5 } }, { "rank": 2, "name": "Docker Sandboxes", "url": "https://www.docker.com/products/docker-sandboxes/", "score": 76.5, "what": "MicroVM sandboxes for running coding agents", "verdict": "Where people run Claude Code and Codex in yolo mode, inside a microVM rather than a container. Praised for solid isolation; knocked for workflow friction and closed source.", "tags": [], "signals": { "hn_devs": 6, "hn_comments_12m": 430, "hn_title_points": 1084, "lobsters": 1, "devto_posts": 2, "github_stars": null, "downloads_month": null, "tone": 0.55 } }, { "rank": 3, "name": "OpenShell", "url": "https://github.com/NVIDIA/OpenShell", "score": 57.5, "what": "NVIDIA policy-controlled runtime for agent fleets", "verdict": "NVIDIA's policy layer for agent fleets, used by some teams to run Claude Code under granular permissions. Sceptics call it overcomplicated next to bwrap or plain Docker.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 2, "hn_comments_12m": 64, "hn_title_points": 585, "lobsters": 0, "devto_posts": 2, "github_stars": 21245, "downloads_month": null, "tone": 0.0 } }, { "rank": 4, "name": "Apple container", "url": "https://github.com/apple/container", "score": 52.2, "what": "Apple's lightweight Linux VM containers for macOS", "verdict": "Mac users build agent sandboxes on it because each container is a lightweight Linux VM. Still called a work in progress, needs macOS 26, and you must block network access yourself.", "tags": [ "Open source" ], "signals": { "hn_devs": 1, "hn_comments_12m": 195, "hn_title_points": 1275, "lobsters": 0, "devto_posts": 1, "github_stars": 51090, "downloads_month": null, "tone": 0.25 } }, { "rank": 5, "name": "Dev containers", "url": "https://containers.dev", "score": 49.8, "what": "Open spec for containerised development environments", "verdict": "The no-new-tool answer: run the agent inside the devcontainer you already have. Protects against agent stupidity, not malice, and locking down the network takes real customisation.", "tags": [], "signals": { "hn_devs": 6, "hn_comments_12m": 27, "hn_title_points": 0, "lobsters": 0, "devto_posts": 4, "github_stars": null, "downloads_month": null, "tone": 0.6 } }, { "rank": 6, "name": "E2B", "url": "https://e2b.dev", "score": 49.6, "what": "Hosted Firecracker sandboxes for agent code execution", "verdict": "The hosted sandbox people build products on: per-session agent runs, untrusted code and preview environments. Volumes get praise; the cost is platform lock-in and limited customisation.", "tags": [ "Open source" ], "signals": { "hn_devs": 1, "hn_comments_12m": 60, "hn_title_points": 12, "lobsters": 0, "devto_posts": 9, "github_stars": 14761, "downloads_month": null, "tone": 0.5 } }, { "rank": 7, "name": "nono", "url": "https://github.com/nolabs-ai/nono", "score": 47.7, "what": "Kernel-enforced Landlock/Seatbelt sandbox for coding agents", "verdict": "Runs agents under Landlock or Seatbelt without a VM, so installed tooling still works, and can inject phantom credentials. Liked, but policy config is fiddlier than Anthropic's srt.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 3, "hn_comments_12m": 9, "hn_title_points": 8, "lobsters": 0, "devto_posts": 2, "github_stars": 4886, "downloads_month": null, "tone": 0.33 } }, { "rank": 8, "name": "smolvm", "url": "https://github.com/smol-machines/smolvm", "score": 45.4, "what": "Lightweight cross-platform microVMs for agent sandboxing", "verdict": "Where users land after trying containers and OS sandboxes; praised as light and solid enough to build a whole agent harness on. Some fear a YC rug pull.", "tags": [ "Open source" ], "signals": { "hn_devs": 1, "hn_comments_12m": 141, "hn_title_points": 13, "lobsters": 0, "devto_posts": 0, "github_stars": 7020, "downloads_month": null, "tone": 0.33 } }, { "rank": 9, "name": "yoloai", "url": "https://github.com/kstenerud/yoloai", "score": 39.6, "what": "Agent sandbox with copy-and-diff apply workflow", "verdict": "Copies your workdir into a sandbox, hides API keys, and shows a diff before changes land. Most HN mentions come from its author; one tester calls the workflow not yet smooth.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 2, "hn_comments_12m": 72, "hn_title_points": 2, "lobsters": 0, "devto_posts": 0, "github_stars": 227, "downloads_month": null, "tone": 0.0 } }, { "rank": 10, "name": "Lume", "url": "https://github.com/trycua/cua", "score": 35.2, "what": "macOS and Linux VM CLI for Apple Silicon", "verdict": "A macOS and Linux VM manager on Apple's Virtualization framework, used to sandbox Claude or run macOS-only agents. Users call it reasonably good but report setup hoops and disk resizing glitches.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 16, "hn_title_points": 468, "lobsters": 0, "devto_posts": 1, "github_stars": 31942, "downloads_month": null, "tone": 0.0 } }, { "rank": 11, "name": "Daytona", "url": "https://www.daytona.io", "score": 28.6, "what": "Hosted sandbox infrastructure for AI agents", "verdict": "Mostly named in lists of hosted agent sandboxes alongside E2B and Modal, and used as interchangeable infra in experiments. The one hands-on complaint: credits are near useless until higher usage tiers.", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 75, "hn_title_points": 63, "lobsters": 0, "devto_posts": 2, "github_stars": 71358, "downloads_month": null, "tone": -0.14 } }, { "rank": 12, "name": "Agent Safehouse", "url": "https://agent-safehouse.dev", "score": 27.8, "what": "", "verdict": "", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 38, "hn_title_points": 823, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": 0 } }, { "rank": 13, "name": "Ardent", "url": "https://tryardent.com", "score": 21.5, "what": "Copy-on-write Postgres branches for agents", "verdict": "Gives each agent run its own copy-on-write Postgres clone so a bad query cannot touch production. HN evidence is only the founder's own posts, so there is no independent user signal yet.", "tags": [], "signals": { "hn_devs": 1, "hn_comments_12m": 140, "hn_title_points": 116, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": 0 } }, { "rank": 14, "name": "Sprites (Fly.io)", "url": "https://sprites.dev", "score": 16.7, "what": "", "verdict": "", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 27, "hn_title_points": 5, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": 0 } }, { "rank": 15, "name": "container-use", "url": "https://github.com/dagger/container-use", "score": 16.5, "what": "Dagger's containerised environments for coding agents", "verdict": "Dagger's tool for giving each coding agent its own container environment. The one real HN comment objects that it requires docker group access by default.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 27, "hn_title_points": 91, "lobsters": 0, "devto_posts": 1, "github_stars": 4087, "downloads_month": null, "tone": 0 } }, { "rank": 16, "name": "Modal Sandboxes", "url": "https://modal.com/docs/guide/sandboxes", "score": 15.9, "what": "", "verdict": "", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 16, "hn_title_points": 10, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": 0 } }, { "rank": 17, "name": "Vercel Sandbox", "url": "https://vercel.com/docs/vercel-sandbox", "score": 9.9, "what": "", "verdict": "", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 2, "hn_title_points": 15, "lobsters": 0, "devto_posts": 1, "github_stars": null, "downloads_month": null, "tone": 0 } } ] }, { "id": "local", "title": "Local models & inference", "tools": [ { "rank": 1, "name": "Ollama", "url": "https://ollama.com", "score": 100.0, "what": "Local model runner with paid cloud tier", "verdict": "The easy path to local Qwen and Gemma, and increasingly a paid cloud plan for hosted open models like DeepSeek Flash and GLM. Few complaints beyond local models feeling limited.", "tags": [ "Open source", "Fading" ], "signals": { "hn_devs": 61, "hn_comments_12m": 1987, "hn_title_points": 8641, "lobsters": 1, "devto_posts": 126, "github_stars": 186256, "downloads_month": null, "tone": 0.6 } }, { "rank": 2, "name": "llama.cpp", "url": "https://github.com/ggml-org/llama.cpp", "score": 88.5, "what": "C/C++ inference engine and llama-server", "verdict": "A single fast binary that people move to from Ollama for control, usually serving models via llama-server. Complaints are about ROCm build friction and new flags that can cut speed.", "tags": [ "Open source", "Fading" ], "signals": { "hn_devs": 27, "hn_comments_12m": 2075, "hn_title_points": 3159, "lobsters": 1, "devto_posts": 29, "github_stars": 133652, "downloads_month": null, "tone": 0.39 } }, { "rank": 3, "name": "MLX", "url": "https://github.com/ml-explore/mlx", "score": 75.2, "what": "Apple's array framework for Apple Silicon ML", "verdict": "The Apple Silicon route to local inference, via MLX-Serve, LM Studio or custom engines. Praised for speed with less heat than a GPU box; some still prefer llama-server on Mac.", "tags": [ "Open source" ], "signals": { "hn_devs": 10, "hn_comments_12m": 715, "hn_title_points": 2846, "lobsters": 1, "devto_posts": 8, "github_stars": 29089, "downloads_month": 544885, "tone": 0.61 } }, { "rank": 4, "name": "LM Studio", "url": "https://lmstudio.ai", "score": 69.5, "what": "Desktop GUI app for running local models", "verdict": "The GUI on-ramp for local models, often paired with Aider, Continue or Cline and praised for easy setup and model switching. Newer gripe: it now steers downloads to its agentic app.", "tags": [ "Fading" ], "signals": { "hn_devs": 18, "hn_comments_12m": 596, "hn_title_points": 1405, "lobsters": 0, "devto_posts": 11, "github_stars": null, "downloads_month": null, "tone": 0.52 } }, { "rank": 5, "name": "Open WebUI", "url": "https://openwebui.com", "score": 69.1, "what": "Self-hosted chat UI for local and API models", "verdict": "The usual self-hosted chat front end over Ollama, llama.cpp or APIs, extended with plugins and Open Terminal. Feature-rich, but wiring up tools like search is painful and setup feels heavy.", "tags": [ "Fading" ], "signals": { "hn_devs": 18, "hn_comments_12m": 118, "hn_title_points": 261, "lobsters": 0, "devto_posts": 8, "github_stars": 158019, "downloads_month": null, "tone": 0.43 } }, { "rank": 6, "name": "vLLM", "url": "https://github.com/vllm-project/vllm", "score": 68.4, "what": "High-throughput GPU inference and serving engine", "verdict": "The pick for batching and throughput on dedicated GPUs, run stock by self-hosters with multi-GPU rigs. Seen as harder to deploy than llama.cpp, with reports of engine bugs on long contexts.", "tags": [ "Open source" ], "signals": { "hn_devs": 5, "hn_comments_12m": 924, "hn_title_points": 2272, "lobsters": 4, "devto_posts": 36, "github_stars": 96519, "downloads_month": 1857414, "tone": 0.24 } }, { "rank": 7, "name": "Unsloth", "url": "https://unsloth.ai", "score": 57.2, "what": "Quantized model builds, fine-tuning and desktop app", "verdict": "Best known as the source of quantized GGUFs people actually run, with even IQ1 DeepSeek quants holding up. Unsloth Desktop is gaining fans; complaints are tight context limits and