Know which agents are getting better — and which are just getting expensive.

See the ranked brief and the evidence behind each decision. Why Morse ↓

1. 1

   Instrument

   Add the SDK — Python or TypeScript — or point an OpenTelemetry exporter at Morse. Typed agent, LLM and tool spans start flowing (HTTP and database spans with SDK 0.7.0); telemetry batches asynchronously and never blocks the request.

   AnthropicVercel AI SDKOpenAI AgentsLangGraphLangChainor plain OTLP
   PythonTypeScript

   pip install morse-ai

   import morse_ai
   
   # One line — reads MORSE_API_KEY from your environment
   morse_ai.init()
   
   # Cost, quality, and traces appear as your agent runs

   Invites arrive by email during private beta. Generate your API key during onboarding. No credit card, now or at launch.

2. 2

   Map success

   Tell Morse what a successful run is for each agent. Outcomes can be labeled from your code, confirmed in the UI, or both. This is the step the brief is built on.

   From your codeConfirmed in the UI

   booking-agent

   Booking confirmed

   fare-reconciliation-agent

   Fares reconciled

   support-triage

   Ticket in the right queue

   with morse_ai.run("support-triage"):
       ticket = route(request)
       morse_ai.set_outcome(
           success=ticket.routed_ok,
           outcome="routed",
       )

   TypeScript: morse.setOutcome(). Unlabeled runs stay unknown — never counted as failures. Until enough runs are labeled, the brief shows coverage instead of a success rate.

3. 3

   Read your first brief

   Start with the ranked list. While labeled telemetry builds a baseline, verdicts stay neutral or withheld. Open a trace to check the evidence; compare weekly briefs once there is enough history to judge whether a change helped.

   Demo org · illustrative data

   1. fare-reconciliation-agent45.2% · -8.1 ptsNeeds attention
   2. triage-agent78.8% · -11.8 ptsNeeds attention
   3. seat-booking-agent96.5% · +7.4 ptsNeeds attention