Independent verification
for coding agents
Your agent writes the code.
Tiden verifies it’s what was actually asked for.It sits beside your agent and keeps the requirement as a durable record, versioned and merged like code. Your change is checked against that record, not against the session that produced it.
$ curl -fsSL https://tiden.ai/install.sh | bashCoding speed went up. Delivery failed to keep pace.
Agents generate features in minutes. Then someone, usually you, spends the rest of the day verifying that the thing actually works.
Your pipeline still runs.
It just can’t tell you the agent built the right thing.
Intent drift
What the agent built and what the ticket asked for slowly stop being the same thing, and no one notices until it matters.
Regressions hide
Working features quietly break because no one tested the part of the system the agent didn’t mention.
Agent velocity is wasted
Engineers spend their day undoing and re-prompting instead of designing what to build next.
What the code was for
When the suite goes green, all you have learned is that the agent agreed with itself. Tiden keeps the other half, versioned exactly like the code.
api/public/v1/limiter.gofunc (l *Limiter) Allow( ctx context.Context, key WorkspaceKey,) bool { q := l.tiers.For(key) return l.win.Count(key) < q}// TODO: rate limits
TIDEN-377 Billing tier caps volumeFree returns 429 above 60 requests per minute.Pro returns 429 above 600.Every 429 carries Retry-After in seconds.Unlimited for all tiers.branch · 1 conflict
How Tiden works
Every agent session becomes a closed loop: eight steps beside your agent, with no new process for your team. You set the thresholds and the gates, and a person resolves anything the loop can’t. The verdict comes from checks that ran in your environment, not from a model reading the diff.
What your agent reads
Tiden hands the agent a working contract before it writes a line: what is in scope, and the one place a mistake will not be caught.
$tiden intent start "Add rate limiting to the public API"Working contractTIDEN-412 · draftIn scope TIDEN-118 Public API authenticates by token2 tests TIDEN-204 Reporters retry on 5xx7 tests TIDEN-377 Billing tier caps volumeno testsCoverage gaps TIDEN-377 has nothing testing it. A change here ships unverified unless this session adds a test.
What’s in the product
Intent capture
Session hooks record what you asked the agent to build. Distilling a session produces an intent branch.
Requirements that branch like code
Propose changes on isolated branches, preview the diff, merge with conflict resolution. Main stays clean.
A living test repository
Suites, cases and step trees linked to the requirements they verify, structured so agents can keep them current.
Reporter ingest
One idempotent endpoint for any test framework, reconciled against stable external IDs. Re-runs never create duplicates.
Agents on the inside
Tiden’s own agents crawl your product and docs, then propose requirement changes on their own branches.
Error tracking, closed-loop
DSN ingest groups production exceptions into issues tied to releases, so incidents flow back into requirements.
Connects to the tools your team already uses
Ship agent code you can prove is right
What was asked, what was built, what was verified, what shipped. Shipping stops being a judgment call and becomes a matter of record.
or write to hello@tiden.ai