Skip to content
Book a demo

AI governance

An audit trail of everything the AI did? What SAGARIS records, and what it does not

SAGARIS keeps an audit trail of what its AI agent and writing agents do. Every action the agent plans, proposes, is refused or completes is recorded with its inputs, evidence and outcome, and when a person approves or rejects one, the record names who. Its writing agents keep the prompt they sent and the text they got back. It is not everything: some model calls, such as call transcription, sit outside it.

Checked against the SAGARIS product code on 25 September 2026.

One agent tool call, as the activity trail lists it

  1. plannedagentsend_email, proposalwritten: yes
  2. held for approvalagentsend_email, proposalwritten: yes
  3. decidedadminapproved, reason kept, nothing sentwritten: yes
  4. completedagentupdate_contactwritten: yes
  5. stopped by the outbound switchagentenroll_in_sequencewritten: yes

All stages: planned, held for approval, proposed, completed, failed, refused, stopped by the outbound switch, skipped as a duplicate.

Read in the audit workbench with View the audit log. Payload: admins only. Not tamper-evident.

The audit log reference

What SAGARIS records about what its AI did

There is not one log, there are several, and each answers a different question.

  • What the AI agent did

    Each tool call the agent makes is written to the activity trail at the stages listed.

    payload: admins only

    Written toactivity trail

    Read by: Audit log viewers

  • Who approved or rejected it

    Who decided, when, any reason and a receipt; approving an email proposal still sends nothing, and the receipt says so.

    Written toactivity trail

    Read by: Audit log viewers

  • What the writing agents were given and wrote

    Thirteen agents do the writing: research, digest and message-writing agents keep the prompt they sent and the text they got back.

    (thirteen registered; five serve the Social module, which is not generally available)

    Written toharness traces

    Read by: Any member

  • Model calls made through the gateway

    A usage row is written before the call starts; if that row cannot be written, the call does not happen.

    Written tousage ledger

    Read by: Audit log viewers

  • Changes to how the AI is allowed to behave

    Before and after values for settings such as the workspace autonomy level, plus API keys created and revoked.

    Prior values are redacted unless the prior value is itself part of the record.

    Written toconfiguration trail

    Read by: Audit log viewers

  • Data exports

    Each export attempt is recorded, allowed or refused; an export whose record cannot be written is refused.

    The export record is written but there is no screen that lists it yet.

    Written toexport record

    Read by: No screen yet

Who can read it

Reading is split by permission, and the split is not the same for every record. Audit log viewers are the members whose role holds View the audit log: Admin, Director (workspace-wide), RevOps admin and Billing admin by default, and any custom role given it.

Any member of the workspace

writing-agent traces
full
activity trail
needs permission
usage ledger
needs permission
configuration trail
needs permission

Audit log viewers

activity trail
summary
usage ledger
full
configuration trail
full
activity payload
admins only

The payload needs a full Admin. A request for it without that is refused outright rather than answered with the payload quietly removed.

From inside the product, not the public API

in the product
signed-in
public API
no operation
API key
cannot pull it
export
none today
A seismograph drum with a fine ink trace on its paper, in soft window light.

What the trail does not do

Check an action against this list before assuming the trail covers it.

  1. It is not tamper-evident

    No stored hash links; the verification endpoint answers that no persisted chain is available.

  2. There is no retention control

    There is no configurable audit retention period. The call-recording setting does not apply here.

  3. Not every model call is in the usage ledger

    Transcription, generated speech and the classifier that decides how a dialed call was answered bypass the gateway.

  4. Records are mostly written after the action

    For most trails, a failed write does not stop or undo what the AI did. Exports are the exception.

  5. Deleting an account changes the trail

    A deleted person's rows in workspaces they did not own stay but no longer name them; deleting an owner's account deletes the workspace and its trail.

  6. Some AI writing and one setting are outside it

    The prompt behind an inbox reply draft is not kept, and switching inbox Auto-Mode on or off is not written to the configuration trail.

Questions buyers ask next

More questions? Book a demo
  • SAGARIS keeps an audit trail of what its AI agent and its writing agents did, though not of every model call the platform makes, and this page names the gaps rather than claiming full coverage. Ask each vendor what its trail leaves out.

    Each agent tool call is recorded at the stages listed above with its parameters, the policy verdict and the evidence it used, a decision on an agent action names the person who made it, and the research, digest and message-writing agents keep the prompt they sent and the text they got back. When you compare AI sales tools, ask each vendor, SAGARIS included, to show you the record rather than describe it.

  • Yes. A decision on an agent action stores who decided, when, any reason they gave and a receipt of what actually happened, and writes a matching row to the activity trail under that person's name. A reply draft that a rep sends records which rep sent it. A running sequence and a launched inbound campaign send on schedule; the AI agent proposes; drafts wait for a person.

  • For the writing agents, yes. Each model call they make is traced with the prompt as sent and the raw response, each kept in full, along with tokens, cost and the contact, sequence and step it was for. Any member of the workspace can read these traces.

  • Any member can read the writing-agent traces. The activity trail in summary, the model usage ledger and the configuration trail need the View the audit log permission, which Admin, Director (workspace-wide), RevOps admin and Billing admin hold by default and a custom role can be given. The recorded input and output on an activity row need a full Admin. It is read inside the product; the public API does not expose it and there is no export of the trail today.

  • No, and this page says so plainly rather than leaving you to discover it during a security review. The audit rows carry no stored hash links, and the chain verification endpoint returns a not-implemented response instead of a result, so there is nothing that would prove a row had not been altered. If your compliance programme needs tamper evidence specifically, the audit trail does not provide it today, and you should treat it as a record of what happened rather than proof that the record is unaltered.

  • There is no configurable audit retention period, so no period is promised in either direction: neither a guaranteed minimum to rely on, nor a guaranteed deletion to point a privacy reviewer at. Call recordings have their own separate retention setting, and that setting does not apply to the audit trail, which is worth knowing if you are mapping either one to a retention policy.

Bring the audit question your reviewer asks and we will show which rows answer it.

Book a demo