# OBSERVABILITY

## See every step. Trust every release.

Trace every agent decision, evaluate quality at every stage, and monitor production behavior - with the visibility enterprise deployments require.

## Trace Everything

### Trace every decision, not just the final output.

See immutable traces of prompts, agents, tool calls, branching paths, retries, and model responses across the full execution path.

### Performance Metrics

Avg latency, cost per run, tool call time, steps taken, context utilization, prompt and completion token counts - all at a glance.

### Deep run inspection

Click any run to see the full conversation trace - inputs, outputs, tool calls, timestamps, and which version ran it.

### Filter and export

Filter by status, user, or date range. Download your entire run history for offline analysis, compliance audits, or reporting.

### Run History - Support Agent Pro Last 7 days

- **1.2s** AVG LATENCY  
- **$0.004** AVG COST  
- **4.1** STEPS/RUN  
- **68%** CONTEXT USE

1. How do I reset my password? **Completed 0.9s**  
2. Refund request - order #84221 **Completed 1.4s**  
3. Account upgrade help (escalated) **Escalated 2.1s**  
4. How many licenses do I have left? **Completed 0.7s**  
5. Billing statement for March 2026 **Completed 1.1s**

## From trace to eval. Without leaving the platform.

Move from a problematic run directly into testing, scoring, and analysis instead of stitching together separate tools.

### Pre-deployment testing

Run test suites, simulations, and red-team scenarios before release to validate behavior under expected and adversarial conditions.

### Post-deployment evaluation

Score live outputs, detect regressions, and monitor changes in quality as prompts, models, and workflows evolve.

### Faster quality loops

Turn production traces into structured evaluation workflows so teams can improve prompts, agents, and orchestration with evidence.

## Don't just monitor production. Understand it.

Track how quality, reliability, latency, throughput, and spend change across runs, workflows, and releases.

### Metrics that matter

Monitor throughput, latency, token usage, model performance, completion quality, and workflow reliability from one operational view.

### Cost visibility

Attribute spend across agents, workflows, model calls, and execution paths so optimization is tied to actual system behavior.

### Insights and anomalies

Detect unusual execution patterns, rising failure rates, quality regressions, and cost spikes before they become systemic issues.

## Test before release. Catch regressions after.

Validate expected behavior before launch and detect quality drops after changes land.

### Structured test suites

Evaluate workflows against fixed datasets and expected outcomes before promoting changes to production.

### Scenario simulation

Exercise edge cases, failure paths, unsafe prompts, and tool dependency issues under controlled conditions.

### Regression detection

Catch quality drops introduced by model changes, prompt edits, workflow updates, or orchestration logic shifts.

## Auditability built for enterprise operations.

Maintain immutable records of runs, changes, and execution history with the visibility needed for governance and review.

### Immutable Audit Log

Retention: 90 days Tamper-proof

| TIMESTAMP           | ACTOR                  | ACTION             | OUTCOME | TAGS         |
|---------------------|-----------------------|--------------------|---------|--------------|
| 2026-03-31 09:14:02 | agent/summarizer-v2   | workflow.executed   | success | prodv2.4    |
| 2026-03-31 09:14:03 | system                | eval.scored        | pass    | faithfulness: 0.93 |
| 2026-03-31 09:14:11 | admin/wei.chen       | prompt.updated       | committed | agent: classifier |
| 2026-03-31 09:15:44 | agent/classifier-v1   | workflow.executed   | flagged  | review      |
| 2026-03-31 09:16:02 | admin/priya.r         | policy.applied      | enforced | PII-block   |

### Immutable audit logs

Maintain durable records of execution history, configuration changes, release events, and operational activity.

### Policy-aware visibility

Track behavior in a way that supports internal controls, governance reviews, and enterprise oversight requirements.

### Operational accountability

Give teams a defensible record of what ran, what changed, how it behaved, and what outcomes it produced.

## Built into the workflow that runs your agents.

Observability is native to the system that builds, orchestrates, deploys, and governs agents - not bolted on after the fact.

### 01 Richer trace context

Capture visibility from the orchestration runtime itself, including multi-step logic, delegated execution, and runtime state transitions.

### 02 Fewer tool handoffs

Trace, evaluate, monitor, and govern in one connected system rather than stitching together separate observability and quality tools.

### 03 Better operations

Keep debugging, release confidence, cost visibility, and auditability aligned to the same execution layer that runs the work.

## Visibility for agents that actually ship.

Trace execution, test quality, monitor production behavior, and maintain audit-grade visibility in one enterprise-native workflow.