no VRAM-to-RAM splitting.", "tags": [ "Open source" ], "signals": { "hn_devs": 4, "hn_comments_12m": 822, "hn_title_points": 1331, "lobsters": 0, "devto_posts": 2, "github_stars": 79152, "downloads_month": 885608, "tone": 0.2 } }, { "rank": 8, "name": "Lemonade", "url": "https://lemonade-server.ai", "score": 46.3, "what": "AMD's local LLM server for GPU and NPU", "verdict": "AMD's all-in-one local server and the go-to on Strix Halo, with easy installers and one endpoint for chat, images and speech. Owners say it replaced most cloud use; one reports weaker speeds.", "tags": [ "Open source" ], "signals": { "hn_devs": 3, "hn_comments_12m": 35, "hn_title_points": 617, "lobsters": 0, "devto_posts": 4, "github_stars": 6140, "downloads_month": null, "tone": 0.5 } }, { "rank": 9, "name": "SGLang", "url": "https://github.com/sgl-project/sglang", "score": 40.8, "what": "High-performance LLM serving framework for GPU clusters", "verdict": "A datacenter serving engine named alongside vLLM for multi-GPU inference. Recent talk centres on a Cloudflare KV-cache leak involving an SGLang state bug, though some put the blame on Cloudflare's config.", "tags": [ "Open source" ], "signals": { "hn_devs": 1, "hn_comments_12m": 167, "hn_title_points": 208, "lobsters": 0, "devto_posts": 7, "github_stars": 38559, "downloads_month": 2984581, "tone": -0.23 } }, { "rank": 10, "name": "llama-swap", "url": "https://github.com/mostlygeek/llama-swap", "score": 36.5, "what": "Model-swapping proxy in front of llama-server", "verdict": "The proxy home-lab users put in front of llama.cpp to hot-swap models behind one OpenAI-style endpoint. Called bulletproof by fans; critics say it adds hard-to-debug complexity now llama-server can juggle models itself.", "tags": [ "Open source" ], "signals": { "hn_devs": 3, "hn_comments_12m": 53, "hn_title_points": 2, "lobsters": 0, "devto_posts": 0, "github_stars": 6205, "downloads_month": null, "tone": 0.53 } }, { "rank": 11, "name": "Apfel", "url": "https://github.com/Arthur-Ficial/apfel", "score": 33.0, "what": "CLI and API for Apple's on-device model", "verdict": "Wraps Apple's on-device Foundation Model as a CLI and OpenAI-style server, so the Mac's built-in model answers quick shell questions. Liked as useful; Apple's own first-party server may now overlap it.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 0, "hn_comments_12m": 46, "hn_title_points": 757, "lobsters": 0, "devto_posts": 0, "github_stars": 6592, "downloads_month": null, "tone": 0.42 } }, { "rank": 12, "name": "Jan", "url": "https://jan.ai", "score": 26.0, "what": "Open-source desktop chat app for local models", "verdict": "The open-source desktop alternative to LM Studio: a simple chat window that searches and downloads GGUF models with no lock-in. Often recommended as an easy start; the threads hold no real complaints.", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 16, "hn_title_points": 1, "lobsters": 0, "devto_posts": 0, "github_stars": 45561, "downloads_month": null, "tone": 0.58 } }, { "rank": 13, "name": "KoboldCpp", "url": "https://github.com/LostRuins/koboldcpp", "score": 24.4, "what": "Single-file llama.cpp runner with built-in UI", "verdict": "One executable bundling llama.cpp with a web UI, image generation and speech, favoured by roleplay users for its tunables. Praised for zero-install simplicity, if a bit jack of all trades.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 21, "hn_title_points": 5, "lobsters": 0, "devto_posts": 0, "github_stars": 12189, "downloads_month": null, "tone": 0.25 } } ] }, { "id": "gateways", "title": "Model access & routing", "tools": [ { "rank": 1, "name": "OpenRouter", "url": "https://openrouter.ai", "score": 98.5, "what": "Pay-as-you-go API router across model providers", "verdict": "The pay-as-you-go side door to open-weight models like DeepSeek, GLM and Kimi, used to test models, run side projects and get past refusals. Main gripe: frontier models cost more there than subscriptions.", "tags": [ "Rising" ], "signals": { "hn_devs": 80, "hn_comments_12m": 3866, "hn_title_points": 6626, "lobsters": 1, "devto_posts": 18, "github_stars": null, "downloads_month": null, "tone": 0.62 } }, { "rank": 2, "name": "AWS Bedrock", "url": "https://aws.amazon.com/bedrock/", "score": 58.4, "what": "AWS managed API for Claude and other models", "verdict": "How companies buy Claude and other models at work: API pricing, budgets tracked with AWS spend, stronger privacy guarantees. Works well, but it costs more than Anthropic's own subscriptions.", "tags": [], "signals": { "hn_devs": 6, "hn_comments_12m": 461, "hn_title_points": 976, "lobsters": 0, "devto_posts": 116, "github_stars": null, "downloads_month": null, "tone": 0.36 } }, { "rank": 3, "name": "LiteLLM", "url": "https://github.com/BerriAI/litellm", "score": 55.6, "what": "Python SDK and proxy unifying LLM APIs", "verdict": "The common self-hosted gateway that teams point tools at for cheap multi-model access and token metering. Talk now centres on its PyPI supply-chain compromise and whether it can be trusted.", "tags": [ "Open source" ], "signals": { "hn_devs": 3, "hn_comments_12m": 431, "hn_title_points": 3453, "lobsters": 0, "devto_posts": 45, "github_stars": 62870, "downloads_month": 92239101, "tone": -0.14 } }, { "rank": 4, "name": "OpenCode Go / Zen", "url": "https://opencode.ai/go", "score": 54.3, "what": "OpenCode's subscription and pay-per-use model access", "verdict": "The budget pick: $10 a month buys generous open-weight usage, mostly DeepSeek Flash, and users rarely hit limits. The one knock is free Zen models rotating into the paid tier.", "tags": [ "Rising" ], "signals": { "hn_devs": 32, "hn_comments_12m": 35, "hn_title_points": 0, "lobsters": 0, "devto_posts": 1, "github_stars": null, "downloads_month": null, "tone": 0.65 } }, { "rank": 5, "name": "Claude Code Router", "url": "https://github.com/musistudio/claude-code-router", "score": 35.9, "what": "Proxy routing Claude Code to other models", "verdict": "The usual way to keep Claude Code's harness while swapping in Kimi, Grok, DeepSeek or local Qwen, handy when quota runs out. One user found tool-call translation never fully lined up.