Runs Hermes on open-weight models: coded a game, built his website, does server maintenance and security
Open Agents
Open-source AI agents you can run yourself.
11 actively developed agents. For each one: stars, license, where it runs, what it costs and its latest release, plus real use cases and complaints, all with source links.
Data checked · GitHub API, official READMEs and sites, X posts

Self-improving personal agent from Nous Research: it writes skills from experience, keeps curated memory, and runs on many backends.
- GitHub stars
- 254,311
- Latest release
- v0.22.2Oct 15, 2026
- Runs
- Local + cloud
- X posts this month
- 23of 123 in our brief
Top features
- Closed learning loopautonomous skill creation, self-improving skills, FTS5 cross-session search
- Kanban board for coordinating multiple agents
- Batch trajectory generation for training tool-calling models
What people are building with it
"Hermes Agent is driving my Microduck!", a robot walking 0.58 m from a single sentence
Fully local on "Qwen3.8-Flash-Next on an RTX 5060 Ti (16GB) + offload 96GB RAM"
Basis: in our X Brief (Sep 16 – Oct 16, 2026), 23 of 123 tallied posts were about Hermes, the most of any agent. OpenClaw was close behind with 22. That's a narrow lead in one month of posts, not a quality ranking. OpenClaw has more GitHub stars.
- OpenBot is source-available (PolyForm-Noncommercial-1.0.0), not OSI open source: non-commercial use only; not OSI open source.
- Crush is source-available (FSL-1.1-MIT), not OSI open source: converts to MIT later; not OSI open source.
- Fields we couldn't confirm are labelled Unverified. Stars, licenses and releases come from the GitHub API on Oct 16, 2026.
- Use cases are user posts. Self-reported numbers aren't verified, and posts that look like a coordinated campaign are hidden by default.
At a glance
Free and open source (MIT)
Latest v0.22.2 · Oct 15, 2026
Free (MIT)
Latest v2026.10.4 · Oct 16, 2026
Free (MIT) with BYO keys
Latest v1.19.3 · Oct 16, 2026
Open source is free
Latest v1.27.1 · Oct 12, 2026
Free for individuals
Latest desktop-v0.0.47 · Oct 16, 2026
Free (Apache-2.0)
Latest rust-v0.0.56 · Oct 7, 2026
Free (Apache-2.0)
Latest v1.54.0 · Oct 8, 2026
Free (Apache-2.0)
Latest v0.86.0 · Aug 9, 2025
Free to use with BYO keys
Latest v0.98.1 · Oct 9, 2026
Free tier ($0, BYOK or local)
Latest jetbrains/v7.1.10 · Oct 15, 2026
Free to use under a non-commercial license
Latest v0.42.0 · Oct 16, 2026
Comparison table11
| Agent | Stars | License | Runs | Cost | Latest release |
|---|---|---|---|---|---|
| Hermes Agent | 254,311 | MIT | Local + cloud | Free and open source (MIT) partly unverified | v0.22.2Oct 15, 2026 |
| OpenClaw | 392,167 | MIT | Local + cloud | Free (MIT) | v2026.10.4Oct 16, 2026 |
| OpenCode | 214,211 | MIT | Local | Free (MIT) with BYO keys | v1.19.3Oct 16, 2026 |
| OpenHands | 91,056 | MIT | Local + cloud | Open source is free | v1.27.1Oct 12, 2026 |
| Cline | 70,404 | Apache-2.0 | Local | Free for individuals | desktop-v0.0.47Oct 16, 2026 |
| Open Interpreter | 68,589 | Apache-2.0 | Local | Free (Apache-2.0) | rust-v0.0.56Oct 7, 2026 |
| goose | 55,342 | Apache-2.0 | Local | Free (Apache-2.0) | v1.54.0Oct 8, 2026 |
| Aider | 49,584 | Apache-2.0 | Local | Free (Apache-2.0) | v0.86.0Aug 9, 2025 |
| Crush | 28,675 | FSL-1.1-MIT · source-available | Local | Free to use with BYO keys partly unverified | v0.98.1Oct 9, 2026 |
| Kilo Code | 27,629 | MIT | Local + cloud | Free tier ($0, BYOK or local) | jetbrains/v7.1.10Oct 15, 2026 |
| OpenBot | 579 | PolyForm-Noncommercial-1.0.0 · source-available | Local | Free to use under a non-commercial license partly unverified | v0.42.0Oct 16, 2026 |
Stars
Chart data as a table
| Agent | Stars | Checked |
|---|---|---|
| OpenClaw | 392,167 | 2026-10-16 |
| Hermes Agent | 254,311 | 2026-10-16 |
| OpenCode | 214,211 | 2026-10-16 |
| OpenHands | 91,056 | 2026-10-16 |
| Cline | 70,404 | 2026-10-16 |
| Open Interpreter | 68,589 | 2026-10-16 |
| goose | 55,342 | 2026-10-16 |
| Aider | 49,584 | 2026-10-16 |
| Crush | 28,675 | 2026-10-16 |
| Kilo Code | 27,629 | 2026-10-16 |
| OpenBot | 579 | 2026-10-16 |
Source: GitHub REST API (repos)
Real use cases
42 posts from users, linked to the source. Self-reported numbers aren't verified. Categories are assigned by Ratelimit Now from the post text.
