Pilot is a local, terminal-based AI agent for your computer. Tell it what you want done in plain English — organize a folder, run your tests, find a file from last week — and Pilot inspects your machine, proposes a plan, asks for approval on anything risky, and executes it. Entirely on your machine.

npm install -g pilot-agent
pilot

Screenshots

| Command palette (Ctrl+K) | Providers | | --- | --- | | | |

| Permissions & risk model | Diagnostics (pilot doctor) | | --- | --- | | | |

| Credentials (BYOK, never displayed) | Non-interactive CLI | | --- | --- | | | |

Why Pilot

• Local-first. The agent loop, tool execution, memory, and config all run on your machine. Use Ollama for a fully local model, or bring your own key for OpenAI, Anthropic, OpenRouter, or Gemini.
• Your keys stay yours. API keys are encrypted at rest on disk and are never sent to a Pilot-owned server (there isn't one), logged, printed, or placed into model prompts. See docs/security.md.
• Nothing risky happens without you. Every filesystem write, delete, move, or shell command is risk-classified and shown to you before it runs. You can approve once, approve for the rest of the session, or deny.
• Everything is logged. A local, human-readable audit trail (pilot audit) and task history (pilot history) record exactly what Pilot did.

Install

npm install -g pilot-agent
pilot init

Quick start

# Fully local, no API key needed
ollama serve
pilot config provider ollama
pilot

# Or bring your own key
pilot config provider anthropic
pilot config keys set anthropic
pilot "find all screenshots related to my current project"

CLI reference

pilot                          # open the interactive TUI
pilot "organize my downloads"  # run a task directly
pilot run "run tests in my project"

pilot init                     # create ~/.pilot/config.json
pilot status                   # show current provider/model
pilot doctor                   # verify provider connectivity

pilot config provider [name]   # get/set provider
pilot config model [name]      # get/set model
pilot config keys list
pilot config keys set <provider>
pilot config keys delete <provider>

pilot history                  # recent tasks
pilot audit                    # local action audit log

Supported providers

| Provider | Requires API key | Notes | |------------|:-----------------:|---------------------------------| | Ollama | No | Fully local, default provider | | OpenAI | Yes | | | Anthropic | Yes | | | OpenRouter | Yes | Access to many hosted models | | Gemini | Yes | |

Tools

Pilot's agent loop can call:

• Filesystem — fs_list, fs_read, fs_write, fs_move, fs_delete, sandboxed to the workspace root (see docs/security.md).

• Terminal — terminal_exec, with dangerous commands classified dangerous and provider-credential env vars stripped from the child process.

• Process — process_list, process_kill.

• Browser — browser_open, browser_click, browser_type, browser_extract_text, browser_screenshot, browser_close, backed by headless Chromium via Playwright. One browser session is reused across a task and closed automatically when the task ends. Chromium binaries aren't downloaded by npm install — run this once:

  npx playwright install chromium

Every command panel from the TUI's command palette (Ctrl+K) — Providers, Credentials, Permissions, Settings, Diagnostics, History, Memory — has a dedicated screen; there's also a CLI equivalent for each (pilot config …, pilot doctor, pilot history).

Architecture

See docs/architecture.md for how the agent loop, tool registry, and TUI fit together, and docs/providers.md for adding a new provider.

Development

git clone https://github.com/pilot-agent/pilot
cd pilot
npm install
npm run dev -- "list files in the current directory"
npm test

License

MIT — see LICENSE.