Skip to content
AGNT
Latency

Booking Response Time

2.8s

End-to-end booking confirmation in under 3 seconds.

Improving

Booking response time measures the wall-clock duration from when a user sends a booking intent message to when they receive a confirmed reservation response. This spans the full pipeline: NLU intent extraction, venue availability lookup via the tool executor, A2A envelope dispatch through ClawPulse (when the venue runs its own agent), confirmation persistence to PostgreSQL, and final message delivery back to the user's Telegram or WhatsApp channel.

The 2.8-second median is measured across all booking types — direct reservations (where AGNT holds the venue calendar) and A2A-routed bookings (where a venue agent processes the request). Direct bookings are faster at ~1.5s because they skip the network hop. A2A bookings add envelope signing, transit, and venue-side processing but still land well under the 15-second gateway timeout.

This metric matters because booking is a conversion event. Every additional second of latency erodes user trust and increases drop-off. We track p50, p95, and p99 separately — the 2.8s figure is the p50 (median). The p95 sits at 5.1s, driven primarily by cold-start venue agents on the A2A path.

Methodology

Measured via structured log spans from message receipt (webhook handler timestamp) to delivery confirmation (channel sender ACK). Spans are tagged with booking_type (direct vs a2a), venue_id, and platform. Aggregated hourly into percentile buckets in Redis, with daily rollups persisted to PostgreSQL for trend analysis. Outliers beyond 15s (gateway timeout) are excluded as failed bookings and tracked separately in the circuit breaker metrics.

People also ask.

See it in action.

2.8s seconds — real numbers from production. Try the live scan demo or explore more benchmarks.