Data & Privacy
TatsuCode is designed for local-first use with user-controlled provider access.
Privacy Principles
- No telemetry from the desktop app — TatsuCode itself sends nothing to Studio Tatsu about your usage. (The marketing website at tatsu.ai uses Google Analytics, disclosed in its privacy policy. The two surfaces are separate; the app does not piggyback on the website.)
- No TatsuCode server in the network path — there is no Tatsu Code server. Your prompts travel from your machine straight to the provider you've configured (Anthropic, OpenAI, GitHub, OpenRouter, your local Ollama, etc.).
- Data is sent only to providers you explicitly configure. Each configured provider sees what you send to it, the same as if you used that provider's own client.
- Local project / session state stays on your device. Sessions, projects, skills, and history live in
%UserProfile%\.tatsu\. Credentials are encrypted at rest incredentials.enc.
What Is Stored Locally
TatsuCode stores only what is needed for operation and continuity, all under %UserProfile%\.tatsu\:
- App settings —
config.json - Provider connection state — which providers you've configured and which model is active
- Credential material —
credentials.enc(encrypted at rest, ACL'd to your Windows account; an oldercredentials.jsonfrom earlier versions is auto-migrated to the encrypted form on first launch) - Session and project history — kept locally as long as you keep it; clear with
/clearor/remove-tatsu - Skills, plugins, and runtime caches
This page intentionally avoids deeper storage-format internals to reduce unnecessary exposure.
What May Be Sent to AI Providers
When you use a hosted provider (Claude Pro / Max via the claude CLI, ChatGPT / Copilot via OAuth, OpenRouter, or any custom external API), the provider receives the request content required to answer your prompt — same as if you used that provider's own client. This is unavoidable for any tool that talks to a hosted model.
What's transmitted:
- Your prompts
- File content you reference (for example via
@file.ts) - Images, screenshots, or attachments you include
- Whatever conversation history TatsuCode forwards to keep the model in context for the current turn
What's not transmitted:
- Other files in your project that weren't referenced or read
- TatsuCode telemetry or metadata about you (TatsuCode adds none)
- Anything sent through a TatsuCode server (there isn't one)
If you want zero data leaving your machine, configure a local-only provider (Ollama, LM Studio, llama.cpp, or any OpenAI-compatible local endpoint) and avoid hosted ones.
For provider-specific data handling, follow each provider's own privacy policy. TatsuCode does not modify, weaken, or extend any provider's terms.
Credentials Security
- Credentials are stored locally in
%UserProfile%\.tatsu\credentials.enc, encrypted at rest and ACL'd to your Windows account. - For Claude Code specifically, your Anthropic OAuth tokens live in
~/.claude/where theclaudeCLI itself stored them — TatsuCode never sees those tokens. - Keep your machine account secured.
- Never share credential files.
- Rotate API keys periodically.
- Disconnect providers you no longer use.
Session and History Management
Use built-in commands to control retained data:
/session
/session-rename
/new
/clear
/remove-tatsu
Recommended:
- Delete old sessions you no longer need
- Use named sessions only when needed
- Start fresh sessions for unrelated tasks
DevBrowser Privacy Model
The built-in DevBrowser is isolated from your regular browser profile.
- Separate browsing context
- No access to your main browser extensions/profile
- Designed for development workflows, not personal browsing
Before Sharing Screens or Recordings
- Verify no sensitive code or secrets are visible
- Check provider/settings screens before demoing
- Sanitize logs or screenshots used in public channels
If You Need Full Data Cleanup
Use:
/remove-tatsu
Then verify local app data is removed according to your organization’s policy.
Standards and Governance
For instruction governance in projects:
- AGENTS.md reference: agents.md
- Skills reference: agentskills.io
These help teams define consistent behavior without exposing internal implementation details.
Next Steps
- Settings — user-facing configuration
- Providers — provider and access setup
- Troubleshooting — common issue recovery