Agentica CLI
An open coding agent for the terminal, browser and desktop that works with models from OpenAI and Claude to Ollama
- 359 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The agent writes and edits files and runs shell commands, with the sandbox enabled by the --sandbox flag
- The gateway opens a web interface and takes instructions from messengers, so protect it with a password and do not expose the port
- Provider keys, sessions and memory sit in ~/.agentica and part of the context goes to the chosen model provider
Install
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/agenticaDetects 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 writes and edits files and runs shell commands, with the sandbox enabled by the --sandbox flag
- The gateway opens a web interface and takes instructions from messengers, so protect it with a password and do not expose the port
- Provider keys, sessions and memory sit in ~/.agentica and part of the context goes to the chosen model provider
This is third-party code. Review the repository files before installing.
What it does
Agentica is an open coding agent that installs with uv tool install agentica and starts with the word agentica in a terminal. The same agent is available in a browser and in a desktop app, which need the agentica[gateway] package. The agent reads and edits files through read_file and apply_patch, searches code, runs commands, browses the web and looks at images. The model is your choice, from OpenAI and Claude to DeepSeek and local Ollama, and /model switches it during a session. The agent remembers the project. Preferences live in AGENTS.md, project facts in long-term memory, and solutions it finds are compiled into SKILL.md for later similar tasks. For long dialogues, large command output is saved to a file, old tool results are cleared, and when the window fills up the agent writes the goal and unfinished work to a notes file and continues in a fresh window. /goal drives a long task, /worktree isolates the work and /resume brings a past session back. Several terminals become a team: the task sub-job reads code read-only, delegate starts a separate process and peer messages link two open sessions. Through the gateway the agent answers from messengers, from Telegram and Slack to WeChat and Feishu. There is a Python SDK, and a Node client, @agentica-ai/sdk.
Who it is for. For developers who want a coding agent tied to no single model vendor, and anyone who wants to steer sessions on their computer from a messenger.
Good fit when
- You want to work with DeepSeek, Qwen, Kimi or local Ollama the way you do with Claude Code or Codex
- A long task keeps hitting the context limit: you need big output saved to a file and the goal carried into a fresh window
- A session runs on your computer while you are away: through Telegram or Feishu, @list shows what is still running and you can send an instruction
- You want a desktop app and a browser interface with separate accounts, each with its own sessions and memory
- A Python project needs an embedded agent with tools, memory and MCP: the SDK fits
Not a fit when
- You only need Claude Code with its native subscriptions: Agentica has a different tool set and its own settings
- You cannot install an unsigned installer: the desktop builds are unsigned and macOS and Windows show a warning on first launch
- You have no key from any provider and no local model: the agent does not work without one
- Your system has no Python 3.10 or newer
Example request
Find out why the unit tests fail in this repository, fix it and record in the project rules how to run the testsLimitations
- A key from at least one model provider (OPENAI_API_KEY, ZAI_API_KEY and others) or a local model is required. The priority is environment variables, then .env, then config.yaml.
- The desktop installers are unsigned. On macOS you need to run sudo xattr -rd com.apple.quarantine /Applications/Agentica.app once, on Windows click to run anyway, and on Linux make the AppImage executable.
- The main docs are in Chinese; an English version is in README_EN.md.
- The WeChat, WeCom and Feishu integrations target Chinese messengers, while Telegram and Slack are supported too.
- Data lives in ~/.agentica: sessions, memory and configuration.
How to disable. Remove the CLI with uv tool uninstall agentica. The gateway is part of the same package if you installed it. Data, memory and sessions live in ~/.agentica; delete the folder when you no longer need them. The desktop app is removed the usual way for your system. docker compose down stops the Docker container.
FAQ
How do I install Agentica and start the web version?
Run uv tool install agentica for the console. For the web and desktop version run uv tool install "agentica[gateway]" and then agentica-gateway. The page opens at http://127.0.0.1:8881/chat. On first launch a default account is created and its random password is printed to the terminal.
Which models does Agentica support?
OpenAI over Chat Completions and the Responses API, DeepSeek, Claude, Zhipu GLM, Qwen, Kimi, Doubao and Ollama. The model switches with /model during a session. Several fallback models can be set for errors.
How do I restrict what Agentica can do?
Start the agent with the --sandbox flag: without it, it writes files and runs shell commands directly.
How do I control Agentica from a messenger?
Connect the gateway to WeChat, WeCom, Feishu, Telegram, QQ, DingTalk or Slack following the Gateway documentation. The message @list shows open sessions on the computer, and @session_name followed by text sends an instruction to that terminal.
How do I run Agentica in Docker?
Copy .env.docker.example to .env, set OPENAI_API_KEY and run docker compose up -d --build. The interface opens at http://127.0.0.1:8881/chat, data sits in a named volume and the current folder is mounted into the container as /workspace.
Discussion
No questions or reviews yet.
Related
CLI toolsA teaching terminal coding agent in Python, built as a "nanoGPT for agents": read it, take it apart and fork it
A tmux-based TUI that runs Claude Code, Codex and other agents in one list: status, quick prompts, worktrees and diff review
An open source coding agent for the terminal and desktop with build and plan modes
A multi-agent harness that verifies results by artifacts: stop hooks, an independent judge and an event log for Claude Code, Codex, Cursor and more
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
A self-hosted AgentOS for one workspace: plan-mode chat, private knowledge base, MCP tools, skills, memory and a sandbox
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| Agentica (this entry) | no | 359 | |
| OpenCode | no | 223,136 | a terminal agent on any model, without a messenger gateway or self-accumulating skills |
| Hermes Agent | no | 258,548 | a self-learning agent with messengers and cron jobs |
| Cline | no | 71,040 | an agent for VS Code, JetBrains and the terminal without a mobile gateway |
README in short
The main page, in Chinese, opens with five reasons to pick the project and a comparison with Codex on open tasks. The task, delegate and peer mechanisms get their own table, and a changelog closes the page.