The Audit tier — paid upgrade
The base product gives your agents clean, read-only, permissioned access. The Audit tier is the record of that access: a complete, tamper-evident, examiner-ready log of every query an agent makes against your client data — built for the CCO’s monthly review and for the day an examiner asks, “show me what your AI tools have touched.”
New to agentdatasync? Start with the free Scorecard and the CCO questions on the home page. Scope, as always: agents read, humans decide — the Audit tier records access; it never moves money, trades, or advises.
One line per tool call, appended at the moment the call completes — whether it succeeded, returned nothing, errored, or was refused. Nothing is sampled and nothing is batched after the fact.
| Field | What it records | Why the CCO wants it |
|---|---|---|
agent_id |
The identity of the calling agent (e.g. meeting-prep, client-qa) and the firm user it acted on behalf of. |
Accountability. “Which tool, acting for whom” is the first question after any incident. |
ts |
UTC timestamp, to the second, taken when the query executed. | Reconstructs a timeline that can be matched against your other records. |
tool |
The MCP tool invoked: list_households, list_clients, get_client, search_notes, or list_accounts. |
Shows the kind of access — a notes search is a different risk conversation than an account listing. |
args_hash / args_summary |
A SHA-256 hash of the exact arguments, plus a redacted human-readable summary (IDs and search terms; never note bodies or free text). | The hash proves later exactly what was asked, without turning the log itself into another pile of sensitive client text. |
records_touched |
The client and household IDs the query read or attempted to read. | Scoping an incident becomes a lookup, not an investigation: these clients, these households, nothing else. |
result_count |
How many records came back — including zero. | Volume is the anomaly signal: a query pattern that suddenly returns hundreds of records looks different here immediately. |
outcome |
ok, error, or a denial such as denied_out_of_scope when an agent asked outside its permissions. |
Denials are evidence your permissions are working — and they are logged with the same fidelity as successes. |
duration_ms |
Execution time of the call. | Operational health, and corroborating detail when reconstructing an event. |
The full field specification, including the reference server’s on-disk format, is in the docs. Field names above are the Audit tier’s canonical names; the downloadable reference server writes the core subset (timestamp, tool, arguments, result_count, duration_ms) so you can verify the approach yourself.
Five lines from a fictional firm’s morning. Note the fourth line: the permission system refused an out-of-scope request from the client-Q&A agent, and the refusal is on the record.
Illustrative excerpt — fictional firm, synthetic records, real field format.
A log nobody reads is a liability with good formatting. The Audit tier is built around a review rhythm a part-time CCO at a small firm can actually sustain.
MONTHLY
A monthly review pack summarizes the period: queries by agent and by tool, the users agents acted for, total records touched, and every denial and error. The CCO reviews it, signs off, and files it — the sign-off itself becomes part of your compliance record.
CONTINUOUS
The log is screened for the patterns a human would miss in raw JSON: an agent querying far outside its usual hours, a sudden spike in result counts, repeated denials (an agent — or a user — probing the permission boundary), and first-time access to a household. Flags go to the CCO; the underlying lines are one click away.
ON REQUEST
The complete log exports as plain JSON Lines plus a human-readable summary, filtered by date range, agent, or client. No proprietary viewer, no vendor call required — your examiner reads the same file you do. Designed to support your books-and-records obligations under Rule 204-2: records retained five years, the first two years easily accessible, arranged and indexed for retrieval, and safeguarded against loss or alteration (sourced: 17 CFR §275.204-2, incl. paragraph (g) electronic-storage conditions). This supports your program; it does not by itself guarantee compliance — your firm remains responsible for its own recordkeeping determinations.
The audit log only tells half the story; the other half is what each agent was allowed to do in the first place. Permissions are set per agent and per data domain, and every change to the matrix is itself logged. A typical starting configuration:
| Data domain | Meeting-prep agent | Client Q&A agent | Ops reporting agent |
|---|---|---|---|
| Households & client identity | Read | Read | Read |
| Contact details (email, phone) | Read | Read | Counts only |
| Client notes (search) | Read | Own clients only | Denied |
| Account types & listings | Read | Denied | Read |
| Write, export, or bulk download | Denied | Denied | Denied |
Illustrative configuration. Your firm’s matrix is set during onboarding and reviewed with the CCO. The full summary, including how denials are surfaced, is in the docs. There is no write row to grant: the tools are read-only for every agent.
The Audit tier is priced per firm — not per seat — based on firm size and log volume. We are not publishing a rate card while we onboard our first firms, and we would rather tell you that plainly than invent a number. Talk to us and we will quote your firm directly.
Founder note: final Audit tier pricing is pending a founder decision and will be published here when set.
The Scorecard is the honest first step: it shows you — and your CCO — exactly what shape your data is in before anyone talks about tiers. Technical reviewer? The docs include the log format and a reference server you can run today.