gptme
gptme is an open-source personal AI agent under MIT that runs anywhere a terminal runs, from a laptop to a headless server or a CI pipeline. It executes shell and Python locally, reads, writes and patches files, browses the web, and can run in a Docker or Wasmtime sandbox. It works with Anthropic, OpenAI, Google, xAI, DeepSeek and OpenRouter, or fully locally through llama.cpp, and it is both an MCP client and server.
What does gptme do?
gptme describes itself as "a personal AI agent that runs anywhere a terminal runs", including SSH sessions, tmux, headless servers and CI pipelines, and as "a great coding agent, but general-purpose enough to assist in all kinds of knowledge-work." Its tools are local: shell and ipython execute code, read, save, append, patch and morph handle files, a browser tool uses Playwright, and vision, screenshot, computer, tmux, gh, subagent and rag round out 17 named tools. Version 0.33.0 added sandboxed Python and shell execution through Docker or Wasmtime.
Providers listed are Anthropic, OpenAI, Google, xAI, DeepSeek and more, OpenRouter for over 100 models, and "fully local via llama.cpp", with the note that local models need no key. It can also sign in with an existing ChatGPT Plus or Pro plan or a SuperGrok plan. MCP runs in both directions: gptme is a client, and gptme-mcp-server exposes its shell, Python and file tools to Claude Desktop, Cursor and other clients. ACP support makes it a coding agent inside Zed and JetBrains, and there is a vim plugin, a web UI with a hosted instance, a desktop app, a non-interactive mode with JSON output for CI, and systemd scaffolding for scheduled agents.
Key facts
- Licence: MIT, copyright Erik Bjäreholt, with no added conditions.
- 17 tools named in the README: append, browser, chats, choice, computer, gh, ipython, morph, patch, rag, read, save, screenshot, shell, subagent, tmux and vision.
- Providers: Anthropic, OpenAI, Google, xAI, DeepSeek and more; OpenRouter for 100+ models; fully local through llama.cpp with no key; sign-in with a ChatGPT Plus or Pro or a SuperGrok subscription.
- Bidirectional MCP: a client, and gptme-mcp-server exposing its tools to Claude Desktop, Cursor and other clients.
- Sandboxed execution through Docker or Wasmtime since v0.33.0.
- ACP support for use inside Zed and JetBrains; a vim plugin; a web UI (gptme-webui) with a hosted instance at chat.gptme.org; a desktop app for Linux, macOS and Windows.
- Non-interactive mode and JSON output for CI; gptme service init scaffolds a systemd service for scheduled agents on Linux.
- Install: pipx install gptme, uv tool install gptme, or pip, with browser, tui, acp and all extras; Python 3.10 or newer. Latest tagged release when read: v0.33.0 on 2026-08-19; 4,417 stars on 2026-09-18.
What does it replace, and where does it fall short?
gptme replaces Claude Code as a terminal agent that runs the model of your choice, including a local one, and adds what a general assistant needs: a browser, screenshots, a desktop control tool and scheduling. Its systemd scaffolding and headless mode let it stand in for Devin on the always-on side, running tasks on a server on a schedule. It is the pick for a general-purpose local agent on the open-source alternatives to Claude Code page.
Where it falls short: it is a single-maintainer project at 4,417 stars, and its README's in-progress list still names tree-based conversations, automatic local context through RAG, advanced evals, the desktop app and the gptme.ai cloud as unfinished. Its terminal recordings date from 2023 by its own note. For a larger community behind a terminal agent, OpenCode.
How does gptme run?
pipx install gptme is the recommended route, or uv tool install gptme, or pip install gptme, with extras such as gptme[browser] for Playwright, gptme[acp] for editor integration, or gptme[all]. It needs Python 3.10 or newer and credentials for one provider: an API key, a ChatGPT Plus or Pro or SuperGrok login, OpenRouter through browser sign-in, which the README calls the fastest path with no card, or a local llama.cpp server with no key. gptme.ai is a managed cloud service in early access, and the README says the same stack stays self-hostable with gptme-server and gptme-webui.
Who is gptme for?
A developer or a technical user who wants one agent for coding and general work, running locally or on a server, with the model swapped freely and the ability to expose its tools to other agents over MCP. A team that needs a polished, widely used terminal agent should start with OpenCode or Crush.