"Hermes + DeepSeek Flash … I used it to gamify learning about Linux Networking Skills" for a teenager
Agent 'Neura' made a booking in Dubai on his behalf using voice and calling tools
Fully local on "Qwen3.8-Flash-Next on an RTX 5060 Ti (16GB) + offload 96GB RAM"
A Japanese-language user ran goose with a local Qwen3.8-Flash-Next served from Strata
Opus orchestrating spare Codex resets "through Herdr, T3, Hermes, OpenClaw, whatever works"
Tips for applying Grok Bot workflows to OpenBot: "Give it skills (crucial)"
Runs a custom automated research and briefing workflow
"Yap-to-content": 30–40 minutes of voice notes to Telegram, and "My OpenClaw agent structures my thoughts"
Built a Chrome extension in under 5 minutes to re-enable right-click on a bank site, using Muse Spark 1.3
A consultant who has "created hermes and openclaw for clients (they just know it's an agent)"
Grasper, built at a Cline hackathon, "blocks fake installs before they run"
"My @openclaw agent called a Japanese restaurant to make a reservation … With @twilio and @ChatGPT voice"
"We embedded @goose_oss into our platform, connected the platform through MCP, and made it accessible from the web using ACP."
Moved a day's workflow from Cursor to Kilo Code to test it on a real task list (thread)
A user who likes it the most "as alternative to GrokBot + using private models from @AskVenice"
"Hermes Agent is driving my Microduck!", a robot walking 0.58 m from a single sentence
Rolled goose out internally for offline use; 10 users after one month
A Claude subscription used inside Hermes via the Direct SDK plugin, with an install video
One VPS running "my Hermes agent, Claude Code, Codex"
Runs Hermes on open-weight models: coded a game, built his website, does server maintenance and security
Moved most of his work onto agents, OpenClaw first
"made this sci-fi bunker in @threejs using union alpha in @cline took approximately 1 hour and 580M tokens"
Burned 1B tokens in 30 days with Hermes; thread on what he built
Exploring orchestrator plus subagents using open models (cloud orchestrator, local subagents)
Built a retro-style note-taking app in Godot by training his own agent on goose
Mac assistant connected to iMessage, browser and phone; automated voice calls to restaurants via OpenAI realtime + Twilio
Wired the Kanban view to Hermes for an automated short-form content pipeline for his app
Uses the Kilo CLI with high-visibility diff views before edits
An OpenHands PR prompted a session-ID validation change in his CodeFlowMu project
Fixed a pagination bug in a Python service with Aider, keeping every step as an undoable commit
Used GLM heavily on Kilo Code
Does development overnight with OpenHands agent loops
Persistent memory with the Honcho plugin
Tips on DeepSeek's direct API vs OpenCode Go: "if i'm spending around $20–25/month on deepseek, i'd stay on the direct api"
A live benchmark of OpenCode Go models
Claimed result: "Hermes alone: $4.29, 23 min. Hermes + oh-my-hermes: $0.66, 5 min"
figures self-reported, unverified
Comparison test: "Kilo Code + Opus 5.5: 21.8 min, 87 tool calls, $4.29 omp + Opus 5.5: 22.8 min, 73 tool calls, $4.39"
Built a pentesting agent for recon, vulnerability hunting and reporting (video)
Built 'Hermes Mobile', an unofficial open-source Android client (chat, model switching, files over Tailscale) using Hermes
"100+ tested workflows with full code", an example archive
"Hermes Fleet for iOS has been APPROVED for External TestFlight", built to "manage my Hermes Agent fleet from my iPhone"
Tested the agent workflow with early access to Cline Desktop
Every agent in detail
Hermes Agent★ 254k · MIT
What it is
Self-improving personal agent from Nous Research: it writes skills from experience, keeps curated memory, and runs on many backends.
