Observability

Monitor agent performance, trace requests, and set up alerting.

Overview

Hyperfold provides comprehensive observability through metrics, logs, traces, and alerts. Monitor agent performance, track negotiations, and get alerted to issues before they impact revenue.

Observability data is stored in Google Cloud Operations Suite (formerly Stackdriver) and accessible via CLI, API, and web console.

Key Metrics

Track key performance indicators:

# View agent metrics
$ hyperfold metrics agents
AGENT METRICS (24h)
NEGOTIATOR: sales-bot-01
  Sessions:          1,247
  Conversions:       412 (33.0%)
  Revenue:           $67,840
  Avg Order Value:   $164.66
  Avg Discount:      12.3%
  Margin:            46.2%
  Response Time:     p50: 120ms  p95: 340ms  p99: 890ms
  Error Rate:        0.3%
  LLM Tokens:        2.4M tokens ($48.00)
FULFILLMENT: fulfillment-01
  Orders Processed:  389
  Shipments Created: 385
  Avg Processing:    4.2 minutes
  Carrier Split:     USPS 45%, FedEx 35%, UPS 20%
# Export metrics to JSON
$ hyperfold metrics agents --output=json --since=7d > metrics.json

Metric Categories

Category	Metrics
Business	Revenue, conversions, AOV, discount rate
Performance	Response time, throughput, error rate
Cost	LLM tokens, API calls, infrastructure
Agent	Sessions, negotiations, bundle uptake

Dashboards

# Open metrics dashboard
$ hyperfold dashboard open
> Opening https://console.hyperfold.io/dashboard...
# Available dashboards:
# - Agent Performance
# - Revenue Analytics
# - Negotiation Insights
# - Inventory Status
# - System Health
# Create custom dashboard
$ hyperfold dashboard create --name="Sales Overview" \
  --widgets='[
    {"type": "counter", "metric": "sessions_today"},
    {"type": "counter", "metric": "revenue_today"},
    {"type": "chart", "metric": "conversion_rate", "period": "7d"},
    {"type": "table", "metric": "top_products", "limit": 10}
  ]'
# Widget types: counter, chart, table, gauge, heatmap

Distributed Tracing

Trace requests across services:

# View traces for a session
$ hyperfold traces get --session=sess_abc123
TRACE: sess_abc123
[0ms] Session started
  │   buyer: openai_gpt4_buyer
  │   customer: cust_xyz (Gold)
[45ms] ├─ ACP Search "waterproof running shoes"
  │   │   products_found: 8
  │   │   semantic_confidence: 0.91
[120ms] │   └─ Vector Search (Vertex AI)
  │       │   query_embedding: 1536 dims
  │       │   results: 8
  │       │   latency: 75ms
[890ms] ├─ ACP Quote prod_aero_x2 @ $120
  │   │   context: {tier: gold, history: 12 orders}
[920ms] │   ├─ Get Product (Firestore)
  │   │   │   cache_hit: true
  │   │   │   latency: 12ms
[980ms] │   ├─ Calculate Dynamic Price
  │   │   │   floor: $92, target: $162
[1050ms] │   ├─ LLM Decision (OpenAI)
  │   │   │   tokens: 847
  │   │   │   decision: counter_offer
  │   │   │   latency: 130ms
[1100ms] │   └─ Response: counter_offer @ $162
[32000ms] ├─ ACP Checkout Finalize
[32200ms] │   ├─ Process Payment (Stripe)
  │   │   │   status: succeeded
[32500ms] │   └─ Create Order
  │       │   order_id: order_789
[32550ms] └─ Session complete
            outcome: conversion
            revenue: $162.00
            duration: 32.5s

Alerting

Set up proactive alerting:

# Configure alerts
$ hyperfold alerts create \
  --name="High Error Rate" \
  --condition="error_rate > 5%" \
  --window="5m" \
  --channels="slack:#alerts,email:ops@company.com"
> Alert created: alert_xyz123
# Alert conditions:
# - error_rate > X%
# - response_time_p95 > Xms
# - conversion_rate < X%
# - revenue_24h < $X
# - inventory_low (product specific)
# List active alerts
$ hyperfold alerts list
ACTIVE ALERTS
NAME                CONDITION              STATUS    LAST TRIGGERED
High Error Rate     error_rate > 5%        OK        Never
Slow Response       p95 > 500ms            OK        Dec 18, 14:30
Low Conversion      conversion < 20%       ALERT     Now
# View alert history
$ hyperfold alerts history --alert="High Error Rate"
# Acknowledge alert
$ hyperfold alerts ack alert_xyz123 --message="Investigating"

View detailed logs with Chain of Thought Logs.