If you’re using OpenAI GPT-5.x or Google Gemini 3 with your own API keys in Heptabase, and you keep seeing a "Something went wrong, please try again later or contact support." message, this guide walks you through the most common causes and how to fix them.
Before using the steps below, check the AI source selected in the chat composer:
If you selected your own OpenAI or Gemini API key, continue with this guide.
If you signed in through ChatGPT Plan, the organization-verification and API-billing steps below do not apply. If the error appears, use Continue if available, switch to another model, and contact Support with the model name, approximate time and timezone, and whether the issue affects one or multiple conversations.
Common symptoms
You may be affected if:
Your OpenAI or Gemini API key shows “Verified” in Heptabase, but:
GPT-5 models always fail with a generic error, or
Gemini 3 Pro always fails with a generic error
Gemini 2.5 works, but Gemini 3 does not
You've subscribed to ChatGPT Plus or Gemini Pro, and you can use ChatGPT 5.1 or Gemini 3 in their web UIs, but the AI agent in Heptabase still fails when using those models.
If you are using your own API key and this sounds familiar, continue with the checks below.
Possible causes
These issues are usually caused by one or more of the following:
OpenAI organization not fully verified
OpenAI requires your organization (even for personal accounts) to complete identity verification before using GPT-5.x via API.Gemini API project billing not set up or on a limited tier
Having a Google AI Pro or similar subscription for the UI is not enough. The API project behind your Gemini key must have billing enabled and sufficient tier access for Gemini 3.Model-specific limitations or instability
Newer models (especially Gemini 3) may be more unstable or restricted than Gemini 2.5.
How to fix the issue
Check OpenAI verification
If GPT-5 models fail but your API key is marked as verified in Heptabase:
Open the OpenAI Platform and go to Settings → Organization → General → Verifications. Verify your organization to access protected models.
If verification is requested, follow the steps shown on the platform. OpenAI may use a third-party service for identity verification.
After completing verification, wait a bit and try GPT-5 again in Heptabase.
If your organization is fully verified and billing is active, GPT-5 should usually start working without changing your API key.
Check Gemini billing and access
If Gemini 2.5 works but Gemini 3 Pro fails despite a paid Google AI plan, the problem is usually with the API project, not the consumer subscription.
Open Google AI Studio and go to API keys.
Check that the key pasted into Heptabase belongs to the project you intend to use and is active—not deleted or disabled.
Confirm that this API project has a billing account attached. A new project may need its billing setup completed before Gemini 3 can be used.
Check the project’s usage tier and quotas (for example, Tier 1 or higher). A free or limited tier may not provide Gemini 3 Pro access even if Gemini 2.5 works.
If needed, paste the key into Heptabase again. First test a query with Gemini 2.5 to confirm the key is valid, then test Gemini 3 Pro.
If Gemini 2.5 works but Gemini 3 still fails after billing is correctly set up, we recommend using Gemini 2.5 as your main model, as it’s currently more stable.
When to contact Support
If GPT-5 or Gemini 3 still shows the same error after you have checked OpenAI organization verification or Gemini API billing and tier access, and re-added your API key, contact support. Include:
The model you were using.
What works and what fails. For example: “Gemini 2.5 works, but Gemini 3 fails,” or “Some questions work, but summarizing a PDF fails.”
Screenshots of the Heptabase error and relevant settings in OpenAI or Google AI Studio.
Our team will look into your issue and help you get it resolved!