Troubleshooting the Chat Agent Innovation Release

Use these fixes to resolve common Hybrid Manager (HM) Chat Agent issues for users and administrators. If you don't see your issue here, contact EDB Support with the request-tracing ID shown in the HM console.

The sparkle icon doesn't appear

Problem: the rainbow sparkle icon is missing from the HM header.

Cause: the chatbot scenario isn't installed. Chat Agent is part of the chatbot scenario.

Fix: check with your HM administrator that the chatbot scenario was selected at install or upgrade time. See HM installation scenarios.

The sparkle icon is disabled with a tooltip

Problem: the icon is visible but grayed out. Hovering shows a tooltip such as The administrator needs to configure an available model to enable the Chat Agent — or another reason.

Cause 1: no model is configured for Chat Agent. This happens when the chatbot-model-creds secret doesn't exist or contains no models.

Fix 1: the administrator must create the chatbot-model-creds secret with at least one model. See Configuring the models. To replace the models on an existing installation, see Changing the models later.

Cause 2: the administrator created the chatbot-model-creds secret after Chat Agent had already started. HM restarts Chat Agent when the model settings in an existing secret change, but not when the secret is first created.

Fix 2: the administrator can restart Chat Agent so that it reads the secret:

kubectl rollout restart deployment/upm-agent-proxy -n upm-agent-proxy

Cause 3: the Chat Agent service is unavailable.

Fix 3: the tooltip includes a reason. Wait a moment and try again, or check with your administrator that the supporting HM workloads are healthy.

Generation stalls or fails

Problem: the Waiting… placeholder bubble stays for a long time, or the response ends with Request failed, please try again!

Cause 1: the configured model is unreachable or has reached a usage limit.

Fix 1: try a different model from the model selector if more than one is configured. Otherwise, check the model provider's status, and confirm the endpoint URL and API key with your administrator.

Cause 2: the prompt is too long. Each user message is capped at 10,240 characters.

Fix 2: trim the prompt and retry.

Cause 3: network egress is blocked.

Fix 3: confirm that the HM cluster has outbound access to the model endpoint, including third-party provider APIs.

Cause 4: the model rejects the configured default thinking level. Not every reasoning model accepts a per-request thinking-level setting, so the turn fails when one is passed.

Fix 4: ask the administrator to remove the thinking-level settings for that model, or switch to a model that documents support for per-request thinking-level configuration. See Setting model capabilities.

Chat Agent doesn't answer my question

Problem: Chat Agent says it can only help with HM topics.

Cause: Chat Agent answers only HM-related questions. Off-topic questions — general knowledge, math, creative writing, and other products — are out of scope.

Fix: rephrase your question to be about your HM estate, an HM feature, or HM documentation.

Chat Agent says it can't perform an action

Problem: Chat Agent reports a 403 or not authorized error and stops, or says you don't have permission to perform the action.

Cause: Chat Agent runs as your user account. If you don't have the role to perform the action in the HM console, Chat Agent gets the same answer.

Fix:

  • Confirm with an administrator that you hold the right organization or project role.
  • Check that you are a member of the project the action targets.
  • For action types that require approval workflow (such as some automation tasks), wait for an administrator to approve.

See Authorization model.

Chat Agent doesn't see my cluster or project

Problem: Chat Agent says it cannot find a project or cluster you know exists.

Cause 1: you aren't a member of the project that owns the resource. Chat Agent only sees resources within projects you can access.

Fix 1: ask an administrator to add you to the project.

Cause 2: you misspelled the resource name. Resource names are unique within a project. Chat Agent resolves a name to an ID by listing the resources first, and a typo causes it to find no match.

Fix 2: copy the resource name from the HM console and retry.

Chat Agent says a topic post-dates its knowledge

Problem: Chat Agent says the topic you asked about is newer than its bundled documentation snapshot.

Cause: each HM release ships with a snapshot of the official HM documentation, with a stamped knowledge cutoff date. Chat Agent reports when the documentation snapshot does not contain an answer.

Fix: refer to the official EDB Hybrid Manager documentation site for newer information. To benefit from a fresher in-product snapshot, upgrade HM to a more recent release.

Chat Agent keeps asking follow-up questions

Problem: Chat Agent keeps asking for more information before completing your request.

Cause: some Chat Agent flows follow a predefined sequence of inputs to complete a specific task — Postgres Cluster lifecycle management is one example. When Chat Agent doesn't have all the required parameters, it keeps asking until it does, rather than guessing or proceeding with incomplete data.

Fix: fill in the requested information to continue, or cancel the form or type an unrelated message in chat — the question clears.

Charts don't render in the response

Problem: the response shows raw chart syntax or a malformed placeholder instead of a rendered chart.

Cause: chart generation requires the model to emit precise structured output. Smaller models often struggle with the syntax and produce charts the renderer can't display.

Fix: switch to a Large or Extra Large model from the model picker if your administrator configured more than one. Otherwise, ask the administrator to add a stronger model. See Model strength tiers.

Reduced capability

Problem: Chat Agent doesn't offer cluster creation, doesn't ask structured follow-up questions, or doesn't start the full interactive flow you've seen before.

Cause: Chat Agent capability scales with the strength of the configured model. A Small or Medium model handles fewer flows than a frontier model, and structured follow-up forms are only used when the model supports them.

Fix: select a stronger model from the model picker if your administrator configured more than one, or ask the administrator to add a stronger model. See Model strength tiers.

Contacting support

If the issue persists, capture these details before contacting EDB Support:

  • HM troubleshooting information — open About from the HM console footer and select Copy troubleshooting information to capture the version, build, scenarios, organization, session, and platform details.
  • The model and model provider in use, if known.
  • A copy of the prompt you sent and the response you received.