[Dashboard Request] AISIX AI Gateway

Dashboard Name

AISIX AI Gateway Monitoring Dashboard

Expected Dashboard Sections and Panels

The exact panel set can be adjusted based on how the Prometheus receiver maps the AISIX metric attributes in the current SigNoz V2 dashboard schema.

Gateway overview

• Request rate — LLM requests per second from aisix_llm_requests_total.
• Success rate — Percentage of LLM requests with outcome="success".
• In-flight requests — Current requests grouped by endpoint or inbound protocol.
• Configuration health — Latest configuration reload state and rejected or partially compatible resources.

Model performance

• End-to-end latency — P50, P95, and P99 from aisix_request_e2e_latency_seconds histogram buckets.
• Time to first token — P50, P95, and P99 for streaming requests from aisix_request_ttft_seconds histogram buckets.
• Requests by provider and model — Traffic distribution using bounded provider and model labels.
• Requests by outcome — Success, client error, rate-limited, and upstream-error traffic.

Token and policy activity

• Input and output token rate — Token consumption over time and by model.
• Cache outcomes and hit ratio — Exact hits, semantic hits, misses, and bypasses when a cache policy is enabled.
• Rate-limit rejections — Rejections grouped by configured scope.
• Guardrail outcomes — Allowed, blocked, and bypassed guardrail activity when guardrails are configured.

Routing and delivery health

• Deployment failure rate — Upstream attempt failures by target model.
• Fallback outcomes — Successful and failed fallback attempts.
• Deployment state — Targets that are healthy or temporarily out of rotation.
• Usage-event delivery problems — Dropped usage events grouped by bounded reason.

Expected Dashboard Variables

• Gateway scrape job
• Gateway instance
• AISIX environment ID
• Provider
• Model

The final variable names will use the exact attributes retained by the SigNoz Prometheus receiver after validation.

Additional Comments or Requirements

AISIX exposes Prometheus metrics on a dedicated /metrics listener. The proposed contribution would use the SigNoz/OpenTelemetry Collector Prometheus receiver; it does not require a new SigNoz integration, AISIX exporter, or application SDK.

The dashboard would be tested with:

• AISIX 1.2.0
• SigNoz Cloud or SigNoz 0.135.0+
• SigNoz Dashboard V2 JSON (schemaVersion: "v6")
• Synthetic traffic covering success, upstream failure, streaming TTFT, cache, rate limiting, guardrails, and fallback behavior

The contribution would include a V2 JSON file, setup README, and sanitized screenshots. Metrics that only exist when a feature is configured would be documented as conditional instead of being presented as universal health signals.

Disclosure: I work at API7.ai, which maintains AISIX. AISIX is an Apache-2.0-licensed independent open-source project and is not an Apache Software Foundation project.

References or Screenshots

• AISIX GitHub repository
• AISIX metrics reference
• AISIX metrics and logs guide
• SigNoz Prometheus metrics guide
• SigNoz dashboard contribution guide

Screenshots will be attached only after the dashboard has been created and imported successfully in a clean SigNoz tenant.