Skip to main content
Every publish creates a version you can roll back to, and every run is recorded with its full transcript. This page covers both.

​
Versions

Agents built in the Agent Builder are versioned. Each Publish creates the next version (v1, v2, …) with:
  • The agent’s files at that moment
  • Its settings: tools, vault, model, connected knowledge bases and analytic models
  • The Release notes you entered, if any
Open the builder’s Versions tab to see them. The version the agent currently serves is marked Live.

​
Roll back

Click Roll back to this version on any version and confirm. The agent’s files and recorded settings return to that version, and the change goes live for everyone immediately. Some settings cannot always be restored. DataGen rolls back everything else and tells you what it skipped, for example when:
  • The tool bundle, sandbox environment, harness or a connected knowledge base was deleted
  • The tool bundle now belongs to a different vault
  • You no longer have edit access to a knowledge base the version had connected
  • A connected analytic model is no longer published
  • The version used Whole workspace exposure; an agent on Connected only cannot go back to it

​
Delete a version

Click Delete this version to remove a version you no longer need. Deleting cannot be undone.
Agents connected from a GitHub repository are versioned by their Git history: push or revert commits in the repository to change what the agent runs.

​
Agent status

The Agents list shows each agent’s status at a glance: The agent page shows the Enabled or Disabled pill, and its Overview tab lists Recent runs. See Triggers for enabling and disabling.

​
Agent Logs

Agent Logs (in the sidebar’s More menu, and pinnable to the sidebar) lists every conversation and run across your agents. Search by title, prompt, agent, run ID or session. On the Runs tab, select running or queued runs to stop them in bulk.

​
Run statuses

​
Triggers

Each run records what started it, for example HTTP Webhook, Scheduled (Cron), Manual Automation, Slack, Microsoft Teams, Discord, Email, Linear, Chat UI, Notebook chat, API key, Public Chat Link or Called by another agent. A conversation whose runs came from more than one source shows Mixed.

​
Run details

Click a run to open Execution Details:
  • Execution Status, Started, Duration and the Model the run used, with where it came from (from API call, from automation, from harness, agent default or system default)
  • The Pull Request link, when an older repo-run automation opened one, and the Error for failed runs
  • Transcript: every event in order — user messages, reasoning, agent replies, tool calls and tool results. Use Show: to filter event kinds, and click an event to see its full content.
  • Summary chips: Wall (end to end, including sandbox boot), API (model time only), TTFT (time to first token), Tokens, Cache hit rate and Cost
  • Run context: the model and prompt, the agent and instruction files loaded, the harness settings, a timing waterfall, usage, and whether the context was compacted
  • App Log: sandbox lifecycle events such as boot, hydration and errors
Click Stop Execution to stop a running run; its session is preserved. Live runs stream new events as they arrive.

​
Conversation details

Open a conversation to see Session Details: every run in the session, the total duration and time range, and the full Conversation Transcript across turns.
Debugging a tool problem? Open the run’s Transcript, find the tool call, and compare its input with the tool result. Run context shows which MCP servers and how many tools were available to that run.
You can also read execution history and output from the terminal. See CLI.

​
Next

Templates

Start from a pre-built agent

Apps

Build notebooks and custom apps on the same data and context