Chat and results#
Chat accepts requests and supported attachments. Tool selection uses registered names, descriptions and schemas. Results may include text, owner-scoped files, media or stored tables.
- Your request
- Registered tools
- Live Activity
- Result
Live Activity and Stop#
The activity stream shows processing, tool and approval events. Stop/cancel requests act on current session tasks. They cannot guarantee rollback of a vendor operation already submitted.
History, folders and search#
Conversations and visible messages persist in owner-scoped SQLite storage. Organize conversations into folders, move them between folders and search history. Stored history does not restore a live agent session or pending approvals after restart.
Snippets and quick commands#
Reusable prompt commands support owner-scoped search and folders. They are saved prompts, not a verified scheduler or visual automation builder.
Approvals#
Pending cards show tool/module, risk, target and timing information. Preview can show bound inline PowerShell or fetch a library script with the required read permission. Approve/reject operate on the exact session-owned request.
Tables, files and Excel#
Stored tables support paging, filtering, sorting, display and Excel export. Table data comes from the provider or stored file; AI presentation labels remain distinct from authoritative columns/data. Uploaded/result video playback is not surveillance analytics.
Administration#
Administrator pages cover users, roles, permission effects/scopes, module settings/health/diagnostics, MCP server definitions and audit visibility. Personal preferences and personal MCP connections have owner-specific boundaries.
Language, help and certificates#
Per-user English/Ukrainian/Russian preferences, Help Center navigation and certificate help/download are implemented. Codex connection and model preferences are available when that module is configured and enabled.