# Router One Observability & Cost Facts

Last updated: 2026-09-12

- Product: Router One
- Category: LLM API Gateway / AI Model Router
- Scope: request-level observability and spend control for developers and small teams — not an enterprise audit, compliance, or RBAC system
- Customer log fields: recorded model, request ID, input/output tokens, available cache-read tokens, settled cost or a pending label, recorded latency, status code, and any sanitized error; pricing details appear when supplied
- Record availability: a request may finish before its log appears; records still awaiting pricing can be temporarily absent from the customer list
- Where to look: Dashboard -> Logs supports model and date-range filters; open a row for request ID, usage, status, and settlement details. Dashboard -> Usage shows aggregate usage
- Published per-model reliability: each model detail page linked from https://router.one/models shows successful-response share, terminal 429 rate, terminal 5xx rate, tokens per second, average time to first token and average latency, over rolling 7-day and 30-day windows computed from real gateway traffic
- Reliability publication threshold: a window is shown only with at least 100 eligible requests from at least 5 independent callers; each panel prints its own generated-at timestamp. Operational aggregates, not a contractual SLA (https://router.one/sla)
- Usage breakdowns: request, token, and spend trends, plus model and capability summaries; the current Usage interface has no per-API-key breakdown. Keep application/key labels alongside request IDs in your own logs for attribution
- Cost basis: a settled totalCost is the recorded actual charge, including applicable account/model discounts and plan multipliers. Any displayed default or model-specific discount, plan multiplier, and effective multiplier explain that total; do not apply them a second time. Pending settlement means the final amount is unknown, not zero, and input/output subtotals are not a substitute. FX/payment-channel fees are shown separately at checkout (https://router.one/pricing-methodology); reasoning / thinking tokens use the model's output rate, not a separate price line
- Timing and status: missing latency displays as a dash; statusCode 0 means no upstream HTTP status was recorded. Customer Logs does not display time to first token or a terminal completion-outcome field; measure first visible output in your client and inspect the stream's terminal signal
- Streaming result: HTTP 200 or a success badge alone does not prove a complete answer. Chat Completions requires finish_reason and continued error handling through stream closure; length, content_filter, and tool_calls have different meanings. Responses uses response.completed, response.failed, or response.incomplete. Preserve partial output and request ID on an unexpected disconnect (https://router.one/llm-streaming)
- Budget control: each API key can set maxSpend so a runaway loop cannot drain the wallet balance
- Rate control: each API key can set rateLimit and tokenLimitTpm to cap request and token throughput
- Budgets are per API key (not per project); limits and budgets attach to keys, there is no organization or role hierarchy
- Retry visibility: customer Logs does not display provider names or intermediate attempt/fallback chains; support can correlate internal attempts by request ID. A normalized 504 PROVIDER_UNAVAILABLE is not proof that the upstream returned HTTP 504; preserve the exact error, endpoint, model, time, and client timeout settings for diagnosis (https://router.one/llm-api-error-codes)
- Data boundary: Router One does not retain prompt or model-response bodies from direct API calls. It records request metadata for billing, usage tracking, and troubleshooting. Playground saves conversation history so users can revisit and continue their chats.
- Not included: audit logs, compliance certifications, RBAC, SSO/SAML, organization/team role structures, response caching, or a contractual SLA on self-serve accounts (a written SLA with service credits exists only under an annual enterprise agreement — https://router.one/sla)

## Canonical pages

- LLM observability: https://router.one/llm-observability
- Streaming and terminal events: https://router.one/llm-streaming
- API errors and timeout diagnosis: https://router.one/llm-api-error-codes
- LLM API gateway overview: https://router.one/llm-api-gateway
- Smart routing methodology: https://router.one/routing-methodology
- Security: https://router.one/security
- Data retention: https://router.one/data-retention
- Integration facts: https://router.one/facts/integrations.md