User Management
Reviewing detections and watching accounts
The AI advisory also lists linked accounts to review separately from accounts with an AI abuse judgment. These candidates share an exact signup or login IP with a flagged account in the recent observation window. Each entry links to the account and explains which signup or login observation matched. Connections are one hop only, common addresses are discounted, and coverage warnings appear when a scan is capped or fails. A connection does not give the candidate its neighbour's AI verdict or probability.
In a user's Overview, Signup classification explains the Users table's Suspected label. It shows the matched signal names, plain-language explanations, classification rule, evaluation time and selected signup-chart window. Use the range selector to match the Users chart you are investigating. This is the chart's current heuristic classification, not a historical saved verdict or the AI advisory. Missing classification and incomplete scans are shown explicitly.
In Risk evidence, Observed means the source returned data. It is not a safe-account judgment. In Sign-ins, matching JA4 values are described as TLS fingerprint matches; they do not prove the same device or operator.
The AI assessment can show an account profile with signup-method and network flags, email-shape signals, time to two-factor setup and first spend, model spending shares, and counts of other accounts sharing recorded traits. Missing profiles or linked-account data remain unknown. Shared network, JA4 or mail-domain counts do not establish common ownership.
Factors identified by the model lists up to three fixed evidence statements and the model's confidence that each statement describes the input. These percentages are not each factor's contribution to the overall verdict. A missing factor list does not clear the account; older assessments and unknown results may have none.
In Admin > Detections, start with Review evidence. Inspect the linked accounts and their Activity before recording a judgment. A suggested response describes the detector's recommendation; recording a review does not apply that response to account access. The How this page works guide combines these steps with the shadow-mode and verdict explanations. You can dismiss it and restore it from the button with the same name.
Use Group detections by account to review related rows together. Selecting rows opens a bulk verdict action; the dialog can also grade other loaded, unreviewed detections for the same single account. Group counts distinguish the current page from all loaded rows. Save and next follows the displayed order. Revision conflicts and failed writes remain visible, and confirmed successful writes are retained when a later request fails. A verdict and its history show the reviewing admin when available.
If you need more time, choose Watch or Watch campaign on a campaign detection (or Watch these accounts on the dashboard's AI advisory) and give the watch a name and reason. The watch saves the exact group you selected. You can find open watches on the admin dashboard and in Detections even after the AI advisory expires, changes verdict, or is reviewed. Groups with omitted participants cannot be saved as a watch; open an individual detection with a complete group instead. A capped activity scan can still be watched when its selected participant list is complete.
Open a watch to see its owner, current account status and credits, profile and Activity links, and the last manual check. Choose Refresh account data to reload the details, Mark checked after reviewing activity, or Assign to me to take ownership. Close the watch when monitoring is complete; the Closed list lets you reopen it later.
Watches are manual: they do not send automatic alerts, run an AI assessment, change a verdict, or suspend an account. Membership does not expand when new detections arrive. Deleted accounts are removed, and a deleted owner's assignment is cleared. Watches remain until explicitly closed and their records do not expire automatically.
Watches are available only to full administrators, because a watch shows each member's credits and suspension state. Delegated permissions such as support:read or users:moderate do not open, create or change a watch. Every change requires an audit reason.
The Users tab is the primary interface for administering user accounts on the Bike4Mind platform. It is a single server-driven table with search, filters, built-in views, column sorting, grouping by email domain, and CSV export. The header shows the total number of accounts and how many were active this week.
The view, filters, sort, grouping, page, and page size are kept in the URL, so reloading the page or sharing the link restores the same list. The search text is not put in the URL, because it often holds an email address.
Coordinated API completions
The Detections filter Coordinated API completions surfaces repeated ordered groups in retained successful API-completion settlements. The hourly observation job reads the preceding 24 hours, with bounded results and processing. It can discover accounts outside an existing flagged group. The evidence explains the number of participants, repeated rounds, and incomplete coverage when a cap is reached.
Start with Review evidence, open each participant's Activity, and use Watch for follow-up. Repeated ordering supports coordination; team schedulers can produce it too. Settlement order is not request start order, and this source does not cover failed requests or all API routes. The observation alone does not establish abuse or a shared operator. These detections cannot trigger automatic containment or change automatic risk-review approval. Human review remains available. Their summary shows Review only with no progress bar toward automatic enforcement.