GitHub repo · hermes-agent.nousresearch.com · @NousResearch
X account: maker account; @Hermes_Agent is an unrelated crypto project, not official
Install
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bashWindows: iex (irm https://hermes-agent.nousresearch.com/install.ps1)Android/Termux APT repo
raw.githubusercontent.com/NousResearch/hermes-agent/HEAD/README.md · checked
Models
Any model: Nous Portal (300+ models), OpenRouter, OpenAI, custom endpoints and others; switch with `hermes model`.
raw.githubusercontent.com/NousResearch/hermes-agent/HEAD/README.md · checked
Cost
Free and open source (MIT). Bring your own keys, or use a Nous Portal subscription, which bundles models plus a Tool Gateway (web search, image gen, TTS, cloud browser). Portal lists per-token model prices.
Unverified: Nous Portal subscription price not capturedportal.nousresearch.com · checked
Features
- Closed learning loop: autonomous skill creation, self-improving skills, FTS5 cross-session search raw.githubusercontent.com/NousResearch/hermes-agent/HEAD/README.md
- Kanban board for coordinating multiple agents x.com/minhng92/status/2107093452962074776 · user report
- Batch trajectory generation for training tool-calling models raw.githubusercontent.com/NousResearch/hermes-agent/HEAD/README.md
Use cases
"Hermes + DeepSeek Flash … I used it to gamify learning about Linux Networking Skills" for a teenager
Fully local on "Qwen3.8-Flash-Next on an RTX 5060 Ti (16GB) + offload 96GB RAM"
Claimed result: "Hermes alone: $4.29, 23 min. Hermes + oh-my-hermes: $0.66, 5 min"
figures self-reported, unverified
Built a pentesting agent for recon, vulnerability hunting and reporting (video)
Built 'Hermes Mobile', an unofficial open-source Android client (chat, model switching, files over Tailscale) using Hermes
"100+ tested workflows with full code", an example archive
"Hermes Fleet for iOS has been APPROVED for External TestFlight", built to "manage my Hermes Agent fleet from my iPhone"
"Hermes Agent is driving my Microduck!", a robot walking 0.58 m from a single sentence
A Claude subscription used inside Hermes via the Direct SDK plugin, with an install video
One VPS running "my Hermes agent, Claude Code, Codex"
Runs Hermes on open-weight models: coded a game, built his website, does server maintenance and security
Burned 1B tokens in 30 days with Hermes; thread on what he built
Wired the Kanban view to Hermes for an automated short-form content pipeline for his app
Complaints and issues
- Antivirus false positives quarantining bundled uv.exe on Windows (acknowledged in README)
- "It broke all of my pipelines. Now I'll have to adapt my scripts again" (on v0.22) @ProDrifterDK ·
- "When are we getting always-on agents???" @schlekx ·
- One agent is "stuck on v0.21 and refuses to upgrade … Could not resolve the stable source channel" @DougVos ·
- The round-robin across free search providers "didn't work for me in two recent Hermes installations" @richardjhd ·
- Update failed · GitHub issue
- Asks why Hermes "*requires* the built in memory provider remain enabled when using a different provider" @hypermageia ·
- Desktop wishlist ("a mobile app. this was the loudest ask by far") @witcheer ·
- Setup on Omarchy: "it pops up the browser window BEHIND the terminal" @jetpen ·
- muse-spark turns end mid-task on finish_reason=stop · GitHub issue
- Hermes shows passwords in Telegram chat (11 reactions) · GitHub issue
FAQ
Do I need a Nous subscription?
No. It works with any provider; Nous Portal is an optional single-login alternative to collecting separate keys.
raw.githubusercontent.com/NousResearch/hermes-agent/HEAD/README.md · checked
Does it run on Windows?
Yes, natively via the PowerShell installer; WSL2 also works.
raw.githubusercontent.com/NousResearch/hermes-agent/HEAD/README.md · checked
OpenClaw★ 392k · MIT
What it is
Personal AI assistant that lives on your own machine, with a local Gateway that connects channels, tools, devices and swappable model harnesses.
