Detect & fix agent issues in production.

Kensa finds issues in production traces, gives your coding agent the evidence to fix them, and monitors for regressions after deployment.

Agents fail quietly. Then fail again.

You cannot trust that the next prompt, model or tool change you make won't break existing workflows or introduce new bugs. You uncover them when they've already had a visible impact, or worse: through your customers. And the evidence you need to fix them is buried in traces nobody reads.

How it works

Kensa surfaces agent failures, hands your coding agent the evidence to fix them, and monitors in production so they don’t come back.

  1. 01

    Connect traces

    Bring your own traces via an OpenTelemetry exporter or Langfuse, or simply use Kensa to instrument your agent. Traces are redacted before processing.

    Your agentOpenTelemetryKensa SDKLangfuseKENSA
  2. 02

    Detect issues

    Kensa flags potential issues supported by evidence from your redacted traces. It builds a catalog of issues your agents have exhibited in the past.

    Traces2 issuesRefund without identity check
  3. 03

    Fix the code

    Your coding agent reads the issue over MCP and works on the fix in your repository. You control review and deployment.

    agent.ts41424344454647⏺kensa_mcp.get_issue("#42")⏺read(agent.ts)·✢✳✶✻✽fixing identity check…⏺update(agent.ts)- refund(order)+ verify(customer)+ refund(order)›
  4. 04

    Monitor for regressions

    Report the deployed fix. Kensa watches later production traces and shows whether the same issue recurs.

    TracesFix deployed10 traces20 traces30 traces40 traces50 traces60 traces70 traces80 traces90 traces100 traces110 traces120 traces130 traces140 traces150 traces160 traces170 traces180 traces190 traces200 tracesNo recurrence

What it is (not)

Kensa is for builders and engineers who want to catch flaky agent behavior in prod before it’s too late.

What it is

  • Evidence from real runs

    Every issue is grounded and tracks back to a production trace.

  • Context for your coding agent

    Your coding agent reads the issue and its evidence over MCP to investigate and fix it.

  • Monitoring after a fix

    After a fix is deployed, see whether the issue recurs in production.

What it isn’t

  • Not another dashboard

    You can look at the charts, but we'd rather your agent close the loop without you having to leave your terminal.

  • Not an auto-merger

    Your coding agent proposes the change. You control review and deployment.

  • No benchmaxxing

    Benchmarks often don't translate to production. We care about reliability in the real world.

FAQ

If you already have tracing enabled, connect Langfuse or your existing OpenTelemetry exporter. Or simply use the Kensa Python or TypeScript SDKs to instrument and start tracing today. Optionally, connect your GitHub repository for better context on your agent’s job and what good looks like.

Start with a trace you already have.

Fix your agent issues today.

Get started for free