Audit log
The Audit log tab on the Organization page is a record of what happened in your organization, newest first. Use it to answer "who did this, and when?" and to hand a copy to your own reviewers.
Not seeing the tab? The Audit log is turned on per organization, and only people the organization allows can open it. If you think you should have it, ask your organization's owner, or contact support to have it turned on.
What it shows
Each row is one event. It shows the time, who did it, what they did, the project it relates to (if any), how it ended, and a short line of details such as a role, a file name or a count.
The log covers:
- Sign-ins and access changes - sign-ins, failed sign-ins, invitations sent and accepted, role changes, and API keys and AI app connections created or removed.
- Project, document, issue and check activity - for example a project created or shared, a document uploaded or deleted, an issue's status or severity changed, a check run started.
- AI app tool calls - what an AI app connected to your organization asked Structured to do, and whether it succeeded or was refused.
- Add-in actions - which tool the assistant ran for someone in the Revit, AutoCAD, Archicad or other add-ins, with the approval and the result. What the person asked for and the assistant's summary are not included; owners can read them in the add-in's own activity log. No code is shown.
Times are shown in your own time zone. Details are limited to names, counts and controlled words. Free-text notes and reasons people typed are not included.
Structured staff
When someone from Structured staff takes an action in your organization, for example while helping with a support request, the row reads Structured staff. The log does not say which staff member.
The one exception is an add-in action staff took inside a member's own session while helping them: like the add-in's activity log, the row is listed under that member and its details say acted_by_staff. It still does not say which staff member.
Filter
Narrow the list with:
- From and To - whole days, in your time zone.
- Person - one person in your organization.
- Category - Access, Add-ins, AI apps, Checks, Codes, Documents, Issues, Organization or Projects.
- Source - Activity, Sign-ins, AI apps or Add-ins.
Clear filters puts everything back. Load more at the foot of the list shows the next page.
Export
Export CSV and Export JSONL download what the filters currently select, not only the rows on screen. CSV opens in Excel and other spreadsheets. JSONL is one event per line, for loading into other tools.
A file holds up to 50,000 events. If your selection is larger, the file stops at that limit and the page tells you so. Narrow the date range and export again for the rest.
Each export is itself recorded in the log as an event, so there is a record of who took a copy and when, which format, how many events, and the filters it used (date range, person, category and source). If that record cannot be written, the export does not download.
Who can open it
By default, organization owners can open the Audit log. An owner can let other people open it: on the Settings tab of the Organization page, open Permissions, find Who can see the Audit log tab and tick the roles that should have it.
People who cannot open it do not see the tab. If they follow a link to it, the page says to ask an owner for access.
See How access works for how roles and permissions fit together.
Streaming
Streaming sends your audit log, as it happens, to an endpoint you run: a SIEM, a log collector or any HTTPS receiver. Use it when your security team needs events in its own tools without anyone exporting files by hand.
Not seeing the Streaming section? Streaming is turned on per organization, separately from the Audit log tab. Ask your organization's owner, or contact support to have it turned on. If it is on for your organization but you still do not see it, you have not been given permission to set it up (see below).
Who can set it up
By default, organization owners. An owner can let others: on the Settings tab of the Organization page, open Permissions, find Who can set up audit log streaming and tick the roles that should have it. People who are not allowed do not see the section.
Set it up
- Open the Audit log tab on the Organization page and scroll to Stream to your SIEM.
- Enter the Endpoint URL. It must start with
https://, be reachable from the internet, and carry no user name, password or query string: put any credential your receiver needs (such as an API key) in the auth header (step 4), never in the URL. Any port is fine. - Choose a Signing secret of 16 to 256 characters. Structured signs every delivery with it so your receiver can check the delivery is genuine. Keep a copy: after you save, only the last four characters are shown.
- If your receiver needs a credential, enter an Auth header name and value. Leave both empty to send none.
- Select Send test event to check the endpoint before you save. The page shows whether the receiver accepted it, the status it answered with, or that it could not be reached (the address is not public, or a DNS, TLS or network error: the page does not say which).
- Leave Send events on and select Save.
To change the URL or the header later, edit and save again. Leave the secret or the header value empty to keep the saved one, as long as the URL stays the same: if you change the URL, enter both again (a saved secret or header value is only ever sent to the address it was saved for). Send test event follows the same rule. Delete stream stops sending and removes the saved secrets.
