A coding agent that runs in your terminal. Rendering is @trendr/core, model access is @prsm/ai.

npm i -g picocode

Requires node 24 or newer and ripgrep. Set at least one provider key in your environment: GEMINI_API_KEY, ANTHROPIC_API_KEY, OPENAI_API_KEY, or XAI_API_KEY. Alternatively, use /connect in pico to sign in with your ChatGPT subscription and access Codex models. Models are offered based on the credentials available.

Run pico in a project directory. Type / to browse commands, or run /help for the full reference.

Sessions are stored under ~/.pico as jsonl event logs. AGENTS.md files are read automatically from the working directory up to the git root, and lazily from subdirectories as the agent touches them. User-defined prompt templates live in ~/.pico/commands and .pico/commands, skills in ~/.pico/skills and .pico/skills.

• Multi-provider support
• ChatGPT/Codex subscription support through Codex OAuth
• Headless mode with text, JSON, and JSONL output
• Background shells, with a management UI
• Parallel background agents, with a management UI
• Full MCP support
• Project and global memories with on-demand recall and management
• Fast session, project, model, prompt, and file pickers
• Scheduled wake-ups and recurring agent tasks
• Skills and commands
• Image attachments (drag and drop or via file picker)
• Per-project config for memories, commands, skills, tools, and MCP servers

And quite a bit more, actually.

Pico can extend itself. Describe what you want and whether it should be available globally or only in the current project.

• Ask Pico to create a skill for you
• Ask Pico to create a command for you
• Ask Pico to create a tool for itself