The same ordered group updates its unreviewed observation within a UTC activity day. Reviewed observations remain unchanged; a later activity day can produce a new observation. Use a watch to keep the exact group accessible across observations.
Synchronized API calls
Choose Synchronized API calls in Detections to review accounts with repeated closely timed API-key requests and shared requested-model changes. The scan searches the platform's retained request metadata rather than starting only from existing flagged accounts. Repeated subsets can be found even when unrelated requests overlap. These observations support coordination, not proof of a shared person, matching prompts or misuse; legitimate team schedulers can look similar.
Open Review evidence to see participant counts, matched bursts, requested-model switches, completed calls, daily-cap refusals and unknown or interrupted outcomes. Cohort spending is a snapshot for the scan's 24-hour API-completion window, separate from requests in matched bursts. Grant-funded credits include rewards and other grants; missing attribution remains unknown. Capped spending and unavailable data are explicit.
In a profile's API keys > Recent API requests, you can inspect the correlation id, requested model and completion outcome for newly captured requests. HTTP transport status is separate from completion outcome: HTTP 200 does not prove a successful stream. No response recorded means the request start was captured without a terminal response; it does not establish an active request or a failure. Older records show Not captured for missing completion metadata.
Use Watch campaign for follow-up. If your review supports stopping additional grants, Review blocking future grants opens the existing manual dry run for the exact selected accounts. The preview and an audit reason are required before confirmation. This blocks further signup grants and pauses queued rewards; already-paid credits and account access remain unchanged. Watching, grading or AI assessment does not perform this action automatically.
Search
The search box at the top of the Users tab finds users by name, username, email, user ID, or LLM provider identifier (see Tracing an LLM provider safety notice to an account). Search is debounced to avoid excessive API calls while typing, and it returns to page 1. Press / anywhere in the admin panel to focus it.
Resolving a provider safety notice
AI providers receive a 64-character lowercase hex identifier for each end user instead of the user's ID or email (OpenAI safety_identifier, Anthropic metadata.user_id, and the user field on image-generation requests). When a provider sends a safety or abuse notice that names one of these identifiers, you can resolve it to an account:
- Open the Users tab in the admin panel.
- Paste the full 64-character value into the search box. Upper or lower case both work.
- The matching account appears in the list. Open it to review the profile and activity.
The same lookup is available as GET /api/admin/users/by-provider-id?id=<64 hex>. For an account's own identifier it returns kind: "user" with the account's ID, name, username, and email. The Users tab needs full admin access; the API also accepts a delegate holding support:read (see Delegated permissions). It responds with 400 if the value is not 64 hex characters and 404 if no account matches.
End users of a platform key
A key granted the ai:end-user-id scope serves the end users of another product and names each one with end_user_id (see End users of your product). The provider then receives a separate identifier for each of those end users. Bike4Mind records each one the first time it is used, so you can resolve it the same way:
- In the Users tab search, the value finds the account that owns the key.
- The API returns
kind: "platformEndUser"withaccount(the key owner's ID, name, username, and email),endUserId(the product's own identifier for its end user),apiKeyId(the key most recently used for that end user),firstSeenAt, andlastSeenAt. Ask the account owner about the end user it names; Bike4Mind holds no other details about that person. end_user_idon text-to-speech requests is validated but not sent to any provider, so an end user who only uses text-to-speech has no lookup record and the lookup returns not found for them.
Granting the platform scope
A full administrator grants ai:end-user-id in one of two ways:
- When generating a key for a user, tick AI: Platform End Users under Platform key, alongside the AI scopes the key needs.
- On an existing key, call
POST /api/admin/user-api-keys/<key id>/end-user-idwith{ "enabled": true, "reason": "..." }. Send"enabled": falseto withdraw it. The change is recorded in the audit log. A key holding a dedicated-flow scope (embed, ingest) cannot take it.
If nothing matches, the account may have been deleted, or the identifier may come from a different deployment. Check that you pasted the whole value without extra characters.
Views
The tabs above the table are built-in views. Each one clears the filters and applies its own:
| View | Shows |
|---|---|
| All (default) | Every user |
| New this week | Users who signed up in the last 7 days |
| High risk | Users whose signup risk assessment is high |
| Suspended | Banned users |
Each tab shows its count. If you set filters by hand that match a view exactly, that view's tab is highlighted.
