gptme is one of the earliest open-source terminal coding agents, first released in Spring 2023 before the category had a name. It combines code execution, file editing, shell access, and web browsing via Playwright into a single persistent conversation interface, giving developers a self-contained autonomous assistant that runs entirely from the command line. The project supports local models through llama.cpp integration, making it suitable for air-gapped or privacy-sensitive environments, and it has been extended with MCP support and IDE integrations for Zed and JetBrains.
Key capabilities
Playwright browser automation - gptme ships built-in browser automation using Playwright, enabling agents to navigate web pages, extract content, fill forms, and interact with browser-rendered applications as part of a coding or research task without any separate setup.
Local model support via llama.cpp - Unlike cloud-only agents, gptme can route requests to locally running models through llama.cpp. This allows fully offline operation with no API keys, no usage costs, and no data leaving the machine.
IDE drop-in integration - The agent functions as a drop-in assistant inside Zed and JetBrains IDEs, extending its terminal-first design to developers who prefer editor-embedded workflows without requiring a purpose-built IDE extension.
MCP support - gptme implements the Model Context Protocol, allowing it to connect to MCP-compatible tool servers and expand its capabilities through the same ecosystem used by Claude Code, Cursor, and other modern agents.
Autonomy level
gptme operates at autonomy level 3: it can plan and execute multi-step tasks involving file edits, terminal commands, and browser interactions with moderate autonomy, but it surfaces its steps in the persistent conversation for the developer to follow and intervene. It does not operate fully hands-off on long-horizon tasks; the conversational loop is designed for interactive collaboration rather than unattended execution.
Strengths
- Early pioneer with over three years of real-world use and a long track record of reliability
- Playwright browser automation is built in with no additional configuration required
- Local model support via llama.cpp enables fully offline, private, and cost-free operation
- Zed and JetBrains IDE integration extends the tool beyond the terminal
- MCP support connects gptme to the broader tool ecosystem
- MIT license with no usage fees or subscription requirements
- Broad model support spanning Claude, GPT-4o, Gemini, Llama, and DeepSeek
Limitations
- Smaller community and star count than mainstream tools such as Aider or Claude Code
- User experience is less polished than newer agents designed with onboarding in mind
- Documentation is oriented toward contributors and power users rather than first-time adopters
- No native web UI or GUI; strictly terminal-based outside of IDE integrations
- Lower profile in the market means fewer tutorials, community answers, and third-party guides