Harvey User Manual

Harvey is a terminal-based coding agent backed by a local Ollama server. It provides an interactive REPL where you chat with a large language model while also being able to read files, run commands, search code, and apply suggested changes — all sandboxed to a single workspace directory.

Note: Harvey focuses on local models via Ollama or Ollama servers running on the local network. It has limited support for remote endpoints (Anthropic, DeepSeek, Gemini, Mistral, OpenAI).

Contents

1. Installation
2. Getting started
3. The HARVEY.md system prompt
4. Session walkthrough
5. Keyboard shortcuts
6. Slash commands
   • Session
   • Backends
   • File operations
   • Code assistance
   • Session quality
   • Knowledge base
   • Recording
7. Security — safe mode, permissions, audit log, API key filtering
8. Typical workflows
9. Further reading
10. Models I’m using
11. Model evaluations
12. Using RAGS with Harvey

Further reading

• DOCUMENTATION.md — master index of all Harvey documentation
• CONFIGURATION.md — all harvey.yaml keys including security settings (safe mode, permissions, timeouts).
• ARCHITECTURE.md — detailed technical documentation: component map, core types, conversation model, backend implementations, security system, test coverage, and the full feature roadmap.
• harvey.1.md — man page source (generated from harvey -help).
• GitHub Issues — bug reports and feature requests.