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 33, "hn_title_points": 163, "lobsters": 0, "devto_posts": 3, "github_stars": 38033, "downloads_month": null, "tone": 0.75 } }, { "rank": 6, "name": "llm (Simon Willison)", "url": "https://llm.datasette.io", "score": 34.3, "what": "CLI and Python library for many LLMs", "verdict": "Simon Willison's CLI and Python library for prompting any model through plugins, with exportable logs. Others build plugins and IPython tools on it and praise its docs; core releases have slowed.", "tags": [ "Open source" ], "signals": { "hn_devs": 9, "hn_comments_12m": 10, "hn_title_points": 0, "lobsters": 0, "devto_posts": 16, "github_stars": 12719, "downloads_month": 487919, "tone": 0.36 } }, { "rank": 7, "name": "Vercel AI Gateway", "url": "https://vercel.com/ai-gateway", "score": 27.8, "what": "Vercel's hosted multi-provider model API gateway", "verdict": "Shows up mainly as where new models launch, sometimes free, and as a pay-per-token work account. Gripes: a required credit card, and third-party routing with the usual provider quality issues.", "tags": [], "signals": { "hn_devs": 1, "hn_comments_12m": 25, "hn_title_points": 21, "lobsters": 0, "devto_posts": 42, "github_stars": null, "downloads_month": null, "tone": 0.08 } }, { "rank": 8, "name": "cc-switch", "url": "https://github.com/farion1231/cc-switch", "score": 25.0, "what": "Provider switcher for Claude Code and Codex", "verdict": "A switcher that points Claude Code or Codex at cheaper models like MiniMax or DeepSeek, translating between API formats. Thin evidence, but the users who mention it are happy.", "tags": [ "Open source" ], "signals": { "hn_devs": 1, "hn_comments_12m": 3, "hn_title_points": 3, "lobsters": 0, "devto_posts": 2, "github_stars": 152579, "downloads_month": null, "tone": 0 } }, { "rank": 9, "name": "Portkey", "url": "https://portkey.ai", "score": 20.7, "what": "Hosted AI gateway with routing and logging", "verdict": "An AI gateway that shows up in benchmark comparisons against LiteLLM and Bifrost. The one hands-on report switched from LiteLLM's proxy on the free plan and found it much faster.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 5, "hn_title_points": 11, "lobsters": 0, "devto_posts": 2, "github_stars": 13339, "downloads_month": null, "tone": 0.25 } }, { "rank": 10, "name": "Chutes", "url": "https://chutes.ai", "score": 20.1, "what": "Decentralised GPU inference with TEE-served models", "verdict": "Cheap open-model inference from trusted execution environments, so GPU owners cannot read prompts. One newcomer found no documented route into OpenCode or Claude Code.", "tags": [], "signals": { "hn_devs": 2, "hn_comments_12m": 32, "hn_title_points": 5, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": 0.25 } }, { "rank": 11, "name": "OmniRoute", "url": "https://github.com/cheaper-inference/OmniRoute", "score": 19.9, "what": "Self-hosted local router for model providers", "verdict": "A self-hosted router for model APIs, named once as someone's local counterpart to OpenRouter. Too little HN discussion to judge it.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 0, "hn_comments_12m": 1, "hn_title_points": 0, "lobsters": 0, "devto_posts": 5, "github_stars": 85385, "downloads_month": null, "tone": 0 } }, { "rank": 12, "name": "Cloudflare AI Gateway", "url": "https://developers.cloudflare.com/ai-gateway/", "score": 18.4, "what": "Cloudflare proxy for LLM calls with spend controls", "verdict": "A proxy in front of model APIs for monitoring, caching and spend caps without hosting your own. Complaints: budgets are fixed-window rules and reported prices can be wrong.", "tags": [], "signals": { "hn_devs": 1, "hn_comments_12m": 13, "hn_title_points": 7, "lobsters": 0, "devto_posts": 4, "github_stars": null, "downloads_month": null, "tone": 0.17 } }, { "rank": 13, "name": "Synthetic", "url": "https://synthetic.new", "score": 14.1, "what": "Subscription API for open-weight models", "verdict": "A flat subscription for open-weight models such as DeepSeek Flash, pitched at OpenCode users. The only HN mention is from a cofounder, so there is no independent signal.", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 28, "hn_title_points": 14, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": 0 } } ] }, { "id": "frameworks", "title": "Build your own agent", "tools": [ { "rank": 1, "name": "LangChain", "url": "https://www.langchain.com", "score": 74.2, "what": "Python and JS framework for LLM apps", "verdict": "Still the default r\u00e9sum\u00e9 keyword and a common base for internal RAG chatbots. Users say it works but adds overhead and latency, and some builders advertise skipping it entirely.", "tags": [ "Open source" ], "signals": { "hn_devs": 4, "hn_comments_12m": 816, "hn_title_points": 719, "lobsters": 1, "devto_posts": 98, "github_stars": 149163, "downloads_month": 170457766, "tone": 0.0 } }, { "rank": 