Get started
Install the GitHub App, dispatch your first lane, sign off on the diff. Fifteen minutes.
Welcome
One workspace. Many agents. Cryptographic audit. Zero stepping on each other.
A startup we work with in São Paulo had two engineers running the same Claude Code session against the same payments service for forty minutes before either of them noticed. Three commits clobbered. Two more contradicting each other. An on-call page that turned out to be the agent rewriting a test it had no business touching. They didn’t have a model-quality problem. They had a coordination problem.
SprintLoop is the operating system you reach for when more than one agent is going to touch the same repo. It enforces lanes (cryptographic scope fences), races multiple harnesses against the same brief, runs every diff past a Review Committee before merge, and writes a signed Ed25519 ledger of everything so a SOC 2 auditor can trace any line of code back to the prompt that produced it.
If you’ve already evaluated GitHub Copilot Workspaces, Devin, or a stack of standalone CLIs and concluded “I want all of this, plus actual concurrency control,” you’re in the right place.
Concepts
Lanes, harness racing, the Context Engine, the Review Committee. Read these in order before integrating.
API reference
165 edge functions covering dispatch, audit, billing, federation, and the Context Engine. REST + bearer.
Security
What’s actually here
The product is one surface — a TeamDock-style workspace where messaging, calling, lanes, agents, repos, and audit live one keystroke apart. The docs cover four things you’ll need to know before integrating:
- Concepts. Five short reads. They’re in narrative order: Lanes (what a lane is), Harness racing (how we pick a winner), the AI SDLC and Context Engine (how the workspace knows what’s good), the Review Committee (how a diff earns its merge).
- Integrations. GitHub, IDE extensions, MCP, notifications. Most teams need three of the four.
- API reference. Auto-generated from the live edge functions; never out of date by more than a build.
- Security. What goes on the wire, what gets signed, what your auditor will ask about.
A note on what we don’t do
We don’t store your source code. The workspace orchestrates lanes against your git remote — GitHub, GitLab, or Bitbucket — and never mirrors the repo content into our database. Provenance entries reference commit hashes; they don’t snapshot the tree. If your auditor asks where your IP lives, the answer is: still in your git host.
We also don’t proxy your model traffic. Bring your own keys for Anthropic, OpenAI, Google, Mistral, or whichever harness you’re racing. SprintLoop signs the prompt, the response, and the tool calls — not the model’s weights and not the inference billing.
Where to go next
If you have a repo and twenty minutes, the Get started page is the fastest path to a green CI signal from a SprintLoop-driven lane.
If you’re evaluating, skim Lanes and Harness racing — that’s the part of the architecture that decides whether SprintLoop fits how your team wants to work.
If you’re an integrator wiring SprintLoop into another tool, the API reference and MCP page are the two you’ll live in.