A daily view of HiveXS spend across jobs. Updated end-of-day from the Anthropic Admin API and our local activity log.
HiveXS runs mostly on cloud AI models. Every call costs something. We track each dollar and attribute it back to the task that triggered it. Three numbers worth watching: total spend this month, spend per task, and spend per client.
Day by day, total cost and how many distinct tasks the spend is split across.
| Date | Spend | Tasks active |
|---|---|---|
| 2026-05-08 | $91.45 ~AUD $126.25 | 5 |
| 2026-05-07 | $194.43 ~AUD $268.41 | 3 |
| 2026-05-06 | $102.51 ~AUD $141.51 | 7 |
| 2026-05-05 | $88.54 ~AUD $122.23 | 5 |
| 2026-05-04 | $24.05 ~AUD $33.20 | 1 |
| 2026-05-03 | $42.29 ~AUD $58.38 | 10 |
| 2026-05-02 | $12.60 ~AUD $17.39 | 6 |
| 2026-05-01 | $47.96 ~AUD $66.21 | 8 |
| 2026-04-30 | $10.06 ~AUD $13.89 | 1 |
Where the money is going. HiveXS-comms = website + content + decks. HiveXS-system = infrastructure + runtime. _orphan = activity hours where no task LOG event was written; close that gap by extending LOG coverage.
| Client | Total | Tasks |
|---|---|---|
| HiveXS-comms | $261.01 ~AUD $360.32 | 10 |
| _orphan | $140.60 ~AUD $194.10 | 1 |
| HiveXS-system | $134.93 ~AUD $186.27 | 15 |
| SHWR | $43.73 ~AUD $60.37 | 7 |
| AUSM | $33.62 ~AUD $46.41 | 3 |
Tasks ranked by cost. Useful for spotting the heaviest jobs and asking whether they are worth what they are costing.
| Task | Client | Cost | Input tokens | Output tokens |
|---|---|---|---|---|
| T-2026-05-06-006-hivexs-website-v1 | HiveXS-comms | $205.56 ~AUD $283.77 | 48.5k | 1120.1k |
| T-2026-05-06-003-hivexs-knowledge-architecture-phase-2-build | HiveXS-system | $49.37 ~AUD $68.15 | 1.1k | 248.6k |
| T-2026-05-05-003-showerloop-favicon-fix | SHWR | $22.96 ~AUD $31.70 | 0.7k | 76.3k |
| T-2026-05-08-001-hivexs-cost-accounting-design | HiveXS-system | $19.89 ~AUD $27.46 | 2.4k | 125.3k |
| T-2026-05-06-001-hivexs-knowledge-architecture | HiveXS-system | $18.50 ~AUD $25.54 | 0.8k | 117.2k |
| T-2026-05-05-002-ausmade-strategy-doc-proof | AUSM | $17.11 ~AUD $23.62 | 0.7k | 83.0k |
| T-2026-05-07-001-linkedin-reply-colin-stevens | HiveXS-comms | $12.69 ~AUD $17.52 | 1.9k | 45.6k |
| T-2026-05-04-004-ausm-onboarding | AUSM | $10.77 ~AUD $14.87 | 0.4k | 123.9k |
| T-2026-05-01-001-focus-group-cause-effect-v2 | HiveXS-comms | $10.24 ~AUD $14.14 | 0.2k | 197.9k |
| T-2026-05-08-002-hivexs-local-model-strategy | HiveXS-system | $10.24 ~AUD $14.14 | 0.7k | 66.9k |
| T-2026-05-05-001-email-phase-3-inbound | HiveXS-system | $9.30 ~AUD $12.84 | 0.3k | 106.5k |
| T-2026-05-01-006-showerloop-onboarding | SHWR | $8.91 ~AUD $12.30 | 0.2k | 102.1k |
| T-2026-05-03-009-system-team-summary-deck-v2 | HiveXS-comms | $7.57 ~AUD $10.45 | 0.1k | 49.8k |
| T-2026-05-03-008-system-team-summary-deck | HiveXS-comms | $7.41 ~AUD $10.23 | 0.2k | 41.0k |
| T-2026-05-06-004-ausm-edm-clean-html-export | AUSM | $5.75 ~AUD $7.94 | 0.2k | 58.2k |
Where the tokens flowed. Cache reads dominate cost on opus-class models because long-running conversations re-read the cached context every turn.
| Model | Input | Output | Cache read | Cost |
|---|---|---|---|---|
| claude-opus-4-7 | 74.3k | 4405.7k | 1001548.5k | $611.29 ~AUD $843.88 |
| claude-sonnet-4-6 | 0.1k | 100.1k | 3696.4k | $2.61 ~AUD $3.60 |
A practical note on the numbers. Anthropic's Admin API gives us hourly token usage per model. We split that hour's cost across whichever HiveXS tasks were active in the same hour, weighted by LOG event count. Tasks with no LOG activity get nothing; orphan hours (no task active) go to the _orphan bucket. Approximate but directionally accurate. Phase 2 (per-call attribution at the OpenClaw runtime layer) would make this exact.