Saved views
To keep a combination of filters, sort, page size and grouping, click Save view and give it a name. The view appears as its own tab after the built-in ones and is stored on the server, so it follows you to any device. Tick Share with all admins to show it to every admin; only you can change or delete it.
Each saved tab you own has a ... menu with Rename, Share with all admins (or Make personal) and Delete. Opening a saved view always starts at page 1. You can keep up to 50 views.
Signup risk review
Open Risk review in the admin sidebar to review accounts flagged by signup risk scoring. Accounts whose signup credits are held by the signup grant risk policy are listed too, marked Credits held. AI access is automatically held while a review is pending, including free models and AI requests through API keys. These users can still sign in and access their existing data. The AI refusal message directs them to contact support to request a review.
Choose Approve for a legitimate account to lift its signup review hold. Approval does not remove a separate manual AI restriction or suspension. Undo, available for 30 seconds, restores the review hold. Approving an account marked Credits held also pays its held signup credits; Undo does not take them back. Later rescores update the evidence but never clear a pending hold or overturn an approval. Choose Suspend for an abusive account to block account access.
Existing pending reviews also hold AI access. If an approved user still cannot use AI, check for separate restrictions, suspension, and normal plan or credit limits. Signup-domain credit holds are independent of account risk reviews.
Columns
| Column | Details |
|---|---|
| User | Avatar, name (or username or email), and email |
| Status | Suspended (banned), Restricted (moderated), New (recently created), or Active |
| Plan | User level |
| Signed up | Account creation date |
| Last active | Most recent activity |
| Credits | Current credit balance |
| 30d spend | Credits spent in the last 30 days |
| API keys | Number of live API keys |
| Risk | Signup risk band: High, Medium, or Low |
Click a row to open the user's profile page. The row's ... menu has Open profile and Copy user ID; using it does not open the row.
Selecting rows
Use the checkboxes to select rows. Shift-click selects a range, and x toggles the focused row. Changing a filter, the view, or the search clears the selection; moving between pages keeps it. Once every row on the page is selected, Select all N matching extends the selection to every user that matches the current filters.
Bulk actions
While rows are selected, a bar under the table offers these actions (on a narrow screen they sit in one Actions menu):
| Action | Effect |
|---|---|
| Suspend | Blocks sign-in, ends live sessions and disables their API keys |
| Restrict | Turns off AI features; users can still sign in and reach their data |
| Adjust credits | Sets, adds or removes credits; every change is written to the credit ledger |
| Revoke API keys | Disables every active API key the users own |
| Add tag | Adds a tag to users who do not have it yet |
| Contain | Suspends the users, sets their credits to 0 and revokes their API keys, together for each user. Optionally also blocks new signups from their email domains |
Each action opens a dialog with a dry run: how many users will change, who is excluded (you and other admins are never included), and how many API keys and credits are affected. A reason is required and is recorded in the audit log for every user.
- A batch is limited to 500 users. If more match, narrow the filters first.
- Block email domains (Contain only) adds the users' email domains to the blocked email domains setting before any user is changed. The dry run lists every domain it would block and the ones already blocked. Well-known public mail domains such as gmail.com are skipped, but the list is not exhaustive, so check the domains before you confirm. It needs permission to edit admin settings.
- If the matching users change between the dry run and the confirm, nothing is applied and the dialog shows the new count to confirm again.
- After a bulk action, an Undo button is shown for 30 seconds. It restores the whole batch and unblocks the email domains the batch added (domains that were already blocked stay blocked). Sessions that were ended stay ended.
Recent activity
The Recent activity button shows a panel of recent user activity across the platform.
User Profile Page
Each user has a full-page profile at /admin/users/<userId>, opened by clicking the user's row in the Users tab or Open profile in the row menu. It is admin-only, like the rest of the admin panel. The Users button at the top returns to the Users tab.
The header shows the user's name, email, avatar, account status, user level, organization, signup method, and created and last-active dates. Adjust credits and Suspend are quick actions; More contains the other account actions.
