FAULTLINE.Tell us what should happen
A wrong tool call, an unfinished task or a missed voice reply. Or a change you need checked before release.
One task. Clear results. Tests you keep.
We run agreed checks
Record the actions and outputs. Compare a proposed fix using the same inputs and normal-use checks.
Your team gets the evidence
Findings, runnable checks and a regression test if the failure reproduces. If it doesn't, we document that too.
€1,500 before tax · Up to 15 engineering hours over five working days once access is ready. Scope and model-call budget agreed first. Fix implementation is separate; reproduction and a successful fix aren't guaranteed.
Public example / voice-agent pipeline
A reconnect sent the same result twice.
We reproduced a reported Pipecat bug, prepared a patch and posted the tests upstream. The original discovery belongs to the reporter.
Read the patch, tests and limits4 failing → 0 failing
Six selected checks. Two passed before; all six passed after the proposed patch.
Local tests with mocked provider connections, not live calls. Not an upstream release, customer testimonial or reliability estimate for other agents.
Talk to Igor
What should your agent do?
Tell me your stack, the task and what needs checking. A few lines are enough to discuss scope. No credentials, private logs or customer data.
Use your mail app or copy the address. Nothing is sent until you press Send.
Two things to know
What access do you need?
We agree on a synthetic example, a minimal test environment or a run by your engineer. Production access isn't needed to start.
What isn't included?
Security audits, certification, training and production changes. Passing checks don't guarantee whole-system reliability. FAULTLINE is independent; framework examples aren't endorsements.