2, "name": "Mastra", "url": "https://mastra.ai", "score": 57.6, "what": "Open-source TypeScript framework for AI agents", "verdict": "The TypeScript agent framework with the warmest comments. A heavy user runs production agents on it, and a bug reporter says the CTO fixed eval scoring bugs within nine hours.", "tags": [ "Open source" ], "signals": { "hn_devs": 2, "hn_comments_12m": 118, "hn_title_points": 708, "lobsters": 0, "devto_posts": 11, "github_stars": 29396, "downloads_month": 10391822, "tone": 0.21 } }, { "rank": 3, "name": "Claude Agent SDK", "url": "https://docs.claude.com/en/docs/agent-sdk/overview", "score": 54.2, "what": "Anthropic's SDK exposing the Claude Code harness", "verdict": "Most discussion is about billing: whether it can run on a Claude subscription, which the docs now confirm. Builders note it inherits Claude Code's heavy system prompts, and telemetry takes hacks.", "tags": [], "signals": { "hn_devs": 2, "hn_comments_12m": 395, "hn_title_points": 164, "lobsters": 0, "devto_posts": 21, "github_stars": null, "downloads_month": 58450543, "tone": 0.08 } }, { "rank": 4, "name": "LangGraph", "url": "https://langchain-ai.github.io/langgraph/", "score": 54.2, "what": "Graph-based agent orchestration library from LangChain", "verdict": "Mostly a r\u00e9sum\u00e9 keyword next to LangChain. Where people describe it, they reach for it when they want deterministic, graph-shaped control over multi-step agent workflows.", "tags": [ "Open source" ], "signals": { "hn_devs": 1, "hn_comments_12m": 465, "hn_title_points": 562, "lobsters": 0, "devto_posts": 102, "github_stars": 44618, "downloads_month": 44769952, "tone": 0.0 } }, { "rank": 5, "name": "DSPy", "url": "https://dspy.ai", "score": 52.8, "what": "Stanford framework for programming and optimizing LLM prompts", "verdict": "Loved by a vocal few for small classification and extraction tasks done in a few lines. Skeptics say plain agent loops beat DSPy-built agents in their own evals.", "tags": [ "Open source" ], "signals": { "hn_devs": 2, "hn_comments_12m": 130, "hn_title_points": 474, "lobsters": 0, "devto_posts": 6, "github_stars": 38967, "downloads_month": 5312050, "tone": 0.21 } }, { "rank": 6, "name": "Microsoft Agent Framework", "url": "https://github.com/microsoft/agent-framework", "score": 50.3, "what": "Microsoft's open-source .NET and Python agent framework", "verdict": "Shows up mostly in .NET and Azure shops, in job posts and as the C# answer for MCP-based agents. Few hands-on reviews; one commenter notes its harness is still single-player.", "tags": [ "Open source" ], "signals": { "hn_devs": 3, "hn_comments_12m": 19, "hn_title_points": 106, "lobsters": 1, "devto_posts": 84, "github_stars": 14577, "downloads_month": null, "tone": 0.2 } }, { "rank": 7, "name": "OpenAI Agents SDK", "url": "https://openai.github.io/openai-agents-python/", "score": 49.1, "what": "OpenAI's lightweight Python multi-agent SDK", "verdict": "Common in r\u00e9sum\u00e9s and enterprise orchestrators. The one detailed user report praises it for wiring sandbox, Slack and knowledge-base integrations while controlling exactly what gets exposed to OpenAI.", "tags": [ "Open source" ], "signals": { "hn_devs": 1, "hn_comments_12m": 55, "hn_title_points": 964, "lobsters": 0, "devto_posts": 9, "github_stars": 30547, "downloads_month": 12449784, "tone": 0.08 } }, { "rank": 8, "name": "Google ADK", "url": "https://google.github.io/adk-docs/", "score": 43.3, "what": "Google's open-source Agent Development Kit", "verdict": "Google's agent kit, used for sequential and looping multi-agent setups. It ships features fast enough to obsolete in-house work, but its LiteLLM dependency drew complaints after LiteLLM's vulnerability.", "tags": [ "Open source" ], "signals": { "hn_devs": 1, "hn_comments_12m": 71, "hn_title_points": 46, "lobsters": 0, "devto_posts": 95, "github_stars": 22135, "downloads_month": 10109331, "tone": -0.08 } }, { "rank": 9, "name": "CrewAI", "url": "https://www.crewai.com", "score": 41.5, "what": "Python framework for role-based multi-agent crews", "verdict": "An early multi-agent framework that now shows up mainly as a r\u00e9sum\u00e9 keyword. The one experience report says agents left to coordinate themselves collaborate awkwardly unless the spec defines roles.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 156, "hn_title_points": 86, "lobsters": 0, "devto_posts": 30, "github_stars": 60609, "downloads_month": 2442764, "tone": 0.0 } }, { "rank": 10, "name": "Pydantic AI", "url": "https://ai.pydantic.dev", "score": 41.4, "what": "Type-safe Python agent framework from Pydantic", "verdict": "A Python favourite among people leaving LangChain. Users like its provider abstractions on their own and use it as a thin base for small headless agents.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 110, "hn_title_points": 298, "lobsters": 0, "devto_posts": 6, "github_stars": 21077, "downloads_month": 5383780, "tone": 0.33 } }, { "rank": 11, "name": "MCP SDKs", "url": "https://modelcontextprotocol.io", "score": 39.8, "what": "Official SDKs for building MCP servers and clients", "verdict": "In practice this means FastMCP: a one-line import for wrapping APIs as MCP servers. Complaints are deployment gotchas like localhost-only DNS rebinding defaults, and doubts that MCP beats plain OpenAPI.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 41, "hn_title_points": 201, "lobsters": 0, "devto_posts": 31, "github_stars": 24793, "downloads_month": 235193557, "tone": 0.08 } }, { "rank": 12, "name": "Vercel AI SDK", "url": "https://ai-sdk.dev", "score": 38.9, "what": "TypeScript toolkit for LLM apps and streaming", "verdict": "The default TypeScript LLM toolkit in r\u00e9sum\u00e9s and provider integrations, with its stream protocol ported as far as Ruby. Comments describe usage, not opinions.