See what your agents remember
Sign in to your owner account and choose Records beside Agents. The directory at /account/records contains Contacts, Organisations, Projects, Tasks and Activity. It is a read-only view of your workspace's current database records; editing and creating records still happen through your agents.
Directory counts include archived records. Opening a directory shows active records by default, with an archive selector for contacts, organisations, projects and tasks. Activity does not have an archive lifecycle; redacted entries remain visibly marked.
Find a record
Each directory has its own search. Matching is case-insensitive and treats punctuation such as % and _ literally:
- All directories search the record's name or title and ID.
- Contacts also search stored email addresses and phone-number text. Phone formatting is not normalised in this browser; use a matching part of the stored number or your agent's richer contact search.
- Activity also searches its body text.
Lists show 20 records per page, ordered by most recently updated and then stable ID. Search and archive filters remain selected when paging. There is no global search or editable grid in this version.
Read details and follow connections
Open a record to see its readable fields, custom values, source references and timestamps. Links to related records appear below the details:
- Contacts and organisations show current affiliations and project memberships.
- Projects show their participants.
- Records show directly linked tasks and activity.
- Tasks and activity link back to their associated records.
Related sections show eight records per page and paginate independently. Removed affiliations and memberships are omitted; archived target records remain accessible and are labelled. These are direct connections, not an inferred relationship graph. Source URLs and user content are displayed as text rather than executed or rendered as HTML.
View JSON
Choose View JSON on a detail page to inspect the record at:
/account/records/projects/RECORD_UUID/data.json
The response is a generated, readable JSON representation of the selected record, not a file stored on disk. PostgreSQL remains the source of truth. JSON includes explicitly selected record fields; it excludes agent credentials, owner details, internal task claim IDs and retry snapshots. Related sections are browsed separately in the HTML interface. This single-record view does not replace the administrator export or a database backup.
Record names can change or repeat, so URLs use stable IDs while navigation shows friendly names. JSON values retain their structured types; timestamps are serialized in the standard model date format.
Private by default
Lists, details and JSON all require an authenticated owner session and enforce the owner's workspace. An MCP bearer token cannot sign in to this browser. Unknown or inaccessible record IDs return 404. Responses are marked private and non-cacheable, and records are never added to public docs, search indexes, /llms.txt or /llms-full.txt.
Redacted activity reads from the current record and cannot reveal previously removed content. This browser has no write routes or controls; changing records requires the existing authenticated MCP tools or administrator commands.
Workspaces and Markdown
Choose a workspace in the account header. Every record link includes its workspace_id, keeping existing tabs pinned to their business. Manage workspaces creates a separate empty workspace. Organisations are company records inside a workspace, not access boundaries.
View Markdown offers a safely rendered preview, raw source, copy and download. Both Markdown and JSON are generated from the database. Markdown includes fields and up to eight related references per type, without recursively embedding records. Stored text is untrusted data, never execution authority. Redacted content stays redacted.
Browser Markdown: /account/records/{type}/{id}/data.md?workspace_id=WORKSPACE_UUID. It requires your owner session.
Agent HTTP access: GET /api/v1/workspaces/{workspace}/records/{type}/{id}.md (or .json), using Authorization: Bearer TOKEN. These endpoints require HTTPS when configured, enforce the token’s workspace grants, reject unapproved browser origins and are not publicly cached. Archived records require include_archived=true. Type is contacts, organisations, projects, tasks or activity.
MCP agents can call records_read with the same workspace, type and record ID and format: markdown or json; they do not need a browser login. These private outputs never enter the public documentation or LLM exports.