GitHub repo · openclaw.ai · @openclaw
Install
curl -fsSL https://openclaw.ai/install.sh | bashWindows: iwr -useb https://openclaw.ai/install.ps1 | iexnpm install -g openclaw@latest (Node 24.16+/26.1+), then openclaw onboard --install-daemonDocker, Nix (docs)
raw.githubusercontent.com/openclaw/openclaw/HEAD/README.md · checked
Models
Hosted and local model providers; models and agent harnesses (Claude, Codex, local models) are swappable plugins.
raw.githubusercontent.com/openclaw/openclaw/HEAD/README.md · checked
Cost
Free (MIT); you pay your model provider.
raw.githubusercontent.com/openclaw/openclaw/HEAD/README.md · checked
Features
- Gateway control plane for sessions, tools, events and channels (iMessage etc.) raw.githubusercontent.com/openclaw/openclaw/HEAD/README.md
- Companion apps/nodes for voice, Canvas, camera, screen raw.githubusercontent.com/openclaw/openclaw/HEAD/README.md
- Run by the OpenClaw Foundation, a 501(c)(3) nonprofit x.com/openclaw · user report
Use cases
Agent 'Neura' made a booking in Dubai on his behalf using voice and calling tools
Opus orchestrating spare Codex resets "through Herdr, T3, Hermes, OpenClaw, whatever works"
Runs a custom automated research and briefing workflow
"Yap-to-content": 30–40 minutes of voice notes to Telegram, and "My OpenClaw agent structures my thoughts"
A consultant who has "created hermes and openclaw for clients (they just know it's an agent)"
"My @openclaw agent called a Japanese restaurant to make a reservation … With @twilio and @ChatGPT voice"
Moved most of his work onto agents, OpenClaw first
Mac assistant connected to iMessage, browser and phone; automated voice calls to restaurants via OpenAI realtime + Twilio
Complaints and issues
- On hosting: "what breaks is rarely the model, its the box: ram blows up mid build, disk fills, logins expire" @kylianCros ·
- Upgrade advice from a user: "Pin the version that works and turn off auto-update for now" @Sonofpeace0001 ·
- GPT-6.1 Sol was "explicitly advertised" in 2026.9.8, but users ended up "investigating their configuration" @TotoshiSensei ·
- "spent 700$ on a mac mini 8 months ago for my openclaw agent, used it for 2 months and haven't touched it since" @treeonchain ·
- "ive gone from openclaw to hermes and now moved everything to grok bot" @peer_rich ·
- "the one i used most was openclaw, and it was a nightmare to set up, even for someone as techy as me" @peduarte ·
- Friends who bought Mac minis "lost interests in about 1-2 months … tokens cost money" @sfbayKL ·
- Security: "doesn't make sense to use if you enjoy privacy, security, or accuracy" @wedtm ·
- "I gave up on tweaking my openclaw after a few months" @thekitze ·
- Codex app-server rewrites the prompt on tool-call turns, busting OpenAI prompt cache · GitHub issue
- Live docs are ahead of the release · GitHub issue
- Bootstrap files in agentDir silently ignored · GitHub issue
FAQ
Where does my data live?
State, memory and credentials stay on your hardware; prompts go only to the model provider you choose.
raw.githubusercontent.com/openclaw/openclaw/HEAD/README.md · checked
Can I use plain npm install from source?
No. The repo is a pnpm workspace; use pnpm install/build.
raw.githubusercontent.com/openclaw/openclaw/HEAD/README.md · checked
OpenCode★ 214k · MIT
What it is
Open-source terminal and desktop coding agent with built-in build/plan agents (repo moved from sst/opencode to anomalyco/opencode).
GitHub repo · opencode.ai · @opencode
Use-case posts tagged possible_campaign came in a cluster of near-identical StepFun Step 5 Preview launch posts on 2026-10-08/09.
