Observability for the systems you build with AI.Explore the platform
ObservaAI
Start free
OPEN TELEMETRY FOR AI

Know what your AI is doing.

Trace every model call. Follow every agent step. Understand token use, latency, and failures before your users find them.

Start monitoring Explore the platform
No credit card OpenTelemetry ingestion Private projects
ObservaAI PREVIEW
SAMPLE DATA
PROJECT / SUPPORT COPILOTOverview
Last 24 hours⌄
REQUESTS12,804↗ 18.4%
TOKENS2.48M1.86M in · 620k out
EST. COST$18.42Across 4 models
ERROR RATE0.42%Within target
TRAFFIC OVER TIMERequest volume
◉ Requests   ◌ Errors
12am4am8am12pm4pmNow
RECENT ACTIVITYLatest traces
View all →
support-copilotgpt-4o1.24s1,634 tokens
research-agentclaude-sonnet2.81s3,915 tokens
knowledge-raggemini-2.53.76s5,470 tokens
Traces arrivingOpenTelemetry · 24ms ago
OBSERVE THE WHOLE JOURNEY
Observe Evaluate Learn Agent workflows
Every stepTrace model calls and agent workflows
Every tokenTrack usage and estimated cost
Every signalFind latency, errors, and regressions
OBSERVE

What did my agent actually do?

Follow a request from your application through every tool call and model response. See the trace, the timing, and the token use in one place.

  • End-to-end trace timelines
  • Provider, model, and service breakdowns
  • Input/output tokens and cost estimates
Explore observability
Trace · 7f4d…b2911.82s
POST /api/assistantsupport-copilot · production1.82s
HTTP request
1.82s
vector.search
280ms
chat.completion
1.21s
MODELgpt-4oTOKENS1,248 in  /  386 outSTATUSCompleted
Connected spans reveal the full request path
SAMPLE SCORECARDProduction scorecardsupport-copilot · last 24 hours
HEALTHY
SUCCESS RATE99.6%+0.8% from yesterday
96score
Request successTarget ≥ 99% · observed 99.6%
Pass
Latency p95Target < 2s · observed 1.74s
Pass
Token budget2 requests exceeded 4k tokens
Review
Based on live trace signalsOpen scorecard
Rules you can explain
EVALUATE

Is my system getting better or worse?

Compare reliability and performance against clear targets. Turn live traces into scorecards for success rate, latency, and token budgets.

  • Error and latency checks
  • Token budget monitoring
  • Model-by-model comparisons
Explore evaluations
LEARN

Keep the traces that teach you something.

Save useful production examples into project-scoped trace sets. Review the outcome, model, token use, and timing without storing prompt or completion text.

Browse traces
Regression setEXAMPLE SET
✓
Support answer · successfulgpt-4o · 1,248 input tokens
Saved
!
Escalation · needs reviewclaude-sonnet · 3,204 input tokens
Review
✓
Knowledge lookup · successfulgemini-2.5 · 4,380 input tokens
Saved
Trace metadata only · No prompt text stored
AGENT WORKFLOWS

From the first span to the root cause.

Agents make systems harder to debug. ObservaAI brings each model call, tool, and error into a single trace so your team can investigate with context.

Follow the full chain

See parent/child spans across retrieval, tools, model calls, and your application.

request→retrieval→model→response

Debug with evidence

Inspect status, timing, provider, model, and safe operational attributes for each step.

!
Slow model callgpt-4o · 2,840ms · research-agent

See the cost of every turn

Connect token use and latency to the exact request and model that produced them.

Input tokens3,204Output tokens711Estimated cost$0.0248
OPEN BY DESIGN

Built on open standards.

Instrument with OpenTelemetry and keep your telemetry portable. Send OTLP/HTTP JSON to a project endpoint, then inspect the same trace across your app.

OpenTelemetry OTLP/HTTP JSON GenAI attributes
Create a project
Your AI appInstrument once
ObservaAILLM telemetry
Traces Usage analytics Your project
Your key routes data to the right project
YOUR DATA, YOUR CALL

Your data stays yours.

Projects belong to your account. Ingestion uses project keys, dashboards stay scoped, and prompt/completion content is removed before storage by default.

Private projects Key-based ingestion Trace metadata controls
Review your instrumentation attributes before sending production telemetry. Retention controls and team access are on the roadmap.
DATA
CONTROLBY PROJECT
OPEN STANDARDS. CLEAR DATA.

Build with the tools you already trust.

Use OpenTelemetry conventions and keep your telemetry understandable. Instrument once and avoid locking your traces into an opaque format.

OpenTelemetryCommon trace format and GenAI attributes
Portable projectsProject-scoped keys and ordinary JSON trace exports
OPEN
TELEMETRYIN · TRACE · UNDERSTAND
“
BUILT FOR THE WORK
When an AI request goes wrong, you need the trace—not another guess.

ObservaAI connects the model call to the application path, token use, and outcome.

Build with your traces
FAQ

Everything you need to get started.

Clear answers about instrumentation, data, and the current preview.

How do I get started?

Create a free account, make a project, and add an ingestion key. ObservaAI accepts OpenTelemetry trace exports at your project endpoint.

What can I monitor today?

Explore requests, model and provider, input and output tokens, estimated cost, latency, errors, services, and full span timelines.

Does ObservaAI store prompts or responses?

No. Prompt and completion content is removed before persistence by default. Operational span attributes and names may still contain sensitive values, so review what your instrumentation emits.

Which telemetry format is supported?

The current receiver supports OTLP over HTTP with JSON encoded Protobuf. OTLP binary Protobuf and gRPC are not available yet.

Can my team share projects?

Projects are currently private to the account that created them. Team invitations and shared workspaces are planned for a later release.

YOUR NEXT TRACE IS A START

Make your AI systems easier to understand.

Start with one project. Instrument your app. See what happens after every request.

Create your account