gptme is a free and open-source terminal AI agent that combines shell command execution, code writing, file editing, web browsing, and vision capabilities in a single interactive or autonomous workflow. Originally created in 2023 as one of the earliest terminal-based AI coding agents, it provides an unconstrained local alternative to Claude Code, Codex, and Cursor agents with full control over which LLM provider and model to use. The tool runs as a Python CLI that accepts natural language instructions, translates them into tool calls, and executes them locally with real-time streaming output. MCP discovery and dynamic loading allows gptme to connect to databases, APIs, and any MCP-compatible service.
Beyond interactive use, gptme is designed for persistent autonomous operation where AI agents run continuously across sessions. The reference implementation Bob demonstrates this capability with over 1,700 completed autonomous sessions — opening pull requests, reviewing code, fixing CI failures, managing task queues, maintaining behavioral lessons, and even posting on Twitter and responding on Discord. The plugin system introduced in v0.30.0 enables extensibility through custom tools, while background jobs support allows agents to monitor long-running processes. Editor integration with Zed and JetBrains IDEs turns gptme into a drop-in coding agent where the editor sends requests and gptme executes with its full toolset.
Licensed under MIT with over 4,300 GitHub stars, gptme maintains an active monthly release cadence with recent additions including form-based user input tools, improved context management, and the gptme-agent-template for bootstrapping custom autonomous agents. The project occupies a unique position as both a personal AI assistant and an autonomous agent platform — simple enough to use as a terminal copilot for quick tasks, yet powerful enough to run unsupervised production workflows. The Python codebase is available on PyPI for easy installation and supports all major LLM providers including OpenAI, Anthropic, and local models through various backends.