What this tells you: Which agents are consuming the most budget. Taller bars mean higher cumulative spend for that agent across all interactions. Compare agents to find where your money is going — a high-cost agent may be handling more requests, or it may be using a more expensive model.→ Action: If one agent dominates cost, consider whether it can use a cheaper model or if its workload can be reduced through better routing.
Error Rates
What this tells you: The percentage of failed interactions per agent. Any agent above 5% error rate needs investigation. Zero-error agents are stable. Errors may indicate model timeouts, rate limits, malformed inputs, or provider outages.→ Action: Click the agent filter in Recent Events to see the specific error details for failing agents.
Compression Trend
What this tells you: How much redundant data AgentFlow is stripping out of inter-agent communication over time. Higher percentages = more waste eliminated = more cost saved. A consistent range of 50–80% indicates healthy compression. Dips may indicate new data patterns the system is still adapting to.→ Insight: Compression directly reduces your token spend. 60% compression on a £100/month pipeline saves £60.
Latency Over Time
What this tells you: Response time for each agent interaction, colour-coded by agent. Dots clustered low = fast, consistent performance. High outliers indicate slow responses that may be degrading your pipeline. Look for patterns — if one colour is consistently higher, that agent is your bottleneck.→ Action: Agents consistently above 1500ms may benefit from model switching, caching, or parallel execution.
Intent Distribution
What this tells you: How your API traffic breaks down by communication purpose. Dominant intents represent your most common workflows. Rare intents may indicate underused features or edge cases worth investigating.→ Action: Frequent intents are prime candidates for response prediction — if the pattern is predictable, AgentFlow can skip the model call entirely.
Tokens In/Out by Agent
What this tells you: Token consumption per agent, split by input (received) and output (generated). High input tokens mean the agent is receiving large contexts — a sign that upstream agents may be over-communicating. High output tokens mean the agent generates verbose responses. A large gap between In and Out indicates either summarisation (good) or bloat (investigate).→ Insight: Agents with high In tokens benefit most from AgentFlow's change-only transmission — if 90% of the input is unchanged, 90% of the cost disappears.
Recent Events
Loading events...
API Key Management
********
Rotating invalidates your current key immediately. Save the new key — it won't be shown again.
Subscription
Cancel your subscription. You will retain access until the end of your current billing period.
Telemetry Intelligence Engine
Automated analysis of your pipeline execution data
Compression Summary
What this tells you: The total data reduction AgentFlow achieves across all your agent communication. The estimated savings shows what you would be paying without compression.
Loading...
Bottleneck Analysis
What this tells you: Which agents consume the most time in your pipeline. Bottlenecks are agents with latency significantly above average, candidates for parallel execution or model switching.
Pipeline Patterns
What this tells you: Which agent sequences produce the best results, ranked by composite score combining success rate, cost, and speed.
Redundancy Detection
What this tells you: Agent pairs producing similar outputs, suggesting one may be redundant. Consider merging to reduce cost.
Route Predictions
What this tells you: The recommended agent sequence for each intent type, based on historical performance.
Cacheable Responses
What this tells you: Agent-intent combinations where responses are stable enough to cache, skipping the model call entirely.
API Explorer
Full interactive API documentation. Click any endpoint to expand details and test live.