See exactly where your Anthropic bill goes.
Claude Recall computes per-session, per-project, per-tag cost analytics locally from your Claude Code session metadata. Which session cost you fifteen dollars. What your weekly trend looks like. Whether the experimental project ate more than the production one this month. Three questions the Anthropic console cannot answer.
What it answers
- Which session cost me fifteen dollars?
- What is my weekly Claude Code spend?
- Is my usage trending up or down?
- Which project burns the most tokens per outcome?
- Are my long sessions actually more productive than my short ones?
The command
From the CLI:
recall statsIn the web UI: open the dashboard, click Stats.
Anthropic console vs. Claude Recall stats
| Question | Anthropic console | Claude Recall |
|---|---|---|
| Total monthly spend | Yes | Yes |
| Per-session cost | No | Yes |
| Per-project breakdown | No | Yes |
| Per-tag breakdown | No | Yes |
| Trend over time | Limited | Yes |
| Computed locally without sending data anywhere | N/A | Yes |
Codex counts too
If you run the Codex CLI as well, Claude Recall indexes those sessions beside your Claude Code ones, carries a Codex badge, and prices them from a separate OpenAI rate table. Same views, same breakdowns, one number for the week across both agents.
The rate table is deliberately strict, because a cost report you cannot trust is worse than none:
- Priced by exact model id, never by family. A model we have no published rate for is not quietly billed at its nearest sibling's price.
- Every rate carries the date it was read. Providers publish prices without an effective date, so the table records when we captured each one, in dated windows. When a price does change, the new rate opens a new window and your older sessions keep costing what they cost at the time.
- Estimates are labelled. The two Codex model ids OpenAI publishes no rate for are estimated and marked
(est.)in the output. Cache-write tokens with no published rate are counted as unpriced rather than given an invented number. - It is a list-price estimate. Claude Code on a Max plan and Codex on a ChatGPT plan are both billed differently from list. The number tells you where your tokens went, not what your card was charged.
Privacy
Cost numbers are computed from the token counts inside the session files Claude Code and the Codex CLI already write to your disk. Claude Recall reads those files locally; nothing gets sent to a server. Your spending pattern stays on your machine.