Install
curl -fsSL https://opencode.ai/install | bashnpm i -g @opencode/cli@latestbrew install anomalyco/tap/opencodescoop/choco install opencodebrew install --cask opencode-desktop
raw.githubusercontent.com/anomalyco/opencode/HEAD/README.md · checked
Models
Many providers; OpenCode Go offers curated models such as Kimi K3, Kimi K2.7 Code, GLM-5.3-Flash, DeepSeek V4.1 Flash, Qwen3.7 Plus, MiniMax M3, Claude Haiku 5.5, GPT 6 Luna.
opencode.ai/go · checked
Cost
Free (MIT) with BYO keys. OpenCode Go costs $10/month; Go Plus costs $40/month with higher limits.
opencode.ai/go · checked
Features
- Two built-in agents switchable with Tab raw.githubusercontent.com/anomalyco/opencode/HEAD/README.md
- Desktop app as well as TUI raw.githubusercontent.com/anomalyco/opencode/HEAD/README.md
Use cases
Tips on DeepSeek's direct API vs OpenCode Go: "if i'm spending around $20–25/month on deepseek, i'd stay on the direct api"
A live benchmark of OpenCode Go models
Exploring orchestrator plus subagents using open models (cloud orchestrator, local subagents)
2 posts flagged as a possible coordinated campaign (hidden by default)
Built a free Adobe Acrobat alternative (merge, sign, redact PDFs) from one message to Step 5 Preview in 67 min
Built a 3D landing page with ~9,000 cursor-chasing particles on OpenCode's free week
Complaints and issues
- "paid for the $10/month @opencode Go sub & tried Qwen 3.8 Max, burnt up my 5 hr limit & 40% of weekly usage in just 30 min" @tphuang ·
- A workspace UI change with "basically no info on why the ui changed" @xammen_ ·
- "OpenCode is one huge bug. If not for the OpenCode Go subscription, it would already be in my garbage bin." @burkov ·
- "why is there insufficient quota ? My 5 hour limit is still at 3%" @bxnjxmin27 ·
- "slow, stuck, stop the answer in the middle" (with a specific model) @AlonLaleza30319 ·
- Asks to revert silent removal of Go privacy wording and provider attribution (60 reactions) · GitHub issue
- Users want the legacy layout kept (82 reactions) · GitHub issue
- GPT-5.6 Luna returns model not found with ChatGPT OAuth (109 reactions) · GitHub issue
FAQ
What does OpenCode Go cost?
$10/month (Go) or $40/month (Go Plus), with per-model request limits.
opencode.ai/go · checked
OpenHands★ 91.1k · MIT
What it is
Self-hosted control center ('Agent Canvas') for coding agents and automations; runs OpenHands or ACP agents like Claude Code, Codex, Gemini CLI, OpenCode.
GitHub repo · openhands.dev · @OpenHandsDev
Install
Desktop/CLI per docs (Quickstart); local GUI, TUI, CLI and SDK
raw.githubusercontent.com/OpenHands/OpenHands/HEAD/README.md · checked
Models
Any LLM (BYO model); can also drive third-party ACP agents.
raw.githubusercontent.com/OpenHands/OpenHands/HEAD/README.md · checked
Cost
Open source is free. Cloud Individual is free, with BYOK or OpenHands models at cost. Enterprise is custom-priced (SaaS or self-hosted VPC).
openhands.dev/pricing · checked
Features
- Prebuilt automations (e.g. reports to Slack, decomposing GitHub issues) raw.githubusercontent.com/OpenHands/OpenHands/HEAD/README.md
- Switch between local, remote and cloud backends raw.githubusercontent.com/OpenHands/OpenHands/HEAD/README.md
Use cases
An OpenHands PR prompted a session-ID validation change in his CodeFlowMu project
Does development overnight with OpenHands agent loops
Complaints and issues
- The same user later: "writing code for both claude agent sdk / codex sdk would have been better than trying to make openhands work" @ayushunleashed ·
- "it malfunctions and gets stuck in between the work manytimes. Openhands feels really unreliable." @ayushunleashed ·
- Mac installer produces a damaged binary · GitHub issue
- Agent Canvas extremely slow with more than 3 conversations; ECONNREFUSED · GitHub issue
FAQ
Open source vs Cloud?
Open source is single-user and local; Cloud adds hosted runtime, APIs, Jira/Slack integrations.
openhands.dev/pricing · checked
Cline★ 70.4k · Apache-2.0
What it is
Autonomous coding agent available as a VS Code/JetBrains extension, CLI, SDK and (since Sept 2026) a desktop app.
