The pipeline keeps its receipts. All of them.
Per-issue cost ledger. Full audit trail. Every prompt sent, every dollar spent, every state transition retained. The answer to “why did this PR cost what it did” lives in the ledger, not in someone’s head.
Every dollar attributed.
Each prompt, each tool call, each completion is logged with its model, its token count, and its dollar amount. The ledger rolls up to the issue, the agent, the tenant, and the time window. Export to your warehouse, or read it directly.
- Per-issue rollup with full prompt-level breakdown
- Per-agent comparison across issues
- Per-tenant attribution for multi-team workspaces
- CSV / API export for finance and BI tooling
| Agent | Issues | Tokens | $ |
|---|---|---|---|
| builder | 184 | 24.1M | $612.40 |
| inspector | 184 | 8.3M | $184.22 |
| surveyor | 184 | 6.1M | $148.66 |
| architect | 184 | 5.7M | $138.30 |
| sentinel | 184 | 1.2M | $32.18 |
| marshal | 184 | 0.8M | $18.42 |
| mayor | 184 | 0.3M | $7.99 |
Every decision recorded.
Each state transition is stamped with the agent, the timestamp, the input prompts, the model output, and the artifacts produced. You can replay an issue’s full history end to end, or open a single transition to see exactly what the agent was looking at.
- Every prompt and every response retained
- Tool call inputs and outputs preserved
- Replayable transition timeline per issue
- Retention configurable per tenant (default: lifetime)
A queryable interface, not a dashboard.
The ledger and audit trail are queryable Postgres views, exposed through a stable API. Build your own dashboard. Wire it into your BI tool. Alert on spend per agent. The point is that the data is yours, structured, and not trapped behind a vendor UI.
API reference// query the ledger for an agent's spend
SELECT agent, SUM(cost_usd)
FROM cost_ledger
WHERE tenant = 'colony'
AND ts >= now() - '7 days'
GROUP BY agent
ORDER BY 2 DESC;See your team’s ledger.
The pilot ships with full read-access to the ledger and audit trail for your repos.