OpenRouter ships Activity dashboard and Analytics API for per-agent, per-model cost breakdown

OpenRouter has introduced an Activity dashboard and a beta Analytics API that give teams per-agent, per-model, and per-request visibility into AI costs. The dashboard aggregates metrics such as total spend, request counts, token volume, cache hit rate, and blended cost per million tokens, with sparklines and comparisons to previous periods. It also breaks down usage by top users, apps, models, and BYOK versus credit spend, and includes a Trends view that sorts by movement to spot rising or falling usage patterns.

Adopting this requires engineers to enable private input/output logging if they want prompt and completion details, since those only exist when logging was enabled at request time. The API is in beta, so it may change, and teams will need to learn its query syntax and integrate it into their existing dashboards or workflows. The dashboard itself is new, so there is a learning curve for navigating Explore, saving views, and drilling into logs.

The tool stops working well when historical data lacks the necessary logging: per-message token counts are estimates derived from message size, and prompt/completion details are absent if logging was off. The API's beta status means it might not be stable for production use, and the dashboard's reliance on recorded usage means any gaps in data collection will show up as incomplete views. Additionally, the guardrails feature only shows what was blocked, redacted, or flagged, so it does not provide full context on all requests.

For engineers, the main value is the ability to move from an aggregate cost spike to the specific requests behind it. Clicking any bar or slice lands in filtered logs, and the Generation detail view shows cost components, provider routing, latency, and attribution. The API extends this to automation, letting agents query their own usage or pull numbers into custom dashboards, which is useful for continuous cost monitoring and alerting.

The guardrails view adds a security dimension by showing what prompt injection and sensitive-information rules blocked, redacted, or flagged, and which rules are doing the work. This helps engineers monitor sensitive data entering prompts and adjust rules. Overall, the feature set addresses the common question of what agents cost and which are worth it, but its effectiveness depends on proper logging configuration and acceptance of the beta API's limitations.