octo-agent CLI
Self-hosted AI agent in a single Go binary: terminal, web UI, chat apps and desktop all run on any compatible model
- 142 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent runs shell commands and edits files, and the chat bridges let you hand it tasks remotely
- Installation uses a script that is downloaded and run from the network, and the installers are unsigned for now
- Project context and tool calls go to the model you chose through its API
Install
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/octo-agentDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent runs shell commands and edits files, and the chat bridges let you hand it tasks remotely
- Installation uses a script that is downloaded and run from the network, and the installers are unsigned for now
- Project context and tool calls go to the model you chose through its API
This is third-party code. Review the repository files before installing.
What it does
octo-agent is an AI agent in a single Go binary of about 40 MB, with no Node, Python or Ruby. Shell, file read, edit and search tools, MCP servers, skills and subagents are on out of the box. You pick the model: DeepSeek, Kimi, Qwen, Anthropic, OpenAI or any address compatible with their protocols. The server and data stay on your machine, there is no telemetry, and the only call per launch is a version check on GitHub that `update_check: false` turns off. The same binary gives eight entry points: a terminal interface, the `octo "prompt"` command for a single pass, a web UI via `octo serve`, a desktop app, a bridge to WeChat, Feishu, DingTalk, WeCom, Discord and Telegram, VS Code and Obsidian extensions and a mobile app in developer preview. Skills in the SKILL.md format are compatible with Claude Code: symlink `~/.claude/skills` and reuse them. To avoid losing files, deletes and overwrites are copied to a recycle bin first (14 days and 10 GiB by default), and commands like `rm -rf /` are hard-denied. An optional sandbox is available: Seatbelt on macOS and Landlock on Linux.
Who it is for. Developers and small teams that want their own agent on their own server with a choice of models and no single-vendor lock-in.
Good fit when
- An agent has to move to a server as one file, without a Node or Python dependency tree
- The model is chosen apart from the tool: DeepSeek or Kimi today, another one tomorrow
- Tasks arrive through Telegram, Feishu or WeChat while the work runs on a home computer
- Scripts and cron run `octo "prompt"` as a one-off agent pass
- The agent's deletions go through a recycle bin so erased files can be restored
Not a fit when
- You already have reliable access to a Codex or Claude subscription: the author himself says to keep using them
- You need a signed build: the installers are unsigned for now and releases are verified by hash
- You need a ready skill or MCP server for your own agent: this is a whole agent, not an add-on
- You need the iOS mobile app from the store: it builds from source and store builds are not out yet
Example request
Add a --json flag to the octo config show command and run the testsLimitations
The macOS and Windows installers are unsigned for now, so the system shows Gatekeeper and SmartScreen warnings; release hashes are described in SECURITY.md. The mobile app is in developer preview: Android installs from an APK and iOS builds from source. Bridges to WeChat, Feishu, DingTalk, WeCom, Discord and Telegram need separate channel setup. An access key is skipped only on the loopback address, so set up protection separately when exposing the server outside.
How to disable. Stop the server, then delete the `octo` binary and the user settings directory. For the desktop app, use the normal uninstall.
FAQ
How do I install octo-agent?
On Linux and macOS run `curl -fsSL https://octo-agent.dev/install.sh | sh`, on Windows `irm https://octo-agent.dev/install.ps1 | iex`. For Go, `go install github.com/open-octo/octo-agent/cmd/octo@latest` works. The binary updates with `octo upgrade`.
How do I connect a model?
Run `octo serve -d`, open http://127.0.0.1:8088 and go through the setup panel: language, model, a short chat about your preferences. In the terminal `octo config` does the same.
Do Claude Code skills work?
Yes, the SKILL.md format is compatible. Symlink the `~/.claude/skills` directory and existing skills load without rewriting.
Does octo send data elsewhere?
There is no telemetry. Without your request only one version check call goes to GitHub, and the `update_check: false` setting turns it off. What remains is calls to the model you configured and the tool calls you asked for.
Can I restore a file the agent deleted?
Yes, ordinary deletes and overwrites through `write_file` or `edit_file` go to a recycle bin first. By default it keeps data for 14 days and up to 10 GiB.
Discussion
No questions or reviews yet.
Related
CLI toolsA Go CLI that configures the AI coding agents you already use: it gives them shared memory, a workflow and deterministic steps without replacing the agents them
A local-first AI agent running on your own machine via llama.cpp that drives a browser, edits files, runs approved commands and remembers context across session
An open coding agent for the terminal, browser and desktop that works with models from OpenAI and Claude to Ollama
A Go coding agent for open models: it turns a chat into tasks on separate branches, checks and merges the result and runs all projects in one window
General-purpose AI agent in one Go binary: console, ACP editors, web UI, Telegram, a cron scheduler and MCP support
A self-hosted AgentOS for one workspace: plan-mode chat, private knowledge base, MCP tools, skills, memory and a sandbox
Comparison with similar tools
Cards that share the main search query, and how they differ.
README in short
The README explains the positioning: an agent for people who want to download and work without long setup, with a choice of model and data on their own machine. It then lists the strengths (one binary, prompt caching, zero telemetry, eight interfaces), core features, installation, first run and a layer diagram from interfaces down to tools. At the bottom are links to the documentation on octo-agent.dev.