What Tekk is

Tekk is your agent for self-improving software. Point it at a codebase and it does three things continuously: it keeps a board of specs (the work), it runs autonomous loops that look for ways to make the software better, and it turns what it finds into proposals you either greenlight or decline.

The difference from a chat assistant is that Tekk doesn’t wait for a prompt. It watches your project — the code, the goals you set, signals from your connected tools — and comes to you with concrete, reviewed suggestions. You stay the decision-maker; Tekk does the noticing and the drafting.

The three things Tekk maintains

• Specs — the unit of work. A spec is a short, structured description of something to build or fix, with acceptance criteria. Tekk drafts them; you approve, edit, or reprioritize them.
• Goals — the direction you set. A goal tells Tekk what “better” means right now (ship onboarding, cut latency, raise activation). Loops orient their work around active goals.
• Proposals — Tekk’s own suggestions, awaiting your decision. Each one explains what it noticed and what it would do. Accepting a proposal turns it into a spec; declining it (with a reason) steers future suggestions.

How you work with it

You can drive Tekk two ways, and most people use both:

1. In the app (app.tekk.coach) — set goals, review proposals, steer the board, and watch the loops run.
2. From your coding agent — connect the Tekk MCP server to Claude Code, Codex, or Cursor and work the same board without leaving your terminal: list specs, implement them, review proposals, read the workspace’s context.

Where to go next

• Ready to wire Tekk into your agent? Start with the MCP server — the most common entry point.
• New to the vocabulary? Read Core concepts.
• Want to understand the autonomous side? Read How the loops work.
• Want Tekk to see your other tools (like Sentry)? See Integrations.