Every decision on record.
An answer from AI is only worth something if you can show how it came about. In Navabi every item carries the whole chain: who asked, what the agent read and answered, why it went to a lawyer, who signed off and what you did with it. Not reconstructed afterwards, but recorded at the moment itself.
- Question askedL. Veldkamp, via Slack · 12 Oct 09:12 · "Can Studio Linde terminate with immediate effect?"
- Agent read and answeredSources: SaaS agreement art. 9, art. 6:237 BW · model v2026.10 · 09:13
- Escalation: money at stakeInvoice dispute of € 12,400 found in Matters · rule "above € 10,000 to a lawyer" · 09:13
- Lawyer reviewedAnswer confirmed, one addition · 11:40 · name and registration in the file
- Client approvalL. Veldkamp · 11:52 · answer sent from own email
Available. The audit log is in every plan and fills from your first question. What is new is in the changelog.
Five lines per decision. Always the same.
Whether a question came from Slack, your own AI, an integration or the workspace: the audit log looks the same and can be exported per matter.
The chain, not just the outcome
Who asked the question and where. Which sources the agent used, with the model version. Which rule triggered the escalation. Which lawyer reviewed, with name and time. What the client decided.
- Exportable per item as a pdf or as data
- Sources with their location, including your own contracts
- Immutable: a correction is a new line, not an overwrite
- Question askedL. Veldkamp, via Slack · 12 Oct 09:12 · "Can Studio Linde terminate with immediate effect?"
- Agent read and answeredSources: SaaS agreement art. 9, art. 6:237 BW · model v2026.10 · 09:13
- Escalation: money at stakeInvoice dispute of € 12,400 found in Matters · rule "above € 10,000 to a lawyer" · 09:13
- Lawyer reviewedAnswer confirmed, one addition · 11:40 · name and registration in the file
- Client approvalL. Veldkamp · 11:52 · answer sent from own email
Ready for whoever comes asking
The AI Act requires companies that use AI to be able to demonstrate human oversight and logging. An accountant wants to know who approved an obligation. A court wants to know what you knew and when. The audit log answers all three without you having to search for anything.
- Human oversight demonstrable per decision
- Retention period of your choice, seven years by default
- Export for accountant, regulator and court
- Export requestedAccountant · 14 Oct
- Audit log 2026 compiled312 decisions · 41 signed off by a lawyer
- Shared via the data roomRead only, 30 days
What your agent may commit to on its own, per team.
A mandate is a rule with a limit attached. Within it the agent acts without a lawyer and records what it did; beyond it the item goes to a lawyer; and some things can never happen. You set the limits, per team.
| Team | May do alone, without a lawyer | Goes to a lawyer | Never |
|---|---|---|---|
| Sales | Payment term up to 45 days, discount up to 15%, term of 12 or 24 months | Liability, exclusivity, the client's own terms, above € 50,000 | Unlimited liability, foreign law, verbal agreements |
| HR | Employment contract from template, addendum for salary or role | Settlement agreement, non-compete outside the list, any conflict | Contract that deviates from the collective labour agreement (cao), dismissal without a file |
| Procurement | Contract within procurement policy up to € 25,000, annual renewal | Indexation above the cap, term over two years, any contract above € 25,000 | Tacit renewal without approval, exclusivity |
| Finance | Reminder with statutory interest, payment plan up to three months | Demand letter, debt waiver, bailiff | Write-off above € 5,000 without board approval |
Example. You set the limits yourself; the agent checks every action against them and the audit log shows within which mandate something happened.
One line from the audit log.
This is what a line looks like when you export it. Every line points to the previous one, so nobody can insert or remove anything afterwards. Not even us.
- Append onlyLines are never changed or deleted. A correction is a new line that points to the old one.
- A hash chain per clientEvery line contains the hash of the previous one. Change a single character anywhere and the chain breaks from that point.
- A timestamp every dayOnce a day we fix the state of all chains in a Merkle root with a qualified timestamp. That proves a line already existed on that day.
- Verifiable without usExport as JSON Lines with the timestamps included. Your auditor or regulator can check the chain themselves.
{
"id" : "log_2026_0412",
"time" : "2026-10-12T09:38:51+02:00",
"tenant" : "tnt_veldkamp",
"matter" : "mt_people_0193",
"channel" : "slack",
"asked_by" : "usr_lisa",
"event" : "signed_off",
"sources" : [
{ "type" : "contract", "doc" : "doc_8812", "version" : "sha256:91c0…7ab2", "passage" : "art. 2" },
{ "type" : "law", "ref" : "BW 7:668", "as_of" : "2026-10-16" }
],
"model" : { "family" : "claude", "region" : "eu", "prompt" : "ask.employment@14", "playbook" : "pb_veldkamp@7" },
"mandate" : "hr.non_renewal",
"lawyer" : "jur_031",
"prev" : "sha256:5d1e…a903",
"hash" : "sha256:7f3a…c21e"
}Short answers.
Is the audit log also evidence in court?
It is a record made at the moment itself, immutable and time-stamped. That is what a court wants to see. Whether it is decisive in a case depends on the case; we cannot promise that.
Does Navabi see my audit log?
A lawyer sees the decisions in the files they work on. Beyond that it is yours. We do not use it for training.
What happens if the agent makes a mistake?
Then it is on record exactly what the agent read and answered, and whether a lawyer saw it. What a lawyer signed off falls under our professional liability; what only the agent did is marked as an AI draft and never went out.
Can I extend a mandate temporarily?
Yes, with an end date. That extension is also in the audit log, with who granted it.