Gptme
Your agent in your terminal, equipped with local tools: writes code, uses the terminal, browses the web. Make your own persistent autonomous agent on top!
Last verified:
What is Gptme?
gptme is an open-source personal AI assistant and agent platform that runs in your terminal and in a web UI, designed to act as an intelligent copilot for coding, shell work, web browsing, and general automation. It exposes a chat-driven interface that can execute shell commands, run Python, edit files, browse and screenshot web pages, and operate as subagents for longer visual workflows. The project targets developers, power users, and teams who want a programmable assistant tightly integrated with local tools and repositories, especially those who prefer working from the terminal or want reproducible, scriptable AI interactions. Key features include a rich toolset (shell, Python/IPython, browser/read URL, computer automation for screenshots and UI interactions), slash-commands and profiles for controlling behavior, non-interactive scripting modes, and integrations like pre-commit hooks and configurable tool selection for safer automation.
Gptme pricing
Pricing model: Freemium
gptme is free and open-source software distributed under an open-source license; there is no proprietary hosted pricing on the documentation site. Users must provide and pay for their own LLM API access (API keys) from supported providers, and any usage costs come from those provider plans rather than gptme itself. Installation and the core feature set are available without paid tiers, while optional third-party services (LLM provider accounts or extras that require additional dependencies) may have separate fees.
Gptme pros
- Runs in terminal and web UI, fits developer workflows
- Rich built-in tools: shell, Python, browser, vision, file editing
- Can execute shell commands directly from chat
- Runs Python code and saves results to files
- Browser tool can read URLs and take screenshots
- Supports non-interactive mode for scripting and automation
- Subagent profiles let you isolate long visual tasks
- Slash-commands for quick control of common actions
- Token and model reporting (tokens, models commands)
- Integrates with pre-commit to catch and fix issues automatically
- Configurable toolset with --tools to minimize permissions
- Supports piping multiple prompts with multiprompt syntax
- Can skip confirmation prompts for automated runs (--no-confirm)
- Saves API keys in config and prompts for missing keys during setup
- Installable via pipx or Homebrew, easy CLI installation
Gptme cons
- Requires external LLM API keys/configuration to function
- Some interfaces or extras are WIP and may lack polish
- Desktop computer automation requires extra permissions and setup
- Browser/vision features need optional extras installed
- Non-interactive automation can be dangerous if misconfigured
- Not a hosted SaaS — user is responsible for environment and costs
- Advanced features like subagents may increase local complexity
- Limited GUI compared to full-featured commercial copilots
Frequently asked questions about Gptme
How do I install gptme?
You can install gptme locally using a Python package manager such as pipx or via system package managers; the docs recommend installing with pipx (pipx install gptme) or using uv tool install gptme, followed by running gptme to start an interactive chat.
Which tools does gptme provide out of the box?
gptme ships with a variety of tools including a Shell tool to run shell commands, a Python/IPython executor to run code and write files, browser/read URL capacity for web pages and screenshots (with extras), image processing/vision tools, and file editing capabilities that allow the assistant to create and patch files in your workspace.
How do I supply an LLM API key?
During first run, if no API key is set in the environment or config, gptme will prompt you to enter one and will save it into its configuration file; you can also set API keys via environment variables as described in the getting started docs.
Can gptme modify files in my repository and run pre-commit checks?
Yes — gptme will include mentioned files as context when asked to modify them, it applies patches to files, and if a .pre-commit-config.yaml exists it will run pre-commit checks automatically after modifications (configurable and can be disabled via environment toggles).
How do I control which tools the assistant can use?
Use the --tools flag to specify exactly which tools are available to the assistant (for example --tools read,append) or to add tools on top of defaults using --tools +computer, enabling safer, least-privilege workflows for automation and testing.
What is non-interactive mode and when should I use it?
Non-interactive mode (--non-interactive) runs prompts and terminates when finished, which is useful for scripted tasks and automation; it implies --no-confirm so actions proceed without interactive confirmations, making it suitable for CI or batch runs but requiring careful use to avoid unintended changes.
Does gptme support visual desktop automation?
Yes — with the extras installed, gptme can perform desktop-style automation via a 'computer' tool that can click, type, and take screenshots; for long visual tasks it encourages using subagent profiles to avoid bloating parent context with repeated screenshots.
How do I undo changes or view conversation history?
gptme provides slash-commands like /undo [n] to undo recent actions and /log to display the conversation log; it also includes an /edit command to edit the conversation in your editor and other utility commands such as /tokens and /models to inspect usage.
Can I run gptme in CI or automation pipelines?
Yes — use the --non-interactive flag to run scripted prompts in a one-shot fashion and combine it with --no-confirm to skip confirmations; this mode is intended for automation and is explicitly noted as suitable for scripting workflows.
What safety or confirmation controls exist?
gptme includes confirmation prompts by default and a --no-confirm flag to skip them when you trust the automation; you can also restrict available tools with --tools, run in isolated subagents for risky visual tasks, and configure environment toggles for behaviors like pre-commit checks.