Skip to main content

Agent History

Agent History is where you review everything your agents have actually done — full conversation transcripts, every tool call with its inputs and outputs, and success evaluations. Use it to measure quality version by version, debug unexpected behavior, and export or sync conversations for downstream analysis.

What it's for​

  • Quality evaluation — per-month stats (total conversations, evaluated, successful, success rate) and a version-by-version performance breakdown tell you whether your latest published version is better than the last.
  • Debugging — the conversation viewer shows exactly what the user said, what the agent replied, and every tool request/response pair, so you can pinpoint where a run went wrong.
  • Downstream workflows — export history as JSON or CSV, or configure Conversation Sync to land transcripts in an AutessaDB table automatically.

From the dashboard sidebar, open Agentic Workflows → Agent History. Pick an agent in the Select Agent dropdown (owned, shared, and company agents are all listed), set a Start Date / End Date range (defaults to the current month), and optionally tick Include My Sandbox Conversations to see your own test runs alongside real traffic.

Key concepts​

Stats and version breakdown​

The This Month's Stats dashboard summarizes the selected range: total conversations, how many were evaluated, how many succeeded (including a "successful with caveats" bucket), and the overall success rate. The collapsible Version Performance Breakdown splits those numbers per published version (and the draft), with a Browse shortcut to open that agent version — so a regression after a publish is immediately visible. See Versions.

Evaluations​

Each conversation row carries an evaluation status: successful, successful with caveats, unsuccessful, or not yet evaluated. Evaluations come from two places — automated evaluation (scored on dimensions like groundedness, task completeness, argument faithfulness, and efficiency) and human evaluation, which you record in the viewer's Evaluation Details panel with statuses like Successful, Successful (Not Efficient), Successful (Not All Steps), or Unsuccessful, plus notes. A human evaluation always takes precedence over the automated one.

The conversation viewer​

Clicking View Details on a row opens the conversation viewer. It shows:

  • An overview card — conversation ID, close date, agent version, API key, message and tool counts.
  • The full transcript — user messages (including audio transcriptions and attached-image markers) and agent replies in order.
  • Tool call cards — each tool request and its response grouped into one expandable card with the tool's input variables, the result (with a raw/JSON pretty-print toggle), and debug logs.
  • For Critic agents, each rejected attempt with the critic's reasoning and failed criteria, followed by the approved final message.
  • A toggleable Evaluation Details side panel for reviewing or recording evaluations.

Conversation Sync​

The Sync with AutessaDB button opens Conversation Sync, where you define configurations that automatically write closed conversations into AutessaDB tables. Each configuration filters by agent (or all agents, optionally including sandbox conversations) and maps conversations onto one or more tables: a match column holding the conversation ID, a column receiving the conversation JSON (with or without tool calls — including them can significantly increase stored size), an insert-or-skip behavior when no matching row exists, and any additional columns filled from sync variables such as @@{conversationId}, @@{closeDate}, @@{agentId}, @@{agentVersion}, @@{apiKeyId}, @@{conversation}, and @@{shortened_conversation}.

Because synced conversations land in a normal AutessaDB table, everything downstream is available: triggers can process each new transcript, knowledge bases can make them searchable, and scheduled jobs can report on them.

How to review a conversation​

  1. Open Agentic Workflows → Agent History and select the agent.
  2. Adjust the date range if needed, then find the conversation row — the icon shows its evaluation status at a glance.
  3. Click View Details to open the transcript. Expand any tool card to inspect its exact inputs and outputs.
  4. Optionally open Show Evaluation to record a human evaluation and notes.

How to export history​

  1. On the Agent History page, open the export dialog.
  2. Pick a time range (last one, two, or three months).
  3. Choose JSON for full conversation details including logs and evaluations, or CSV for a summary table (date, version, message and tool counts, API key, conversation ID). The file downloads immediately.

How to sync conversations to AutessaDB​

  1. Click Sync with AutessaDB on the Agent History page.
  2. Click Create New Configuration, name it, and choose which agents it covers (leave the filter empty for all agents).
  3. Add a Table Update: pick the target table, the match column for the conversation ID (a SHORT_TEXT or LONG_TEXT column), and the column to receive the conversation JSON.
  4. Decide whether unmatched conversations should insert new rows (upsert) or be skipped, map any additional columns using the sync variables, and save. New conversations sync automatically while the configuration is enabled.