To edit the user's name or email, open More > Edit profile. The username is displayed read-only. Change the fields in the modal and click Save Changes. Wait for the success notification before closing the modal; the profile header refreshes after a successful save. If saving fails, keep the modal open, correct any validation error, and retry.
The page has six tabs. The selected tab is kept in the URL, for example ?tab=activity or ?tab=admin (Overview is the default and has no parameter), so a link can open a specific tab.
| Tab | Contents |
|---|---|
| Overview | Account details, access, subscription, recent activity, and internal notes |
| Activity | The user's recent activity, including recorded API requests |
| Credits | Credit ledger and spending |
| API keys | The user's API keys and recent recorded requests |
| Sign-ins | Sign-in history and sessions |
| Admin log | Administrative actions recorded for the user |
Use More for role, product-access, subscription, compliance, restriction, and deletion actions (see Admin Actions).
Old links of the form /admin?userId=<userId> redirect to the user's profile page. An unknown or malformed user id shows User not found; any other loading error shows a retry button.
Recorded API requests
Open API keys > Recent API requests to inspect each recorded call's time, method, endpoint, response status, latency, IP and key. Filter by key and UTC date range, use Load more requests for older records, or Refresh to check for new activity. All keys also includes retained requests from deleted keys. Endpoint query strings and fragments are omitted from this view.
The API filter in Activity shows these same requests alongside the other timeline sources. Adjacent matching calls within a minute are grouped; the count covers loaded activity. Click a request to open API keys for individual records. A failed response is marked for attention; it does not establish abuse.
Records are retained for 90 days and cover only recorded API-key requests. An empty result does not establish that no requests were attempted. If a read fails, use the retry action rather than treating the error as inactivity. Check the selected key, UTC date range and retention window when a request is missing.
Risk flags
Under the profile statistics, named risk badges show activity that needs a closer look. Click a badge to see its rule, the exact UTC window, and the counts or values behind it. There is no combined risk score, and a flag does not automatically restrict the account.
| Badge | Rule |
|---|---|
| Repeated failed sign-ins | At least 5 failed identity checks or blocked sessions in 24 hours. Failed passwords are not included in this audit source. |
| New IP or device | A successful sign-in in the last 7 days uses an IP or browser and operating system combination absent from earlier retained successful sign-ins. At least 3 earlier successes are required. |
| Credit spend spike | Spending over the current UTC day and 6 previous days reaches 3 times the preceding 30-day average rate. The baseline needs at least 7 active days and 100 credits, and recent spend must reach 200 credits. |
| Prior moderation | At least 1 image moderation incident in 30 days, excluding incidents reviewed as false positives. |
For sign-in flags, View in timeline opens the Activity tab filtered to sign-ins and the UTC calendar days covering the evidence window. View credits opens the Credits tab for a spend spike. Open compliance opens the moderation history. Sign-in comparisons use retained audit history, so they cannot establish whether an address or device appeared before that history.
No risk flags means none of these rules fired. It does not establish that the account is safe or that every kind of sign-in failure was recorded. If the flags cannot be loaded, use Try again; a read failure is shown separately from an empty result. Access requires an administrator session; API keys cannot read the IP evidence.
Internal notes and profile tags
On a user's profile, Overview > Internal notes stores plain-text support context.
Reading requires support:read; adding, editing, and deleting require
users:moderate. Full administrators have both permissions (see
Delegated Admin Permissions). Notes show their
original author and time, and View history shows attributed revisions.
Deletion hides a note while retaining its history; the Undo notification restores
it. Legacy inline notes remain visible but read-only. Impersonated sessions cannot
write notes.
In Access > Tags > Edit, add or remove free-form tags with suggestions from
existing tags. Saving requires settings:manage. Tags are limited to 100 distinct
values of 1 to 64 characters. Role and product-access grants use their dedicated
actions and cannot be changed through the free-form tag editor. Successful saves
refresh the user list and tag filters. Activity and Admin log record note actions
without note bodies, and tag actions with before/after values.
If the notes card is missing, check the signed-in administrator's support:read
permission. Use Reload notes or Reload history after a read failure. A stale
note version or tag snapshot is rejected: reload the profile and reapply the change
to the current state. If a tag is too long or duplicates another tag, correct it
before saving.
