Repository navigation
Product Overview
Tepegoz is an agentic browser: a real desktop browser with an AI agent that can inspect pages, form a plan, use browser tools, and pause before risky actions.
The product is built around one idea: autonomy should be observable, bounded, and reversible. The agent can help operate the browser, but permission, policy, and secrets are enforced outside the model.
The desktop app is the product. The current build does not require a Tepegoz account or managed backend. Users bring their own AI provider keys, and provider calls go directly from the user's machine to the selected provider.
The renderer and visited web pages are untrusted. Important decisions happen in the main process: policy classification, dangerous-action approval, secret storage, and network boundaries.
The agent shows its plan, tool calls, decisions, token usage, and approval requests. State-changing, destructive, financial, or tainted actions are gated by human-in-the-loop approval.
The project intentionally distinguishes:
- built and exercised,
- landed but not measured,
- deliberately inert,
- planned but not built.
This matters because an agentic browser can affect logged-in accounts, credentials, local files, and money. A vague "AI does it for you" claim is not enough.
- Desktop browser shell.
- Agent console.
- Settings and provider-key management.
- Downloads, uploads, history, bookmarks, tasks, and macros.
- Network privacy routing.
- First-party extension system.
- Developer-facing packages and architecture decision records.
README.mddocs/website/home.mddocs/website/features.mddocs/architecture.md