slaude Docs
GitHub

Field notes

2026-07-08 — Per-channel token and model metrics

Context: Token consumption was previously aggregated strictly by kind (input/output/cache_read/cache_creation). To enable downstream cost calculation (e.g., in Grafana), the metrics needed granularity down to the specific Slack channel and the active LLM model.

Decision: Augment the slaude_tokens_total Prometheus counter with two new labels: channel_id and model.

Mechanism:

  • manager.ts fetches the active slack_channel_id from the SQLite session store using the sessionId when the SDK yields a result.
  • Extracts the active model from the SDK message, falling back to SLAUDE_MODEL.
  • Passes { channel_id, model, kind } to metrics.tokensTotal.inc().
  • Exposing the model and channel labels allows downstream aggregators to multiply token counts by per-model rates to estimate cost accurately without hardcoding pricing within the slaude runtime.