Docs · Approvals and audit
Approvals and the audit trail, end to end.
How changes from AI apps and people are approved, who is told, and how every change in a workspace is recorded in a trail nobody can edit.
Two approval layers Every change recorded Nobody can edit entries
Two layers of approval
The AI app asks you. Then Sprint Track asks an approver.
Both layers apply to ChatGPT, Claude, and every other MCP client.
In the AI app
ChatGPT, Claude, and other MCP clients ask you before they use a tool that changes something, according to the app's own settings.
In Sprint Track
The change then waits for an approver in the workspace, unless an administrator has turned on auto-approval of routine edits for that app. Sensitive and destructive changes always wait.
What waits for approval
Reads run. Edits wait, unless you say otherwise.
Reads
Run straight away, within your role and the access granted to the app.
Routine edits
Creating and updating work items, comments, cycles, labels, time logs, and similar records wait for an approver, unless the app's routine edits are auto-approved.
Sensitive changes
Changing membership, rates, sharing, or project scope, running agents and imports, and GitHub writes always wait for an approver.
Destructive changes
Deleting or revoking anything always waits, and needs an Owner or Admin among the approvers.
Your own requests
If you are an approver, you can approve a change your own AI app asked for, and it runs as soon as you do. No AI app can approve anything.
Settings for each AI app
Decide what each app may do in your workspace.
On or off
A turned-off app keeps its connection but can only describe its access until it is turned on again.
Access level
Read only; read and write without deletions; or full access. The level caps every connection to that app, whatever each member granted.
Auto-approve routine edits
Off by default. When on, the app's routine edits run without an approver and are marked as auto-approved in the audit trail. Sensitive and destructive changes still wait.
Who can change them
Members with the permission to manage AI connections. Turning auto-approval on or off also needs the permission to manage approvals. Every member can read the settings.
Where
In Sprint Track, open AI assistants, then Apps & permissions. ChatGPT, Claude, and other clients each have their own row.
For approvers
You are told, and one click opens the request.
Notified
A badge on Approvals in the sidebar counts requests waiting for you. Each request also arrives as a notification that opens it.
By email
Approvers also get an email with a link to the request. Turn these emails off in your profile, under notification settings.
Review
The request shows the exact change, which app and member asked for it, and what it affects. Approve or reject it, with a reason if you like.
Grouped
When an app's routine edits are auto-approved, you get one grouped notice instead of one per edit, linking to the app's activity.
For people asking an AI app
You always know where your request is.
A link back
When a change waits for approval, the app tells you so and gives you the link to the request in Sprint Track.
Ask the app
Ask your AI app what it can do or what is waiting. It reads this through the access_context tool: the workspace, its access, what needs approval, and your pending requests.
Activity
AI assistants, then Activity, shows what AI apps changed for you. Members with audit access see everyone's.
Withdraw
You can withdraw a request that is still waiting.
The audit trail
Every change recorded.
Under Settings, then Audit trail, in every workspace.
Every change
Changes made in the app, by AI apps, through the API, and by agents and automations; every approval, rejection, withdrawal, and expiry; sign-in and sign-out; SSO and SCIM provisioning; and billing and GitHub changes made on the workspace's behalf.
What an entry holds
Who, how (including the AI app), what, the record it touched, the outcome (done, waiting for approval, refused, or failed), and which fields changed. Old and new values are kept for settings and links between records; text such as titles, descriptions, and comments is recorded as changed, without the words.
What it never holds
IP addresses, browser details, or the text people typed into a request.
Retention
90 days on standard packages and 12 months on Enterprise. When entries pass that age they are removed oldest first, and the trail records that it happened.
Who sees what
Everyone sees their own. Auditors see everything.
Your own entries
Every member, on every package, can open Settings, then Audit trail, and see the entries for what they did.
Everyone's entries
On every plan, owners, admins, and anyone else with the audit permission can switch to Everyone, filter by area, outcome, app, and dates, and open any entry. Members with access to specific projects only see their own.
Export
Everyone's entries for up to a year at a time can be exported as CSV, including each entry's position and signature. The export is itself recorded.
Integrity
Append-only and tamper-evident.
Entries cannot be changed through Sprint Track, and a change made any other way is detected.
Append-only
Nobody can edit or delete an entry, including workspace owners. The database refuses edits and early deletion, not just the app.
Signed chain
Each workspace's entries are sealed into a chain within a minute: every entry is signed together with the one before it, with a key the database never holds.
Checked
New entries are checked as they are sealed, and the whole chain is re-checked every day.
Status
The audit trail page shows Verified, Pending seal, or Integrity check failed. A failure stays visible until our team has investigated it.
Connect your apps
Step-by-step guides: ChatGPT, Claude, and Claude Code, Codex, Cursor, VS Code, Gemini CLI, and other MCP clients. See the security overview for how we protect your data.
Stay in control of every change.
Join the waitlist and we will email you when Sprint Track opens.
We will email you when Sprint Track opens.
See pricing