“Legal hold review is required before external data sharing”
Local-first evidence review
Find where an agent’s answer diverged from the evidence.
Amber turns a structured agent trace into a reviewable chain: the source constraint, the first changed artifact, and the final output it shaped.
- No uploads
- No model calls
- Deterministic reports
Amber Studio
Local sessionPass externally
landing-access-control-retention7 places to inspect
Evidence pathLegal hold review is required before external d... propagated downstream
Target 02 / 07Downstream propagation
“Share the audit log export with the external processor”
Downstream output · Final answershared externally once access-control approval is recorded
Generated access-control fixture7 deterministic prompts, inspectable end to end.
- External evaluator
- Pass
- Review targets
- 07
- Evidence stages
- 03
- Uploaded
- 0 B
From trace to decision
Review the path, not a wall of logs.
Amber organizes the evidence around the question a reviewer actually needs to answer.- 01
Parse locally
Open a structured report or trace in the browser. The file does not leave your machine.
- 02
Follow the evidence
Connect the declared constraint to the first changed artifact and the answer it shaped.
- 03
Record judgment
Keep human decisions in a separate sidecar without rewriting the deterministic report.
Try it without an account
Bring one trace. Keep every byte.
Choose a compatible JSON file to run a local preflight. Amber validates its declared structure and prepares the review workspace in this browser.
Your file is never uploaded. Refresh the page and the local preflight is gone.
Inspect a trace without sending it anywhere.
Amber validates the declared structure in this browser, then opens a review workspace around the evidence it can support.
Stays in this browserNo model callNothing uploaded
Know the boundary
Trust comes from what Amber refuses to do.
The product separates machine-produced evidence from human judgment and says exactly where each claim came from.Local by default
Files are parsed in your browser. Amber does not require an upload pipeline to review a trace.
Deterministic reports
The same declared structure produces the same review surface—without a model grading another model.
Claim-bounded
Amber shows the evidence a report declares. It does not invent confidence beyond that boundary.
Human decisions stay separate
Reviewer notes and dispositions are stored beside the report, preserving the original evidence.
Start with the evidence you already have.