Every step on record, from file to final answer.
Each run keeps its inputs, step outputs, citations, confidence scores, approvals and credit cost. Answer who did what, when and why without rebuilding the story from logs.
What every run records.
Query any run by id to see each step in order, or filter past runs by type, status and your own metadata.
Inputs and outputs
The files that came in and what each step returned, in order.
Evidence
Citations to page or timecode, and the confidence score on each output.
Decisions
Every approval or rejection, with who made it and when.
Cost
Credits spent per step and per run, so finance sees the same record as engineering.
$ curl https://api.cloudraker.com/v1/runs/plr_01KY... parse completed 3 files, 14 pages extract completed 4 fields cited approval approved ada@example.com fill completed w9-filled.pdf credits 18
Clear on who wrote what.
Model output, human edits and agent actions stay apart, so a reviewer can tell a cited value from a typed one.
Model output
Each value keeps its citation and score until someone changes it.
Human edits
A manual edit drops the model citation and records the person who made it.
Agent actions
Calls made by an agent carry that agent's name, within the permissions you granted.
Keep runs as long as you need, and no longer.
Runs expire on their own by default. Delete a run at once, or pin it for long-term retention.
| What happens | Use it for | |
|---|---|---|
| Default | The run expires automatically | Everyday volume with nothing to keep |
| Delete | The run and its files go at once | Sensitive files after the answer ships |
| Pin | The run stays for long-term retention | Audits, disputes and regulated records |
Questions about the audit trail.
Query it by id, or filter runs by type, status and the metadata you attached when you started them.
Run one file and read its full record.
Every run on your $15 of free monthly usage keeps the same audit trail as production.