GitHub repo · cline.bot · @cline
X account: already in x-accounts.json
Install
npm i -g clineVS Code Marketplace / JetBrains Marketplacenpm install @cline/sdk
raw.githubusercontent.com/cline/cline/HEAD/README.md · checked
Models
OpenAI, Anthropic, Google, OpenRouter, Vercel AI Gateway, Azure/Vertex, Cerebras/Groq, Ollama/LM Studio, any OpenAI-compatible API.
raw.githubusercontent.com/cline/cline/HEAD/README.md · checked
Cost
Free for individuals; pay for inference at cost or BYOK. Enterprise tier for teams (price not listed).
cline.bot/pricing · checked
Features
- .clinerules project rules raw.githubusercontent.com/cline/cline/HEAD/README.md
- MCP support; terminal execution with live output raw.githubusercontent.com/cline/cline/HEAD/README.md
- Cline Desktop with free open-weight models via ClinePass (third-party report) x.com/stretchcloud/status/2099666257217790431 · user report
Use cases
Built a Chrome extension in under 5 minutes to re-enable right-click on a bank site, using Muse Spark 1.3
Grasper, built at a Cline hackathon, "blocks fake installs before they run"
"made this sci-fi bunker in @threejs using union alpha in @cline took approximately 1 hour and 580M tokens"
Tested the agent workflow with early access to Cline Desktop
1 post flagged as a possible coordinated campaign (hidden by default)
Built an interactive 3D mechanical honeybee in three.js from one prompt with Step 5 Preview
Complaints and issues
- "genuinely curious what the abuse pattern was" @bangtanscenez ·
- "put out free samples at costco and someone brought a wheelbarrow" @garageweights ·
- Token burn on local models: "an agent harness burn 20,000 tokens before writing a single line of code" @Aaleks_crypt ·
- "You advertise 'Kimi Free,' yet we run through three messages and the quota is already used up." @velsinhax ·
- Impossible token count in context window · GitHub issue
- Browser automation fails with base64 image URL error · GitHub issue
- Switching Plan to Act clears unsent prompt text · GitHub issue
FAQ
Is Cline free?
Yes for individuals; you pay only for inference, or bring your own keys.
cline.bot/pricing · checked
Open Interpreter★ 68.6k · Apache-2.0
What it is
Now a Codex fork tuned as a coding agent for low-cost open models (Kimi K3, GLM 5.3).
GitHub repo · openinterpreter.com · @OpenInterpreter
Install
curl -fsSL https://www.openinterpreter.com/install | shirm https://www.openinterpreter.com/install.ps1 | iex
raw.githubusercontent.com/openinterpreter/openinterpreter/HEAD/README.md · checked
Models
Any OpenAI-compatible provider via Chat Completions; switch with /model.
raw.githubusercontent.com/openinterpreter/openinterpreter/HEAD/README.md · checked
Cost
Free (Apache-2.0); BYO provider.
raw.githubusercontent.com/openinterpreter/openinterpreter/HEAD/README.md · checked
Features
- Speaks the Codex exec protocol / SDK raw.githubusercontent.com/openinterpreter/openinterpreter/HEAD/README.md
- QA skill that drives web apps in a real browser raw.githubusercontent.com/openinterpreter/openinterpreter/HEAD/README.md
Use cases
No first-hand use cases found in the window we searched.
Complaints and issues
goose★ 55.3k · Apache-2.0
What it is
Open-source, extensible AI agent (desktop app, CLI, API) that installs, executes, edits and tests with any LLM; repo moved from block/goose to aaif-goose/goose.
GitHub repo · goose-docs.ai · @goose_oss
Install
Desktop app for macOS/Linux/Windowscurl -fsSL https://github.com/aaif-goose/goose/releases/download/stable/download_cli.sh | bash
raw.githubusercontent.com/aaif-goose/goose/HEAD/README.md · checked
Models
15+ providers (Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure, Bedrock…), or existing Claude/ChatGPT/Gemini subscriptions via ACP.
raw.githubusercontent.com/aaif-goose/goose/HEAD/README.md · checked
Cost
Free (Apache-2.0); BYO key or subscription.
raw.githubusercontent.com/aaif-goose/goose/HEAD/README.md · checked
Features
- Custom distributions with preconfigured providers, extensions and branding raw.githubusercontent.com/aaif-goose/goose/HEAD/README.md
- MCP extensions x.com/vivekkmkpinn/status/2093786722366537765 · user report
Use cases
A Japanese-language user ran goose with a local Qwen3.8-Flash-Next served from Strata
"We embedded @goose_oss into our platform, connected the platform through MCP, and made it accessible from the web using ACP."