Filtering
Click Filter to add a filter chip. Each chip shows its current value; click it to change the value, or remove it. Clear filters removes them all.
| Filter | Values |
|---|---|
| Status | Active, Suspended, Restricted |
| Plan | DemoUser, PaidUser, VIPUser, ManagerUser, AdminUser |
| Risk | High, Medium, Low (users with no risk assessment match none of these) |
| Signed up | Last hour, day, week, or month |
| Sign-in method | Email, Google, GitHub, Facebook, Okta, SAML |
| Email verified | Verified, Not verified |
| API keys | Has keys, No keys |
| Organization | One or more organizations, or No organization |
| Tags | One or more tags, including Admin for super admins |
The Signed up column shows the first provider linked to the account, or Email when none is linked. The Sign-in method filter matches any linked provider, and Email matches only accounts with no linked provider. An account created with an email code that later linked Google shows as Google, and appears under the Google filter but not Email.
Filters are applied on the server, so they cover every user, not just the current page. Changing a filter returns to page 1.
Grouping by email domain
Set Group by to Email domain to list users under a header per email domain, with the user count for each. Show only this domain on a header filters the list to that domain (subdomains included).
Sorting
Click the User, Signed up, Last active, or Credits column header to sort by it; click again to reverse the order. The default is Signed up, newest first. Changing the sort returns to page 1.
Pagination
The footer below the table shows the range of rows on the page (for example "1 to 25 of 1,234"), a Rows per page selector (25, 50, or 100; 25 is the default), and previous and next page buttons.
Creating a User
Click the Create User button to open the Create User modal. The form is organized into three sections:
Basic Information
| Field | Required | Description |
|---|---|---|
| Username | Yes | Unique username for login |
| Yes | Must contain @ | |
| Full Name | Yes | Display name |
User Settings
| Field | Default | Description |
|---|---|---|
| User Level | DemoUser | One of: DemoUser, PaidUser, VIPUser, ManagerUser, AdminUser |
| Admin Privileges | Unchecked | Grants admin access; automatically linked to the "Admin" tag |
| User Tags | Required (at least 1) | Determines which models are available to the user. For external users, the "Customer" tag is recommended |
Resources
| Field | Default | Description |
|---|---|---|
| Initial Credits | 10,000 | Starting credit balance |
| Storage Limit | 1,000 MB | Maximum storage allocation |
The form requires all basic fields to be filled and at least one tag selected before submission is allowed.
Editing a User
In Full view mode, user fields can be edited inline. Changed fields are highlighted, and an Update button appears in the user card header when changes are detected. Changes are saved via the API when the Update button is clicked.
Available editable fields include:
- Name, username, email
- Admin status and user level
- User tags (predefined and custom)
- Ban and moderation status
- Storage limit and credits
- Subscription details
Login as User
The user profile page header has a Login as User button that allows an admin to impersonate the selected user for debugging purposes.
Send System Message
The Send Message button opens a modal to send a system message directly to the user.
User Authentication
Bike4Mind uses passwordless email authentication. When a user logs in, a one-time code (OTC) is sent to their registered email address — no password is required. Ensure the user's email address on file is correct and that the email service is functioning (see Admin → General Ops → System Health).
User Permissions
The Roles section on the profile page's Overview tab provides:
| Control | Description |
|---|---|
| User Level | Button that cycles through levels: DemoUser, PaidUser, VIPUser, ManagerUser, AdminUser. Color-coded by level. |
| Super Admin | Checkbox to grant or revoke admin privileges |
| Predefined Tags | Checkboxes for system-defined tags |
| Custom Tags | Text input to add comma-separated custom tags, with remove buttons for each custom tag |
To give a user part of the admin area without making them a Super Admin, edit their Admin permissions on the Access card. See Delegated Admin Permissions.
Admin Actions (Ban, Moderate, Delete)
Destructive user operations live in the Danger zone on the Admin tab of the user profile page:
| Action | Description |
|---|---|
| Ban | Checkbox toggle. Bans the user without deleting their account. |
| Moderate | Checkbox toggle. Moderated users cannot use AI features. |
| Delete | Requires typing "DELETE" in a text field, then clicking the Delete button. A confirmation modal appears before the deletion is executed. This action cannot be undone. Deleting from the profile page returns you to the Users tab. |
| Compliance | Opens the user's compliance details. |
Ban and moderation changes are staged like any other field. When you click Update with a ban or moderation change pending, a confirmation dialog names the change before anything is saved; Cancel saves nothing.
