How Tekk works

Set it up once. Then it keeps going.

The first three steps you do once. The last three repeat every day: the loops propose, you decide, your coding agent builds.

1. 1

   Connect your repository

   Sign in with GitHub or GitLab and pick the repo. Tekk reads your code; it never writes to it.

   Getting started →
2. 2

   Plug in the tools you already use

   Connect Sentry, Datadog, PostHog, Stripe and the rest. An error spike, a monitor that fired or a failed payment wakes the loop that owns it. Every connection is read-only by default, and none is required.

   Integrations →
3. 3

   Pick your loops

   Switch on the loops you want: Security, Payments, Reliability, AI engineering and the rest of the twelve. Each one wakes when your code changes, when Sentry fires, or when its area has gone unexamined too long.

   The twelve loops →
4. 4

   Loops run 24/7. Proposals land on your board.

   Overnight, a loop reads what changed, works one method end to end, and brings you at most one proposal: what to change, which files, and why. It lands in the Proposals column of your board.

   How a loop runs →
5. 5

   In the morning, /sweep

   Run /sweep in your coding agent. It re-checks every proposal against the code as it is now, drops what is already fixed, and waits for your go. The ones you greenlight become specs in To Do. Decline one with a reason and the loops learn what is noise for your project.

   The Tekk MCP server →
6. 6

   /blitz, and the loop closes

   Run /blitz and your coding agent (Claude Code, Codex or Cursor) opens one session per spec, in parallel. Each pull request says Closes TEK-123; when it merges, the spec moves to Done, and that merge is the next change the loops read.

   How a merge closes a spec →

Tekk never writes, ships or merges code. The loops find and propose; you decide; your own coding agent builds.