Rolled goose out internally for offline use; 10 users after one month
Built a retro-style note-taking app in Godot by training his own agent on goose
Complaints and issues
- "I've been using goose recently it works but meh. Hermes works" @SterlingCooley ·
- Requests fine-grained regex shell permissions · GitHub issue
- Hard to configure context size in the desktop app · GitHub issue
FAQ
Can I use my ChatGPT/Claude subscription?
Yes, via ACP.
raw.githubusercontent.com/aaif-goose/goose/HEAD/README.md · checked
Aider★ 49.6k · Apache-2.0
What it is
AI pair programming in your terminal that commits each change to git.
No official Aider X account verified (@Aider_AI: 27 followers, unverified). Creator Paul Gauthier is @paulgauthier.
Install
python -m pip install aider-install && aider-install
raw.githubusercontent.com/Aider-AI/aider/HEAD/README.md · checked
Models
Almost any LLM, including local models; examples for DeepSeek, Anthropic, OpenAI.
raw.githubusercontent.com/Aider-AI/aider/HEAD/README.md · checked
Cost
Free (Apache-2.0); BYO key.
raw.githubusercontent.com/Aider-AI/aider/HEAD/README.md · checked
Features
- Every edit is a git commit you can undo x.com/Renderhaus_/status/2092912875677618540 · user report
Use cases
Fixed a pagination bug in a Python service with Aider, keeping every step as an undoable commit
Complaints and issues
Crush★ 28.7k · FSL-1.1-MIT
What it is
Terminal coding agent from Charm with multi-model support across many OSes.
Install
brew install charmbracelet/tap/crushnpm install -g @charmland/crushwinget/scoop/apt/yum/pkggo install github.com/charmbracelet/crush@latest
raw.githubusercontent.com/charmbracelet/crush/HEAD/README.md · checked
Models
Many LLMs plus any OpenAI- or Anthropic-compatible API; Hyper is Charm's official subscription provider.
raw.githubusercontent.com/charmbracelet/crush/HEAD/README.md · checked
Cost
Free to use with BYO keys. Hyper is subscription-based with a free tier.
Unverified: Hyper price not captured (charm.land/hyper returned 404)raw.githubusercontent.com/charmbracelet/crush/HEAD/README.md · checked
Features
- Runs on macOS, Linux, Windows, Android, FreeBSD, OpenBSD, NetBSD raw.githubusercontent.com/charmbracelet/crush/HEAD/README.md
Use cases
No first-hand use cases found in the window we searched.
Complaints and issues
- Shift+Enter for newline doesn't work as expected · GitHub issue
- Wants configurable auto-summarize threshold · GitHub issue
Kilo Code★ 27.6k · MIT
What it is
Open-source coding agent for VS Code, JetBrains, CLI and a web cloud agent, with 500+ models.
GitHub repo · kilo.ai · @kilocode
X account: profile says acquired by Anaconda (unverified elsewhere)
Install
VS Code Marketplacenpm install -g @kilocode/clicurl -fsSL https://kilo.ai/cli/install | bashbrew install Kilo-Org/tap/kiloJetBrains Marketplace
raw.githubusercontent.com/Kilo-Org/kilocode/HEAD/README.md · checked
Models
500+ models with mid-task switching; BYOK or local.
raw.githubusercontent.com/Kilo-Org/kilocode/HEAD/README.md · checked
Cost
Free tier ($0, BYOK or local). Kilo Gateway is $0/month plus usage at exact provider rates. Kilo Pass starts at $19/month (up to 50% bonus credits). Teams cost $15/user/month.
kilo.ai/pricing · checked
Features
- Kilo Marketplace for agents, skills, MCP servers, plugins raw.githubusercontent.com/Kilo-Org/kilocode/HEAD/README.md
- CLI is an OpenCode fork with built-in sandboxing (user report) x.com/artillain/status/2095035744402518393 · user report
Use cases
Persistent memory with the Honcho plugin
Comparison test: "Kilo Code + Opus 5.5: 21.8 min, 87 tool calls, $4.29 omp + Opus 5.5: 22.8 min, 73 tool calls, $4.39"
Moved a day's workflow from Cursor to Kilo Code to test it on a real task list (thread)
Uses the Kilo CLI with high-visibility diff views before edits
Used GLM heavily on Kilo Code
Complaints and issues
- "Does it need kilo cli to run kilo desktop?" @hazeem_dev ·
- "Why are you creating worktrees without permission of user? … had to uninstall for now" @raghav_lrkp ·
- Auto-compaction fails for local Ollama 32K models · GitHub issue
- Model hits output limit while reasoning and produces no output (12 reactions) · GitHub issue
- Compaction experience is 'horrible' · GitHub issue
FAQ
Can I use it without paying Kilo?
