Conversation & usage logs

Conversation logs (/conversation-logs) is the operator read plane for what your users actually did: an insights rollup and activation funnel on top, the full list of conversations below, a per-conversation transcript viewer, and a sibling per-event usage ledger.

Insights & activation funnel

The top band summarises what your users asked for and whether it worked. These figures are AI-generated and refreshed by nightly analysis, so recent conversations show as pending until the next run.

• Activation funnel — the four steps from workspace created to first real conversation, read from your usage log.
• Outcome breakdown — how conversations resolved: fulfilled, partial, failed, unsupported, abandoned, or informational.
• Top capability gaps — the requests your catalog does not yet support.

The conversation list

Below the rollup, every conversation is listed with its outcome, environment, end-user, message count, cost, and timing. A conversation that has not been analysed yet shows as pending.

• Filter by outcome, environment, and time range, or search across conversations.
• Open any row to read its full transcript.
• Export the filtered set as CSV, Excel, or text.

The transcript viewer

Opening a conversation shows its message history, turn by turn, including compact tool-call cards that name the tool and its outcome. Personal data is automatically redacted, and your access is itself written to the audit log. You can export a transcript as a PDF.

Usage logs

Usage logs (/usage-logs) is the per-event side of the same story: a read-only ledger of every billable event, filterable by type, user, and time, and exportable as CSV.

• Event types include conversation starts, messages, tool calls, refusals, cost events, catalog uploads, and MCP connections and tool calls.
• Each row carries its cost in USD and metered attributes — never raw message content.
• Rows deep-link to their conversation, and the filtered set exports as CSV, Excel, or text.

Next steps