Every AI call, with its cost and its owner.
A searchable, exportable record of everything the AI has done in your processes: for audit, for budget and for understanding.
The problem
When AI enters processes, the questions come immediately: how many calls do we make, what do they cost, who generates them, what was sent to the provider? Without a log the answers are guesses, the provider's invoice arrives as a surprise and, in case of a dispute, it is impossible to reconstruct what the model saw. Management does not know which process to charge the spend to, IT does not know which integration consumes the most, and whoever handles compliance has no document to show those asking for evidence. A useful log must be searchable, filterable, exportable and, above all, separate from operational data and protected: FlowSharp builds it that way.
Filters and total
- Total in period: €12.40
Log
- Date · Channel · Model · Outcome · Cost · Attributed to · shield
- AI task · ok · €0.03 · MCP key · shield
- Summary · ok · €0.01 · user · shield
How it works
1Choose the level per provider
Off (default), metadata only or full (messages and response). You can decide provider by provider how much detail to keep, depending on risk and audit needs.
2Every call is recorded
Channel (task, decision, title, summary, Flow Agent, Test, design), outcome, model, tokens, duration, cost estimated from the price list and attribution to a user, API key, MCP key or Flow Agent.
3Filters and analysis
By provider, channel, outcome, shielding, case, task and decision-ladder step, with the total of the filtered cost. A question like "how much does this channel cost?" has an immediate answer.
4You export
CSV (metadata) or JSON (full), with the export itself traced in the audit. Even the act of extracting the data leaves a trace, so exporting is not an invisible action.
Specifications
| Levels | off · metadata only · full |
|---|---|
| Shielding | with Data Shield on, pseudonyms appear, never real values |
| Cost | estimated from the model price list, with filtered total |
| Attribution | user, API key, MCP key, Flow Agent |
| Export | CSV/JSON up to 10,000 rows, traced |
| Retention | from 1 to 3,650 days per organization (30 preset) |
| Access | Admins only; not reachable by agents, MCP or templates |
An operational example (illustrative values)
The AI call log lists every call made by the platform with date, channel, model, outcome, cost, the person or key it is attributed to and a shield indication. In the example, for illustration only, an AI task succeeded at a cost of €0.03 and is attributed to an MCP key; a summary succeeded at €0.01 and is attributed to a user. Both show the shield. At the top, filters by period and channel and a total: in the example €12.40 in the period. The values are invented to show the shape and are not real data. The usefulness is concrete: whoever manages the budget sees how much AI costs and who uses it, whoever manages compliance sees which calls occurred and with what protection. The diagram beside it is not an official screenshot.
How to use the log to verify
Use the log as a periodic control tool. Filter by period and look at the total: if it has grown, sort by channel and by who made the call to see where the increase comes from. Verify that calls with an unsuccessful outcome are few and understandable, because a failed call is often a cost without value. Check that each row has the shield indication: a row without the shield deserves immediate follow-up. Attribution lets you tell apart use by people, MCP keys and internal functions such as AI tasks. For governance, the log is the proof that AI use is traced and proportionate, and lets you account for spending and protections to whoever asks. If a broader record is needed, the data exportable from the log can support periodic reviews of AI use.
Related features
- Data Shield0 sensitive data exposed: reserved fields become placeholders before every AI call.
- Decision metricsAutomatic versus human decisions, fallback rate, escalation rate and system1/system2 split, usable as KPIs.
- AI execution transparencyIn the task detail: model, tokens, duration, shielded values and the question, answer, confidence, field table.
- Agent attributionHistory shows the AI client and the person holding the key; the cost goes to the key.
- Activity logWho, when, through which channel and on what, including logins, failed attempts and two-step verification.
Use cases
Frequently asked questions
Does it record the personal data sent to the AI?
At the full level it records the messages, but with Data Shield on only pseudonyms appear in them.
Can I tell which user or integration uses the most AI?
Yes: each entry attributes the spend to a user, API key, MCP key or Flow Agent.
How long is the data kept?
From 1 to 3,650 days, configurable per organization.
Can AI agents read the log?
No, the log is not reachable by agents, MCP or templates.