Concourse
concourse.run Join founding access
Private beta — by invitation

Don’t trust the agent.
Make it prove the work.

Concourse runs coding agents inside gates you hold. Every run arrives as a proposal — the diff, the passing tests, the evidence behind each acceptance criterion — and nothing merges until it’s earned the right to. Trust becomes a dial you turn, not a default you accept.

No lottery, no spam — we’ll reach out as seats open. We store your email to do that (privacy).
concourse.run/acme · sample workspace
Mission Control
What needs you, what’s moving, what shipped
Sample
Needs your decision
Moving
Nothing merges unreviewed
Every run lands as a proposal. The merge waits behind a gate you hold — by default, you.
Every criterion, backed
Each acceptance criterion links to the diff and the test that proves it. Claims are receipts.
Deterministic gates hard-stop
Tests, coverage, CI and secret-scan are facts, not opinions — they can stop a merge cold.
Earned autonomy

Autonomy is earned, never granted.

Every agent starts fully gated — you confirm what “done” means, approve the plan, review every merge. As the work proves itself, you turn gates off one at a time. The deterministic checks never turn off at all. This is the road we’re building: today, every plan is human-approved.

Right now, this agent is at
Plan trusted
Plans auto-approve when they add no new risk. You still confirm “done” and review every merge.
Plan approval
auto
no new risk
Merge review
you review each
human gate
Fact gates · tests · CI
always hard-stop
never turns off
Drag the dial — gated to earnedit only moves as far as the work has proven
The floor never moves. Tests, coverage, CI and the secret-scan are deterministic facts — they hard-stop every run, at every setting. Autonomy is earned on judgment, never on facts.
See for yourself

Read the receipt before
you read the agent.

When a run is ready to merge, Concourse shows you the proof, not a promise: the satisfied acceptance gate, every changed file, the tests that back each criterion. Approve in one click — or open the diff and look closer.

The acceptance gate, satisfied — coverage, behaviour and CI shown as evidence, not a checkmark you take on faith.
Every changed file and the test that exercises it, one click from the proposal.
Approve & merge, or open the diff — the agent never merges itself at this trust level.
Rate-limit the public API forge/sample-rate-limit · sample run
Needs you
Shape Plan Execute Verify Ship Learn
Acceptance gate fact
coverage ≥ target · 429 + Retry-After on all limited paths
Test suite green · a few added sample
Coverage above target on changed lines ≥ target
429 + Retry-After on every limited path
src/gateway/rateLimit.ts +84 −1
request → 200, then 429 at the limit verified
Evidence bundle · criteria backed open
CI is green. Nothing merges unreviewed.
The safety primitive

A gate knows the difference between
a fact and a judgment.

Most tools collapse “the step ran” and “the result is acceptable” into one green check. Concourse keeps them apart. Deterministic facts can hard-stop a run; a model’s judgment can inform and loop it — but never gets the last word on a merge.

Fact gates Can hard-stop

Tests, coverage, CI, secret-scan. Deterministic and binary — they either pass or they block the merge. No model in the loop, no negotiation.

Test gate fact
coverage ≥ target
Test suite green all green
Coverage above target ≥ target
No secrets in diff
Judgment gates Informs & loops

A reviewer model scores the work and explains its reasoning. A low score sends the run back to revise — it never silently approves a merge on its own.

Review gate judgment
reviewer score ≥ threshold
Error paths covered clear
Naming drifts from the module below bar
Loops back to Forge to revise
One run, end to end

Gated. Inspectable. Resumable.

Every run moves through the same phases — and stops at any one of them the moment it needs you. Pause it, inspect the evidence, send it back a step, resume where it left off. Nothing is a black box.

Shape Plan Execute Verify Ship Learn
Resumable runs
Pause at any gate, inspect, send a step back, resume. Long runs survive interruption.
Evidence bundles
Each proposal carries its proof — criteria mapped to the diffs and tests that satisfy them.
Trace viewer
Every decision the agent made, every gate it crossed, recorded and replayable.
Per-step autonomy
Trust is set per step, not per agent. Grant the safe ones, hold the rest.
Risk-aware routing
Blast radius is scored up front. High-impact changes always route back to you.
Skills, not prompts
Reusable, versioned capabilities — so an agent's good run becomes every run.

Make the agent prove it.
From the first run.

Founding access is a hand-picked private beta — small batches, shaped by the engineers in it. Bring one repo and one skeptical eye.

No lottery, no spam — we’ll reach out as seats open. We store your email to do that (privacy).