What limits does the README state?
From the README: the terminal recordings are from 2023 and show the classic CLI; tree-based conversations, RAG for automatic local context, advanced evals, the desktop app and gptme.ai are listed as in progress; and its comparison table is self-authored.
Questions people ask
Is gptme open source?
Yes. gptme is MIT licensed with no added conditions, and its web UI and MCP server are part of the same open project. The gptme.ai managed service is a separate early-access offering, and the README says the same components stay self-hostable.
Can gptme run fully locally?
Yes. The README says local models via llama.cpp need no key, and that gptme runs anywhere a terminal runs, including headless servers and CI. Code execution can also be sandboxed in Docker or Wasmtime since version 0.33.0.
What tools does gptme have?
Seventeen named in its help: shell and ipython for execution, read, save, append, patch and morph for files, browser for the web through Playwright, vision and screenshot, computer for desktop control, tmux, gh for GitHub, subagent, rag, chats and choice. It also runs as an MCP server to lend those tools to other clients.
How does gptme compare with Claude Code?
gptme covers the same terminal coding loop with any provider or a local model, and adds browsing, screenshots, desktop control and scheduling that make it a general agent. Claude Code is Anthropic's product for Claude models. The open-source alternatives to Claude Code page compares it with OpenCode, Crush, Qwen Code and goose.
Can gptme run inside an editor?
Yes. ACP support makes it a coding agent in Zed and JetBrains IDEs, there is a vim plugin, and its MCP server exposes its tools to Claude Desktop, Cursor and other MCP clients. A web UI and a desktop app for Linux, macOS and Windows are also listed.
Sources
- gptme README and MIT LICENSE: github.com/gptme/gptme, read 2026-09-17.
Compared with the others
On the open-source alternatives to Claude Code page, gptme is the pick for a general-purpose local agent that also codes. MIT, shell and Python locally, file patching, a browser, screenshots, sandboxing in Docker or Wasmtime, llama.cpp for local models, and MCP as both client and server.
Also on that page: OpenCode for a terminal agent with a planning mode, Qwen Code for switching between hosted and local models at runtime, Crush for the widest provider list with local models.
On the open-source alternatives to Devin page, gptme is the pick for a scheduled agent on a server. MIT, with systemd scaffolding for scheduled autonomous agents, sandboxed execution in Docker or Wasmtime, and a browser, shell and file tools.
Also on that page: OpenHands for an always-on agent on your own infrastructure, goose for a general agent with a desktop app and an api, Cline for headless runs in ci and scheduled agents.
More coding assistants and agents
An MIT terminal coding agent with a full-access build agent and a read-only plan agent switched with the Tab key, installers for every platform, a beta desktop app, and the most-starred repository in this directory.
OpenAI's Apache-2.0 coding agent that runs locally in your terminal, with a desktop app and IDE extensions, signed in with a ChatGPT plan or an OpenAI API key, so the code is open and the models are OpenAI's.
Google's Apache-2.0 terminal agent for Gemini models with a free tier of 1,000 requests a day on a personal Google account, built-in search grounding, file and shell tools, MCP support, checkpointing and a GitHub Action.
A high-performance code editor from the creators of Atom and Tree-sitter, GPL-3.0-or-later with Apache-2.0 components, with an agent panel, external agents over ACP, edit prediction, MCP tools, sandboxing, and local models through Ollama or llama.cpp.
An MIT self-hosted control centre, called Agent Canvas, that runs the OpenHands agent or Claude Code, Codex, Gemini and any ACP-compatible agent across local, Docker and cloud backends, with automations triggered by schedules and webhooks.
An Apache-2.0 coding agent for VS Code, JetBrains, the terminal and a desktop app that edits files as reviewable diffs, runs commands with your approval, works with any model provider including Ollama, and connects tools through MCP.
Added September 17, 2026. Every claim above comes from the project's README, LICENSE or model card, read on September 17, 2026, or from the GitHub API on the date shown in the panel. Found an error? Write to reveneau@licheo.com and it is fixed in the next weekly pass. Repository: github.com/gptme/gptme.
Running one of these inside your own environment, with your own data and your own security rules, is the kind of work Reveneau does. Read how a forward deployed engagement works.