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 112, "hn_title_points": 40, "lobsters": 0, "devto_posts": 14, "github_stars": 27801, "downloads_month": 145134192, "tone": 0.0 } }, { "rank": 13, "name": "Strands Agents", "url": "https://strandsagents.com", "score": 38.5, "what": "AWS's open-source model-driven agent SDK", "verdict": "AWS's agent SDK, used with Bedrock for things like log-digging devops agents and a talking Raspberry Pi fish. One user hit a cache TTL issue; another questioned what Strands Harness adds.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 38, "hn_title_points": 137, "lobsters": 0, "devto_posts": 37, "github_stars": 9069, "downloads_month": 37112498, "tone": 0.0 } }, { "rank": 14, "name": "smolagents", "url": "https://github.com/huggingface/smolagents", "score": 33.1, "what": "", "verdict": "", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 33, "hn_title_points": 28, "lobsters": 0, "devto_posts": 5, "github_stars": 30253, "downloads_month": 425605, "tone": 0 } }, { "rank": 15, "name": "Jido", "url": "https://github.com/agentjido/jido", "score": 30.2, "what": "", "verdict": "", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 26, "hn_title_points": 354, "lobsters": 0, "devto_posts": 0, "github_stars": 1905, "downloads_month": null, "tone": 0 } }, { "rank": 16, "name": "BAML", "url": "https://boundaryml.com", "score": 28.8, "what": "", "verdict": "", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 35, "hn_title_points": 27, "lobsters": 0, "devto_posts": 2, "github_stars": 9492, "downloads_month": null, "tone": 0 } }, { "rank": 17, "name": "Pi agent toolkit", "url": "https://github.com/earendil-works/pi", "score": 19.9, "what": "", "verdict": "", "tags": [], "signals": { "hn_devs": 1, "hn_comments_12m": 7, "hn_title_points": 0, "lobsters": 0, "devto_posts": 2, "github_stars": null, "downloads_month": null, "tone": 0 } } ] }, { "id": "evals", "title": "Evals & observability", "tools": [ { "rank": 1, "name": "OpenTelemetry GenAI", "url": "https://opentelemetry.io/docs/specs/semconv/gen-ai/", "score": 55.4, "what": "OpenTelemetry semantic conventions for LLM telemetry", "verdict": "The plumbing under most LLM observability tools rather than something people debate. People instrument API calls so agents can query a local collector, and export LLM traces to vendors like Braintrust.", "tags": [], "signals": { "hn_devs": 3, "hn_comments_12m": 49, "hn_title_points": 170, "lobsters": 3, "devto_posts": 52, "github_stars": null, "downloads_month": null, "tone": 0.08 } }, { "rank": 2, "name": "promptfoo", "url": "https://www.promptfoo.dev", "score": 47.7, "what": "Open-source CLI for prompt evals and red-teaming", "verdict": "The go-to for personal regression evals: people use it to check prompts before switching models and to build benchmarks for their own workflow. Its team has since joined OpenAI.", "tags": [ "Open source" ], "signals": { "hn_devs": 1, "hn_comments_12m": 48, "hn_title_points": 70, "lobsters": 0, "devto_posts": 2, "github_stars": 26936, "downloads_month": 4010231, "tone": 0.31 } }, { "rank": 3, "name": "Langfuse", "url": "https://langfuse.com", "score": 46.2, "what": "Open-source LLM tracing, evals and prompt management", "verdict": "The default open-source pick for LLM tracing, evals and prompt management, all over r\u00e9sum\u00e9s. Users note a processing lag on traces, which a reply blames on outdated SDKs, and limited chart customisation.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 125, "hn_title_points": 272, "lobsters": 0, "devto_posts": 10, "github_stars": 37128, "downloads_month": 23563200, "tone": 0.0 } }, { "rank": 4, "name": "Arize Phoenix", "url": "https://phoenix.arize.com", "score": 42.5, "what": "Open-source LLM tracing and eval platform", "verdict": "Open-source tracing and eval tool people use for LLM-as-judge and production trace analysis. Evidence is thin, and one user found its Claude Code tracing setup heavy.", "tags": [ "Fading" ], "signals": { "hn_devs": 30, "hn_comments_12m": 5, "hn_title_points": 1, "lobsters": 0, "devto_posts": 24, "github_stars": 12191, "downloads_month": 626210, "tone": 0.2 } }, { "rank": 5, "name": "MLflow", "url": "https://mlflow.org", "score": 38.0, "what": "Open-source ML lifecycle and experiment tracking platform", "verdict": "The established ML experiment tracker, now also used for agent tracing. Those who tried MLflow 3's agent observability liked its session capture and metadata like files touched and model used.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 55, "hn_title_points": 23, "lobsters": 0, "devto_posts": 3, "github_stars": 28717, "downloads_month": 19887934, "tone": 0.17 } }, { "rank": 6, "name": "Logfire", "url": "https://pydantic.dev/logfire", "score": 33.2, "what": "Pydantic's OpenTelemetry-based observability platform", "verdict": "Pydantic's tracing service, often paired with Pydantic AI. Users call it very helpful for diagnosing agent bugs; one who tried both found it similar to Langfuse but with fewer eval features.