Activity

View recent queries handled by your agent.

The Activity page shows the last 100 queries your agent has handled. Use it to debug responses and inspect conversation transcripts.

Reading the activity log

Each row in the table represents one completed query. Columns:

Column	Description
Time	When the query was received (MM-DD HH:MM, local time).
User	owner for queries you made yourself; otherwise the subscriber’s name.
Accessed	Surface the query came from — chat, widget, email, userAPI, a2a, etc. Document-ingestion rows also show a Loaded / Failed pill here (see below).
Chat	The short chat id. Click it to filter the list to that conversation; a grouped run can be expanded to show every turn that shares the id.
Doc ID	(owner only) On document-ingestion rows, the ingested document’s ID, shown as a badge. Use it to locate the document on the Data pages. Blank for normal queries.
Credits	Credits deducted for the interaction (- when nothing was charged).
Judge	The LLM-judge score out of 10 when judging is enabled — amber when the score is below your threshold, err if the judge itself failed.

Use the toolbar above the table to filter by channel, search chat id / message text, or toggle Group by chat to fold each conversation’s turns together.

Document-ingestion rows

When a file, crawl, or connector document is ingested it appears as its own row with Ingest in the Accessed column and a status pill:

• Loaded (green) — the document was ingested successfully. The filename is shown beneath the pill.
• Failed (red) — ingestion failed. The filename is shown beneath, and the failure reason is printed inline so you can see what went wrong without leaving the page.

The Doc ID badge on these rows is the id you can search for on the Data pages to inspect or remove the stored document.

Inspecting a query

Click the Details button on any row to open the interaction detail page. It shows:

• Meta strip — the interaction ID (GUID), timestamp, how long the query took to complete, and the credits it cost.
• User — the original user message.
• Answer — the agent’s final response.
• Knowledge Used — the retrieval trace for the query: what was searched, and the passages that were actually passed to the agent.

Click Back to return to the activity list.

Message content (query and response) loads asynchronously when you open the detail page.

Charges breakdown

When itemised charges are available, a Charges panel lists every credit line that made up the interaction — the answer, knowledge retrieval, judging, and any platform operations — each with its own amount, followed by a Total. The Answer and Knowledge Used sections also carry their own Charge: N credits line inline. All amounts are in credits.

Tools unavailable

If any tools failed to load for the run — an MCP server that didn’t connect, or a skill script that errored — a Tools unavailable section lists each one with its error, so a dropped tool is never silently invisible. The section only appears when something was actually dropped.

What subscribers see

Subscribers get the same detail page for their own interactions, with two differences:

• The retrieval section is titled “Your data passed to the agent” rather than “Knowledge Used”, and owner-only retrieval stages are hidden.
• The Charges breakdown, the per-section charge lines, and the Tools unavailable section are owner-only — subscribers don’t see them.

If you’re a platform admin, an extra ⓘ link appears next to Details on the owner activity list. It opens the full internal trace for that interaction.

In practice

Debug a reported bad answer. A user says the agent gave a wrong response. Find their row by User + Time, click Details, and read the exact message and the agent’s reply — precisely what happened, no guessing. Then re-run the user’s question on Retrieval Test to see whether the agent retrieved the right context.

Spot an unexpected traffic source. If the Accessed column shows queries from a surface you didn’t expect (widget on a page you thought wasn’t live, or userAPI), you’ve found something to investigate — a widget left embedded somewhere, or an API key in use.

Manage via the Management MCP

This area is managed through the Management MCP at https://manage.agentheya.com/mcp.

Endpoint	https://manage.agentheya.com/mcp
Tools	get_account_status, list_activity

All tools take owner_id + agent_id as required arguments. See the Management MCP guide for authentication and the full tool catalogue.