The panel under the form shows the newest event delivered, the last successful delivery, how many events have been delivered, and, after failures, how many in a row and when the next retry is.
Example: Splunk HTTP Event Collector
Structured sends one event per line, so point it at the collector's raw endpoint:
- Endpoint URL:
https://your-splunk-host:8088/services/collector/raw - Auth header name:
Authorization - Auth header value:
Splunk <your HEC key>, where the key is the one Splunk generated for the collector
Use a signing secret of your own as well; Splunk ignores the signature headers, but they are there if you put a verifying proxy in front.
What is sent
Each delivery is one HTTPS POST of up to 500 events, oldest first, with Content-Type: application/x-ndjson: one JSON event per line. The events are exactly the ones in the Export JSONL file, so the same fields and the same rules about what is left out apply. A line looks like this (wrapped here for reading; on the wire it is one line):
{"id":"activity:123","occurred_at":"2026-11-07T14:05:00+00:00","source":"activity",
"category":"projects","action":"project.member_invited",
"actor":{"kind":"member","email":"pat@example.com","name":"Pat Example","user_id":"..."},
"project_name":"Harbor Tower","outcome":null,"ip_address":null,"user_agent":null,
"details":{"role":"editor"}}
Every request carries these headers:
| Header | Value |
|---|---|
X-Structured-Timestamp | Unix time in seconds when the delivery was signed. |
X-Structured-Signature | v1= followed by the hex HMAC-SHA256 of <timestamp>. plus the raw body bytes, keyed by your signing secret. |
X-Structured-Delivery | A unique id for this POST. A retry of the same batch gets a new one. |
X-Structured-Event-Count | How many events are in the body. |
User-Agent | Structured-Audit-Stream/1 |
Plus your own auth header, if you set one.
Verify the signature
Check every delivery before you trust it. Sign the timestamp, a dot and the raw body bytes (not a re-serialized copy), compare in constant time, and reject timestamps older than five minutes so a captured delivery cannot be replayed.
from hashlib import sha256
from hmac import compare_digest, new as hmac_new
from time import time
SECRET = b"your signing secret"
MAX_AGE_SECONDS = 300
def is_genuine(headers: dict, raw_body: bytes) -> bool:
try:
timestamp = int(headers["X-Structured-Timestamp"])
signature = headers["X-Structured-Signature"]
except (KeyError, ValueError):
return False
if abs(time() - timestamp) > MAX_AGE_SECONDS:
return False
signed = str(timestamp).encode() + b"." + raw_body
expected = "v1=" + hmac_new(SECRET, signed, sha256).hexdigest()
return compare_digest(expected, signature)
Answer with any 2xx status once you have stored the batch. Anything else counts as a failure.
Delivery guarantees
- At least once. If a delivery fails, or your receiver stored it but answered too slowly, the same events can arrive again. Every event has a stable
id(for exampleactivity:123); deduplicate on it. - About 10 minutes behind. Only events at least 10 minutes old are sent, so that an event recorded a moment late is never skipped. Expect roughly 10 to 11 minutes between something happening and it arriving.
- No history. Streaming starts from the moment you first turn it on. Use Export for anything earlier.
- In order, oldest first, within and across batches.
Failures and pausing
A delivery fails when your receiver answers anything other than 2xx, does not answer in time, or cannot be reached. A redirect (3xx) is a failure too: Structured does not follow redirects.
Structured retries the same batch with growing waits, starting at one minute and doubling up to one hour. After 10 failures in a row the stream pauses: the panel says so, nothing is skipped, and the stream keeps its place. Fix the receiver, then select Resume and sending carries on from where it stopped.
The test event
Send test event delivers a single event, signed and headed like a real one, that is not part of your audit log:
{"id":"test:2f6c1d0e-8a41-4c8e-9d52-1b7d3a9c5e10","source":"test","action":"org_audit.stream_test", ...}
It is a good first check that your receiver parses the line, accepts the headers and verifies the signature. Ignore events with "source": "test" in your searches.
Streaming is itself audited
Setting up, changing, deleting or resuming a stream, and an automatic pause, are each recorded in the audit log under Access (org_audit.stream_created, stream_updated, stream_deleted, stream_resumed, stream_paused). Secrets are never written to the log.