Using BetterContext Copy

Usage & Analytics

Track context query volume, calculate token cost savings, monitor recall latency, and stream telemetry into your monitoring stack.

Deploying AI agents at enterprise scale demands granular observability. BetterContext measures every memory query, citation click, cache hit, and causal evaluation to ensure your AI infrastructure remains fast, cost-effective, and accurate.

Core Performance Metrics

Recall Latency

P50: 140ms | P95: 280ms | P99: 310ms. Optimized localized vector traversal guarantees real-time responsiveness.

Token Savings Ratio

Average 84% token reduction compared to brute-force document dumps, dramatically lowering LLM provider inference costs.

Graph Cache Hit Rate

92%+ cache hits on repetitive agent reasoning steps, preventing redundant embedding queries and API calls.

Understanding Token Economics

In standard RAG, agents dump entire 100k+ token files into context windows for every single question. At scale, this quickly leads to thousands of dollars in wasted inference costs.

Synergistic Packing Economics:
• Standard RAG: 100 queries × 80,000 tokens = 8,000,000 tokens (~$40.00/day on Claude 3.5 Sonnet).
• BetterContext: 100 queries × 3,500 de-confounded tokens = 350,000 tokens (~$1.75/day).
• Annual Savings: Over $13,900 saved per developer seat while boosting output accuracy.

Exporting Metrics to Prometheus & Datadog

BetterContext provides standard OpenTelemetry endpoints for organizational monitoring:

# Enable OpenTelemetry metrics exporter in your daemon config
export BETTERCONTEXT_OTEL_ENDPOINT="https://otel-collector.internal.org:4317"
export BETTERCONTEXT_OTEL_SERVICE_NAME="bettercontext-prod"

Exported metrics include bettercontext_query_duration_seconds, bettercontext_cache_hits_total, and bettercontext_tokens_packed_total.

Next Steps

Was this page helpful?

ON THIS PAGE

Overview
Key Metrics
Token Economics
Exporting Telemetry
Next Steps