Skip to main content
​
Updates

​
🧠 Knowledge signals from chat

Every conversation where a user corrects an answer, clarifies a metric, or states a business rule is a potential source of domain knowledge. Until now, that signal was lost after the chat ended. Knowledge Signals from Chat lets domain admins surface those insights from any conversation and promote them directly to the domain knowledge base with a single review step.⚙️ How it works:
  1. Open any conversation in your domain
  2. Click the brain icon (Knowledge Suggestions) in the chat controls bar
  3. Review the suggested entries; each one is editable before accepting
  4. Accept, Decline, or edit inline, then accept. Nothing is published automatically
Yes. Once accepted, the entry is written to the domain knowledge base and available for retrieval on the next query.
To edit a suggestion, use the inline text editor. Click into the text, make your changes, and then click Accept. The edited version is the one that gets saved.
Navigate to the Domain Knowledge tab. The confirmation notification after acceptance includes a direct link. You can edit or delete the entry there.
The Knowledge Suggestions panel is currently in early access while we refine its performance and quality. Review and edit suggestions before accepting.
Currently available to Domain Admins only. This feature will extend to Explorers in a future release.

​
🗂️ Agent version control

Agent edits no longer modify the live version. Every change creates a draft, leaving the published version running undisturbed until you’re ready. Publishing a new version increments the version number and locks that version. Run logs are permanently tied to the version that executed them, so reviewing a past run always shows the correct graph and configuration.⚙️ How it works:
  1. Edit the agent. A draft is created automatically, and the published version continues running
  2. Test using Preview before committing any changes
  3. Click Publish. The draft is locked as the new published version, and the version number increments
  4. Open any past run in the Run Visualizer to see the exact version used for that run
Agent Versioning 01
No. In-progress runs complete against the version they were dispatched on. Publishing only affects runs triggered after the publish.
Version history is retained and viewable. One-click rollback is on the near-term roadmap.
Each agent has one draft at a time. Concurrent edit locking is on the roadmap.
We always recommend testing your drafts before publishing. Once published, the draft becomes the locked published version. Testing beforehand is especially important for agents with scheduled runs or production recipients.

​
✅ Agent Reviewed runs

Reduce agent non-determinism by marking successful runs as golden examples. When a run is marked as Reviewed, its full execution history (LLM interactions, tool calls, and outputs) is injected into future runs of the same prompt as historical context. This anchors SQL patterns, output formatting, and decision logic to a run you’ve confirmed is correct.⚙️ How it works:
  1. Open an agent and click View Runs to access its Previous Runs
  2. Click Mark as reviewed on a successful run, a green Reviewed chip appears on the run header
  3. On future runs, WisdomAI fetches up to 3 matching reviewed conversations and injects them as context before generating a response
Reviewed Run
Color/StatusMeaning
greenCompleted successfully
redFailed or errored
grayDid not execute (branch not taken)
spinningCurrently in progress (for live runs)
Up to 3, selected by most-recently-reviewed first.
Yes. The action can be toggled. Removing reviewed status removes the run from the reference pool immediately.

​
🔍 Agent Run visualizer

The Run Visualizer renders a full execution trace of any past or live agent run directly on the agent graph: every node color-coded by status, with a detail panel showing inputs, outputs, tool calls, and timing. Because the graph displayed is always the version used for that specific run, it serves as a reliable audit trail even after the agent has been updated.⚙️ How it works:
  • Go to Agents, open an agent, and navigate to Previous Runs
  • Click any run to open its Run Visualizer
  • Click any node to see its input, output, tool calls, and execution duration
  • For live runs, nodes update in real time as the execution progresses
Run Visualizer
Always the version used for that run. If the agent has been edited and republished since, historical runs still show the original version.
Yes. Visual Mode agents show the full node graph. Prompt Mode agents show a structured timeline of LLM calls, tool invocations, and outputs.
You can share a specific run’s visualizer link with colleagues using the direct URL included in email and Slack delivery notifications. Access is still subject to the agent’s permissions.

​
🔌 Dremio connector

WisdomAI now connects natively to Dremio Cloud and Dremio Software via Apache Arrow Flight, enabling natural-language queries against your data lakehouse without manual dialect configuration or JDBC workarounds. WisdomAI handles schema crawling, view detection, and Dremio-specific SQL generation are handled automatically. Enterprise deployments using internal CA certificates are supported via custom certificate upload during setup.⚙️ How it works:
  1. Go to Connections > Add connection and select Dremio
  2. Enter the Flight endpoint host and port
  3. Provide authentication credentials: Personal Access Token for Dremio Cloud; username/password or PAT for Dremio Software
  4. Configure TLS and optionally upload a custom root certificate for internal CA environments
  5. WisdomAI crawls the schema and makes the domain available for natural-language queries
  6. Define the source path scope (Dremio’s multi-level namespace paths are supported)
Dremio
Yes. Dremio Cloud uses a Personal Access Token. Dremio Software supports username/password via Flight handshake or a PAT.
WisdomAI automatically generates SQL in Dremio’s dialect. No manual configuration is required.
WisdomAI applies its standard query result limits. Dremio’s own resource controls also apply.

​
📊 Dashboard improvements

Three updates make editing, exporting, and working with dashboard tiles faster and more reliable.
  • Inline tile renaming: Rename a dashboard tile by hovering over the title and clicking to edit it directly, without navigating through the overflow menu. Viewers see a plain title with no edit option, keeping the interface clean for non-editors.
  • Tile export and copy: Exporting and copying dashboard tiles now includes a reorganized menu, live progress tracking for CSV exports, and image exports that exclude UI chrome.
  • SQL widget trajectory: The Add SQL Tile flow now creates a trajectory at widget creation time, matching the behavior of chat-created widgets. All dashboard widgets now carry trajectory lineage, supporting trajectory-based editing going forward.

​
🔌 BigQuery OAuth available for all users

BigQuery OAuth is now available on all deployments without requiring a per-environment feature flag. When setting up a BigQuery connection, the “Enable OAuth for user queries” checkbox appears in the connection form for all users.

​
🔐 Streamlined email domain allowlist enforcement

The email domain allowlist is now enforced only at account-creation time, improving response times for all requests. A notice in the settings clarifies that changes to the allowlist are not retroactive: existing users are unaffected by updates to the list.

​
⌨️ Cmd+Enter shortcut for SQL generation in Playground

Press Cmd+Enter (or Ctrl+Enter on Windows and Linux) in the Playground to trigger SQL generation without reaching for the Generate SQL button.

​
📄 10× faster chat exports

Exporting large chat histories is now significantly faster and less likely to cause system interruptions, with a 10× reduction in memory usage during the export process.

​
🧠 Domain Health replaces the Suggestions tab

For admins and superusers, Domain Health is now the single place to review and act on modeling suggestions. The Suggestions tab is no longer shown for these roles, and any existing bookmarks to the Suggestions page redirect to the domain Overview.