Observability for AI agents,
inside your own walls.
Conduit traces every agent step, tool call and LLM call (status, errors, tokens and cost) and keeps the data in your infrastructure. It's built in Go, with no npm, no JVM and no supply-chain surprises.
Built for teams that take security seriously
Everything you need to see what your agents are doing, and nothing you'd have to explain to a security review.
Identity you can trust
Agents authenticate with mTLS client certificates. The certificate's CommonName is the agent's identity, and attributes can't spoof it.
Hardened ingestion
Every span is validated before it reaches storage. Rate limits are charged per telemetry item, so batching can't get around them.
LLM-aware traces
Model, tokens, cost and errors per span, rolled up per trace. Turn off prompt storage with one switch.
A UI that's quick to read
One row per trace, filters, stable paging, and a span tree with a waterfall timeline. No JavaScript framework.
Small and auditable
Static Go binaries in FROM scratch images, a strict Content-Security-Policy, and few dependencies.
Production basics included
Login or GCP IAP sign-in, health probes, configurable retention, Kubernetes manifests, and an end-to-end CI smoke test.
Documentation
Start with the architecture, or go straight to deploying and instrumenting your first agent.