Fleet Health in Seconds, Not in Tabs

What this means for you

KAHN is observability for agent fleets. If you run ClaudeCode agents across several repos and you cannot answer “which one needs my attention right now” without grepping logs, KAHN is the dashboard the operator reaches for first.

The pitch

A prompt-DAG scheduler with a read-only inspection surface called Scope. Agents transition from thinking to tool use to converged or failed, and the operator sees the whole fleet as a single sortable list with the regression count weighted heaviest. Apache 2.0. No telemetry. Self-host or run the hosted dashboard.

Who it’s for

The solo operator or platform engineer running more than one ClaudeCode agent at the same time and tired of switching terminal tabs to find out which run died.

Proof points

• Two public surfaces shipped: Scope at kahn.tools and the CLI at kahn.run, each with install and self-host docs
• Reference dashboard sorts by a four-layer severity score: a fresh regression weighs 1000, while failed and partial outcomes are each capped at 200 so a noisy agent cannot bury real signal
• Restraint references named on the record: Linear, Stripe Dashboard, Vercel. Anti-references named on the record: Datadog and New Relic dark plus neon
• Apache 2.0 open core, no phone-home, your agents on your filesystem

mindmap
  root((KAHN))
    Scan
      walk repos
      build DAG
    Schedule
      topological order
      conflict detection
    Verify
      exit codes
      retry loop
    Inspect
      Scope dashboard
      severity sort
      fleet at a glance