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 14, "hn_title_points": 43, "lobsters": 0, "devto_posts": 0, "github_stars": 4537, "downloads_month": 13028487, "tone": 0.25 } }, { "rank": 7, "name": "Opik", "url": "https://www.comet.com/site/products/opik/", "score": 29.4, "what": "Comet's open-source LLM tracing and eval tool", "verdict": "Comet's open-source LLM tracing and eval tool, with little discussion. Mentions call it useful for LLMOps, and one contributor fixed streaming wrappers that were swallowing exceptions.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 5, "hn_title_points": 13, "lobsters": 0, "devto_posts": 1, "github_stars": 23246, "downloads_month": 1741711, "tone": 0.4 } }, { "rank": 8, "name": "Braintrust", "url": "https://www.braintrust.dev", "score": 29.1, "what": "Hosted LLM eval and observability platform", "verdict": "A hosted eval platform that shows up in startup stacks like Curai's. The few opinions call it solid for evals; its top HN story is a confirmed breach.", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 32, "hn_title_points": 6, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": 6574708, "tone": 0.2 } }, { "rank": 9, "name": "Helicone", "url": "https://www.helicone.ai", "score": 28.0, "what": "Open-source LLM gateway and observability proxy", "verdict": "Known as much for its LLM cost page as its gateway. Self-hosters found almost no setup docs, cost analysis needs your own SQL, and the service is now in maintenance mode.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 10, "hn_title_points": 47, "lobsters": 0, "devto_posts": 2, "github_stars": 6273, "downloads_month": null, "tone": -0.1 } }, { "rank": 10, "name": "DeepEval", "url": "https://deepeval.com", "score": 27.5, "what": "Open-source Python framework for LLM evals", "verdict": "A Python eval library used mostly for RAG checks and LLM-as-judge assertions. Advice threads pitch it as the code-first option next to promptfoo's YAML.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 12, "hn_title_points": 3, "lobsters": 0, "devto_posts": 1, "github_stars": 19352, "downloads_month": 2584198, "tone": 0.17 } }, { "rank": 11, "name": "LangSmith", "url": "https://www.langchain.com/langsmith", "score": 27.2, "what": "LangChain's hosted tracing and eval service", "verdict": "LangChain's tracing and eval service, mostly a r\u00e9sum\u00e9 line next to LangGraph, sometimes the eval harness for production agent systems. The one hands-on report dropped it for adding latency.", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 67, "hn_title_points": 21, "lobsters": 0, "devto_posts": 8, "github_stars": null, "downloads_month": null, "tone": -0.08 } }, { "rank": 12, "name": "Inspect AI", "url": "https://inspect.aisi.org.uk", "score": 25.5, "what": "UK AISI's open-source LLM evaluation framework", "verdict": "The UK AI Security Institute's eval framework. HN barely discusses it: one r\u00e9sum\u00e9, one PR link, and someone planning to try it alongside promptfoo.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 12, "hn_title_points": 3, "lobsters": 0, "devto_posts": 1, "github_stars": 3198, "downloads_month": 2896739, "tone": 0 } }, { "rank": 13, "name": "garak", "url": "https://github.com/NVIDIA/garak", "score": 14.4, "what": "NVIDIA's open-source LLM vulnerability scanner", "verdict": "NVIDIA's LLM vulnerability scanner. It appears only in red-teaming tool lists next to PyRIT; most HN hits for the name are about the Star Trek character.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 11, "hn_title_points": 0, "lobsters": 0, "devto_posts": 0, "github_stars": 9918, "downloads_month": 39200, "tone": 0 } }, { "rank": 14, "name": "Weights & Biases Weave", "url": "https://wandb.ai/site/weave/", "score": 12.8, "what": "", "verdict": "", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 0, "hn_title_points": 0, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": 817064, "tone": 0 } }, { "rank": 15, "name": "Laminar", "url": "https://www.lmnr.ai", "score": 12.4, "what": "", "verdict": "", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 0, "hn_title_points": 0, "lobsters": 0, "devto_posts": 0, "github_stars": 3477, "downloads_month": null, "tone": 0 } } ] }, { "id": "assistants", "title": "Always-on agents", "tools": [ { "rank": 1, "name": "OpenClaw", "url": "https://openclaw.ai", "score": 93.0, "what": "Open-source always-on personal AI agent", "verdict": "The personal agent many tried and most have retired. A few keep it for event-driven chores and calendar nudges; the rest cite constant tinkering, crashes after a week and poor local-model support.", "tags": [ "Open source", "Rising" ], "signals": { "hn_devs": 72, "hn_comments_12m": 5090, "hn_title_points": 11707, "lobsters": 1, "devto_posts": 145, "github_stars": 394125, "downloads_month": null, "tone": -0.46 } }, { "rank": 2, "name": "Hermes Agent", "url": "https://github.com/NousResearch/hermes-agent", "score": 82.4, "what": "Nous Research's open-source persistent personal agent", "verdict": "The personal agent with the most loyal users, run on Mac Minis or VPSes, reached via Telegram, often on local models. Detractors cite a dated UI and regressions between updates.", "tags": [ "Open source", "Rising" ], "signals": { "hn_devs": 54, "hn_comments_12m": 507, "hn_title_points": 309, "lobsters": 0, "devto_posts": 204, "github_stars": 261431, "downloads_month": null, "tone": 0.08 } }, { "rank": 3, "name": "Muse", "url": "https://ai.meta.com/muse/", "score": 66.7, "what": "Meta's consumer personal AI agent", "verdict": "Meta's consumer agent. Fans use it for research, forum scraping and group marketing, but most comments cite privacy incidents, ignored permissions, a 0-day, and slowness.", "tags": [ "Rising" ], "signals": { "hn_devs": 21, "hn_comments_12m": 