Skip to content
← Tools / Skills
ToolErik Bjäreholt/10 trustMetrics-checked
gptme logo

gptme

Open-source personal AI agent that runs anywhere a terminal runs.

Visit official site →

About

gptme is an open-source personal AI agent by Erik Bjäreholt that runs anywhere a terminal runs: laptops, SSH sessions, tmux, headless servers, and CI pipelines. Started in spring 2023 (predating Claude Code), it is provider-agnostic and local-first, working with Anthropic, OpenAI, Google, DeepSeek, OpenRouter, Requesty, ChatGPT/Grok subscriptions, or fully local models via Ollama and any OpenAI-compatible server. It ships with shell, Python, web browsing (Playwright), vision, and plugin support, plus MCP integration and multi-agent orchestration.

Its distinguishing design is persistent autonomous agents whose memory lives in a git repo you own, with self-improving agent architecture, cross-harness memory utilities, and guardrails. The author later joined Lovable as a founding hire before leaving to focus on Superuser Labs full-time, and gptme remains in very active development. Limitations: as a community open-source project it has less polish and support than commercial agents, setup requires comfort with terminals and API keys, and its power comes with the usual risks of an unconstrained local agent.

What it can do

  • Persistent autonomous agents with git-backed memory
  • Shell, Python, web browsing, and vision tools
  • 100+ model providers plus fully local operation
  • MCP support and plugin system
  • Multi-agent orchestration and self-improvement

Setup

curl -sSf https://gptme.ai/install.sh | sh (auto-detects uv or pipx), or pipx install gptme / uv tool install gptme; configure a provider API key on first run.

Prerequisites

Python 3.10+; a terminal; an API key for a supported provider or a local model.

Use cases

ai-agentscoding-assistantterminalautomation

Works with

Trust score — 7/10

/10 trustMetrics-checked

Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 68; Trending score 70/100 from researcher assessment.

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.