# docrew -- Full Product Reference

> AI Crew that works, not chats

docrew is a personal AI agent application for desktop and mobile. It runs locally on your machine, processes your files without sending them to the cloud, and connects to 250+ third-party services. Built for people who need AI to do real work -- not just chat.

## Product Overview

docrew differs from cloud-based AI assistants in three fundamental ways:

1. **Local-first**: All file processing happens on your device. Documents, spreadsheets, and images are read and written locally. Only the conversation text is sent to LLM providers.
2. **Fully autonomous**: The agent executes multi-step workflows without asking for permission at every step. It reads files, writes files, runs shell commands, and calls APIs on its own.
3. **Connected**: Via Composio, docrew connects to 250+ services (Gmail, Google Drive, Slack, HubSpot, Notion, Jira, GitHub, etc.) and can act on your behalf.

## Platforms

- **Desktop**: macOS, Windows, Linux. Built with Tauri (Rust backend + React frontend). The agent runtime is a custom Rust implementation compiled into the Tauri binary.
- **Mobile**: iOS and Android. Built with React Native and Expo. Thin client that can send tasks to the desktop agent or run standalone via the cloud proxy.

## Architecture

- **Agent runtime**: Custom Rust, running inside the Tauri desktop shell. Tool-use loop with SSE streaming.
- **LLM providers**: Google Gemini (via Vertex AI) and Anthropic Claude (via AWS Bedrock). Per-conversation model family selection.
- **Proxy**: Fly.io-hosted TypeScript service. Handles authentication, smart routing (auto-selects model tier), billing, and message persistence.
- **Backend**: Supabase (PostgreSQL + Row-Level Security + Edge Functions). Stores accounts, subscriptions, conversations, and messages. Never stores user files.
- **Sandboxing**: All shell and Python execution runs inside OS-level sandboxes (macOS Seatbelt, Linux bubblewrap). Network disabled by default.

## Agent Tools

The agent has access to the following tools:

- `file_read` -- Read files within the workspace
- `file_write` -- Write or create files
- `file_list` -- List directory contents
- `file_grep` -- Search file contents (ripgrep-based)
- `shell` -- Execute shell commands in a sandbox
- `python` -- Execute Python scripts in a sandbox
- `docx_read` -- Extract semantic content from .docx files (custom OOXML parser)
- `xlsx_read` -- Extract content from .xlsx spreadsheets (custom OOXML parser)
- `http_client` -- Make HTTP requests
- `subagent` -- Spawn sub-agents for parallel work
- `connector` -- Interact with 250+ third-party services via Composio
- `show_file` -- Display a file card in the chat
- `send_file` -- Send a file to mobile via Supabase Storage

## Pricing

All plans include AI compute costs (no separate API key needed).

| Plan | Price | Credits | Best for |
|------|-------|---------|----------|
| Free trial | $0 | 1M credits | Trying docrew |
| Starter | $10/month | 7M credits | Light personal use |
| Plus | $25/month | 17M credits | Regular personal use |
| Pro | $50/month | 35M credits | Power users |
| Business | $100/month | 70M credits | Teams and heavy use |

Top-up packages available at the same rate (e.g., $10 for 7M credits, $25 for 17M).

1 credit = 1 microdollar ($0.000001) of LLM compute cost.

## Model Selection

Users choose the model family per conversation:

- **Gemini** (Google): Flash Lite (light tasks), Flash (medium), Pro (heavy). Includes image generation.
- **Claude** (Anthropic): Haiku (light), Sonnet (medium), Opus (heavy). Best for complex reasoning.

Smart routing automatically selects the right model tier based on task complexity.

## Privacy and Security

- All user files stay on the local device. Never synced to the cloud.
- API keys stored in the system keychain (desktop) or secure storage (mobile).
- Row-Level Security (RLS) on every Supabase table.
- All shell and Python execution runs in OS-level sandboxes.
- GDPR-compliant: per-user regional routing (US or EU) for LLM calls.

## Comparison Pages

- docrew vs ChatGPT: https://docrew.ai/compare/chatgpt
- docrew vs Claude (Cowork): https://docrew.ai/compare/claude-cowork
- docrew vs Grok: https://docrew.ai/compare/grok
- docrew vs Gemini: https://docrew.ai/compare/gemini
- docrew vs OpenClaw: https://docrew.ai/compare/openclaw

## Recent Articles

- When AI as Chat Is Not Enough: The Case for Desktop AI Agents
  URL: https://docrew.ai/blog/when-ai-chat-is-not-enough
  Published: 2026-03-08
  Summary: Chat interfaces changed how we interact with AI, but chatting about work and doing work are different. Desktop AI agents are the next step.

## Links

- Homepage: https://docrew.ai
- Download: https://docrew.ai/download
- Pricing: https://docrew.ai/pricing
- Blog: https://docrew.ai/blog
- Privacy Policy: https://docrew.ai/privacy
- Terms of Service: https://docrew.ai/terms

## Contact

- Company: DDDEV LTD
- Domain: docrew.ai