86, "hn_title_points": 1326, "lobsters": 1, "devto_posts": 17, "github_stars": null, "downloads_month": null, "tone": -0.28 } }, { "rank": 4, "name": "Claude Cowork", "url": "https://claude.com/product/cowork", "score": 60.4, "what": "Claude's desktop agent for non-coding work", "verdict": "Claude's agent for non-coding work, used for email, todos, receipts and business operations, including by non-developers. One complaint: it floods teams with AI slop that reviewers must clean up.", "tags": [], "signals": { "hn_devs": 40, "hn_comments_12m": 937, "hn_title_points": 3681, "lobsters": 0, "devto_posts": 38, "github_stars": null, "downloads_month": null, "tone": 0.36 } }, { "rank": 5, "name": "n8n", "url": "https://n8n.io", "score": 60.0, "what": "Fair-code visual workflow automation platform", "verdict": "The workflow automation tool AI and automation freelancers list on r\u00e9sum\u00e9s, used for chores like filing tax records and forwarding email. One comment calls its open-core licensing direction a mess.", "tags": [], "signals": { "hn_devs": 4, "hn_comments_12m": 429, "hn_title_points": 1496, "lobsters": 1, "devto_posts": 107, "github_stars": 209761, "downloads_month": null, "tone": -0.06 } }, { "rank": 6, "name": "NanoClaw", "url": "https://github.com/qwibitai/nanoclaw", "score": 48.2, "what": "Minimal container-isolated OpenClaw alternative in TypeScript", "verdict": "A lightweight, container-isolated OpenClaw alternative some find easier to understand. Several users moved on to Hermes or hand-rolled agents, and one ditched it as useless for email triage.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 3, "hn_comments_12m": 200, "hn_title_points": 1486, "lobsters": 0, "devto_posts": 2, "github_stars": 31091, "downloads_month": null, "tone": -0.13 } }, { "rank": 7, "name": "Home Assistant + LLM", "url": "https://www.home-assistant.io/integrations/conversation/", "score": 47.0, "what": "Home Assistant's LLM conversation and voice integrations", "verdict": "Home Assistant's LLM and voice hooks, often run on local models. People wire it to MCP servers for voice control and EV charging plans, though LLMs can struggle with the unofficial MCP.", "tags": [ "Rising" ], "signals": { "hn_devs": 13, "hn_comments_12m": 93, "hn_title_points": 2, "lobsters": 1, "devto_posts": 3, "github_stars": null, "downloads_month": null, "tone": 0.37 } }, { "rank": 8, "name": "Claude Duet", "url": "https://claude.com/product/claude-duet", "score": 36.8, "what": "Anthropic's proactive agent for professionals, in preview", "verdict": "Anthropic's waitlisted proactive agent, still in research preview. Early users run it for monitoring communications, but find it slow and say OpenAI's Dots is better today.", "tags": [], "signals": { "hn_devs": 3, "hn_comments_12m": 32, "hn_title_points": 792, "lobsters": 0, "devto_posts": 7, "github_stars": null, "downloads_month": null, "tone": -0.27 } }, { "rank": 9, "name": "Dots", "url": "https://openai.com/index/introducing-dots/", "score": 22.4, "what": "OpenAI's persistent personal agent", "verdict": "OpenAI's persistent personal agent. One user built a speculative Lean-verified proof with a Dots swarm, another rates it above Duet; critics dislike the cutesy branding and a confusing product lineup.", "tags": [], "signals": { "hn_devs": 1, "hn_comments_12m": 24, "hn_title_points": 16, "lobsters": 0, "devto_posts": 0, "github_stars": null, "downloads_month": null, "tone": 0.15 } }, { "rank": 10, "name": "Paperclip", "url": "https://github.com/paperclipai/paperclip", "score": 19.7, "what": "Open-source org chart for running agent companies", "verdict": "An open-source tool for running a 'company' of agents with an org chart. HN mentions are passing comparisons, with no hands-on reports.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 0, "hn_comments_12m": 3, "hn_title_points": 4, "lobsters": 0, "devto_posts": 0, "github_stars": 111950, "downloads_month": null, "tone": 0 } }, { "rank": 11, "name": "Claude Code Routines", "url": "https://code.claude.com/docs/en/routines", "score": 19.0, "what": "Scheduled, cloud-run Claude Code agent jobs", "verdict": "Scheduled Claude Code jobs, essentially agent cron. People see it as a cheaper subscription path for PR review bots, but several say a local cron job does the same.", "tags": [], "signals": { "hn_devs": 0, "hn_comments_12m": 15, "hn_title_points": 729, "lobsters": 0, "devto_posts": 4, "github_stars": null, "downloads_month": null, "tone": -0.08 } }, { "rank": 12, "name": "nanobot", "url": "https://github.com/HKUDS/nanobot", "score": 16.6, "what": "Ultra-lightweight open-source OpenClaw alternative", "verdict": "An ultra-lightweight open-source alternative to OpenClaw from HKUDS. None of the matched HN comments discuss it; every hit is about sci-fi nanobots.", "tags": [ "Open source", "New in 2026" ], "signals": { "hn_devs": 0, "hn_comments_12m": 106, "hn_title_points": 308, "lobsters": 0, "devto_posts": 1, "github_stars": 49903, "downloads_month": null, "tone": 0 } }, { "rank": 13, "name": "Eigent", "url": "https://github.com/eigent-ai/eigent", "score": 13.1, "what": "Open-source desktop alternative to Claude Cowork", "verdict": "An open-source desktop alternative to Claude Cowork. HN evidence is effectively nil: one bare link, with every other match being German 'eigentlich' or EigenTrust.", "tags": [ "Open source" ], "signals": { "hn_devs": 0, "hn_comments_12m": 20, "hn_title_points": 51, "lobsters": 0, "devto_posts": 0, "github_stars": 15762, "downloads_month": null, "tone": 0 } } ] } ] }