Cognifity exists because production AI needs evidence, not guesswork. Verdict — our first product — is an Apache 2.0 toolkit for working with agent runs, supported model calls, evaluation, and monitoring with reproducible workflows.
Production AI teams have an evidence gap. Logs, latency charts, and token dashboards rarely show the full work an agent performed or which recorded event explains a failure.
Cognifity focuses on that gap: connect agent work and model calls, make failures inspectable, and help teams test and monitor changes.
Verdict is the first product: open-source observability for engineers running AI agents in real applications. It connects recorded agent work with supported model calls, evaluation, and monitoring.
Teams exploring LLM reliability, model migration, prompt changes, or evaluation workflows can get in touch to compare notes.
Cognifity is led by Rehan Shah, an engineering leader who has built enterprise software and production AI systems. His work now focuses on evidence that helps teams operate AI agents reliably.
He started Cognifity after seeing the same gap repeatedly: teams putting LLM-powered systems into production without a rigorous way to tell when quality, cost, or behavior had changed. Verdict is our first answer: open-source investigation and monitoring for the agent work and model calls teams actually run.
A finding should lead to the affected run, turn, event, or genuine model call. Missing source evidence stays visible.
Verdict is open source. You can run the SDK, evaluation package, inspection CLI, and local dashboard yourself, with your own storage and provider keys.
Local SQLite and dedicated PostgreSQL deployments let teams choose their own storage, access control, and provider keys.
Recorded agent issues, optional judge results, and prospective model-call alerts answer different questions. Verdict labels their evidence and coverage.
The dashboard starts with captured activity and recorded issues, then guides a reviewer toward investigation and monitoring setup.
SDK content capture is bounded and on by default; metadata-only mode is available. SQLite runs locally. Postgres is available when teams need shared storage. API keys stay in your environment.
Whether you're an enterprise AI buyer, a potential investor, a candidate, or another engineer who's felt these gaps — we'd like to hear from you.