Yes: the Free tier runs BYOK or local models with no hosted credit.
kilo.ai/pricing · checked
OpenBot★ 579 · PolyForm-Noncommercial-1.0.0
What it is
Local-first desktop workspace for persistent AI teammates: runs Codex, Claude, Grok, OpenCode, Gemini, Cursor and Cline agents with their own workspaces, queues, file sharing and an embedded browser.
GitHub repo · openbot.run · @OpenBot_
X account: verified account; profile links openbot.run, which links github.com/nightly-labs/openbot. @openbotrun (created 2026-09-20, 3 followers) copies the bio; not treated as official.
Name clash: CopilotKit/OpenBot (copilotkit.ai/openbot, an open-source "AI coworkers" template, about 6.5k stars) is a different project. Earlier data pointed at CopilotKit by mistake; corrected 2026-10-16 after tracing @OpenBot_ to openbot.run to github.com/nightly-labs/openbot.
Install
Desktop app: macOS 13+ (Apple silicon/Intel), Windows 10+ x64, Linux x64/arm64 AppImage (GitHub releases)
raw.githubusercontent.com/nightly-labs/openbot/HEAD/README.md · checked
Models
Uses your existing agents and subscriptions: Codex App Server, Claude Code, Grok CLI, plus OpenCode, Gemini, Cursor CLI and Cline CLI over ACP; custom OpenAI-compatible endpoints and local servers (Ollama, LM Studio).
raw.githubusercontent.com/nightly-labs/openbot/HEAD/README.md · checked
Cost
Free to use under a non-commercial license; runs on the subscriptions you already pay for (per the @OpenBot_ bio).
Unverified: no paid plan or commercial-license price foundx.com/OpenBot_ · checked
Features
- Persistent agents with independent sessions and local workspaces raw.githubusercontent.com/nightly-labs/openbot/HEAD/README.md
- FIFO message queues, agent-to-agent messages and managed file transfers raw.githubusercontent.com/nightly-labs/openbot/HEAD/README.md
- Persistent embedded browser agents can control; optional Computer Use on macOS, Windows and Linux raw.githubusercontent.com/nightly-labs/openbot/HEAD/README.md
- Development preview: agents run with danger-full-access and approvalPolicy never after first-launch consent raw.githubusercontent.com/nightly-labs/openbot/HEAD/README.md
Use cases
Tips for applying Grok Bot workflows to OpenBot: "Give it skills (crucial)"
A user who likes it the most "as alternative to GrokBot + using private models from @AskVenice"
Complaints and issues
- A request to "publish their team setups on the marketplace" @Its_Vivaann ·
- A Portuguese-language user says "Eu usei openbot e gasta MUITO" ("I used openbot and it burns A LOT") @ErickMath1as ·
- "OpenBot is not local-first... not private..! Why do I need to give my personal details?" @DerekColley_ ·
FAQ
Is OpenBot open source?
The code is public, but it is licensed under PolyForm Noncommercial 1.0.0, which allows non-commercial use only. That is source-available, not OSI open source.
raw.githubusercontent.com/nightly-labs/openbot/HEAD/README.md · checked
Does it run offline?
No. It is local-first, but each agent connects to its own provider (OpenAI, Anthropic, xAI, Google, Cursor, Cline).
raw.githubusercontent.com/nightly-labs/openbot/HEAD/README.md · checked
Shut down / superseded
README: 'The Roo Code Extension was shut down on May 15th.' Suggests ZooCode (community fork) and Cline.
Last release v3.54.0 · May 15, 2026 · raw.githubusercontent.com/RooCodeInc/Roo-Code/HEAD/README.md
README: repository no longer actively maintained; final 2.0.0 release. X profile name reads 'Continue (acquired by Cursor)'.
Last release v2.0.0-vscode · Jun 19, 2026 · raw.githubusercontent.com/continuedev/continue/HEAD/README.md
Research agent from Princeton/Stanford that takes a GitHub issue and tries to fix it with your LM of choice.
Last release v1.1.0 · May 22, 2025 · github.com/SWE-agent/SWE-agent