Export
The Export CSV button exports all users matching the current search and filter criteria as a CSV file named users.csv. The CSV includes the following columns:
| Column | Description |
|---|---|
| Name | User's display name |
| Organization | Organization name |
| Username | Login username |
| Email address | |
| Logins | Total number of logins |
| Last Login | Timestamp of most recent login |
| Created At | Account creation timestamp |
| Is Admin | Yes/No |
| Is Banned | Yes/No |
The export includes all matching users, not just the current page, up to 15,000 rows. If the export hits that cap, a warning asks you to narrow the filters to export the rest.
Best Practices
- Use the list for quick scanning and the user's profile page for detailed administration.
- When creating external users, assign the Customer tag to provide appropriate model access.
- Prefer banning over deleting users to preserve audit trails. Deletion is permanent.
- Always confirm the environment banner color before performing destructive actions like user deletion.
- Use Export CSV to maintain offline records of user lists before making bulk changes.
- Use organization and tag filters together to narrow down to specific user cohorts.
- Review user permissions regularly and grant only the minimum necessary access.
- Notify users before making major changes to their accounts.
Common Tasks
Upgrading a User
- Open the user's profile page by clicking their row.
- Click the User Level button to cycle through levels (DemoUser, PaidUser, VIPUser, ManagerUser, AdminUser).
- Alternatively, add specific permission tags for fine-grained access control.
- Click Update to save changes.
Tracing an LLM provider safety notice to an account
Requests to Anthropic and OpenAI carry an opaque per-user identifier: a 64-character hex SHA-256 of the user ID, sent as Anthropic metadata.user_id and OpenAI safety_identifier. When a provider sends a safety notice that quotes this identifier:
- Press Cmd/Ctrl+K on the admin page to open Admin commands. Full admins only.
- Paste the 64-character identifier. The matching account is listed with its masked email and account status:
active,throttled,suspend_pending(the suspension threshold was reached and is waiting for an admin to confirm),suspended(also shown for an account suspended from its profile) ordeleted. - Select it to open the user profile, where you see the full email and can act on the account.
Each lookup in Admin commands is recorded in the admin audit log without the identifier itself. You can also paste the identifier into the Users tab search. No result means the identifier does not belong to a user on this deployment.
Handling Support Requests
- Use Login as User to reproduce the issue in the user's environment.
- Check the user's activity, login history, and account status for clues.
- Document findings and resolution steps.
- Use Send Message to communicate the resolution to the user.
Managing Subscriptions
- Use Grant Subscription in the User Details section for manual extensions.
- Check both Individual and Team subscription status in the Subscription section.
- Monitor credit usage for unusual patterns via the Credits field.
- Manually adjust credits if necessary.
Troubleshooting
User Cannot Log In
- Check if the account is banned or moderated in the Admin Actions section.
- Verify the email address is correct — the one-time code is sent to the email on file.
- If the Email field is blank, the account has no email on file and cannot receive a one-time code. This is a supported account shape: an account created through a social/OAuth provider that did not assert the email as verified is stored without one. That user signs in through the same provider they signed up with, not by email code. To move them onto email login, have them add an address from Profile > Email Address > Add Email and confirm the verification link.
- Confirm the email service is working via Admin → General Ops → System Health.
- Check the user's login history for suspicious activity.
Missing Features
- Verify the User Level is appropriate for the features they need.
- Check organization membership — some features require org association.
- Review user tags for required permissions (e.g., "Developer" tag for API access).
Credit Issues
- Check the current credit balance in the B4M Settings section.
- Review the credit purchase date for recent activity.
- Manually adjust credits if a billing discrepancy is confirmed.
Security Notes
- All administrative actions are logged for audit purposes.
- Login uses one-time codes delivered to the user's registered email — codes are time-limited and single-use.
- Login history tracks device and location information to help identify suspicious activity.
- Regular security audits of user permissions are recommended.
Related Articles
- Admin Dashboard Overview - Overall admin panel navigation
- Email Verification - Managing email verification status
- User Migration - Bulk user migration
- Delegated Admin Permissions - Permission model for delegated admins
- Analytics - User activity analytics