AI providers
Configure a custom OpenAI-compatible provider and choose reply routing
Starter and Pro organizations can save custom OpenAI-compatible chat providers. Only the organization Owner can create, test, edit, enable, delete, or route providers. Bubblo’s managed AI remains available according to the plan’s managed-reply allowance.
Document and search embeddings always use Bubblo’s managed embedding model. A custom provider changes chat generation, not the dimensions of the organization’s existing knowledge index.
Add and test a provider
Open Organization → AI providers and add a provider with:
- A public HTTPS OpenAI-compatible base URL
- An API key
- One or more model IDs and the model to use currently
Private and local network endpoints are blocked. The API key is encrypted at rest and is not returned after it is saved.
A new provider starts disabled. Run its connection test before enabling it. Only one provider can be enabled at a time; enabling one disables the previously enabled provider. Disable the current provider before editing its connection or model settings, then test the changed configuration again.
If the organization has an active Action, the provider test must also confirm tool-calling support before the provider can be enabled for that workload.
Choose reply routing
The routing setting controls which source handles a new AI reply:
| Routing mode | Behavior |
|---|---|
| Use Bubblo AI first | Reserve a managed reply first. If the managed allowance is exhausted, use the currently enabled custom provider. |
| Use current provider only | Send the reply directly to the currently enabled custom provider. An enabled provider is required. |
Saved providers are not an automatic failover pool. Bubblo uses only the current enabled provider and does not switch to another saved custom provider if a request fails. A provider runtime failure puts the conversation into Needs human so a teammate can continue in Inbox; it does not silently try another provider.
Data sent to the provider
Chat prompts, retrieved knowledge, page context, and Action results can be sent to the selected provider to generate a reply. Review the provider’s own data-handling terms before enabling it. Embedding requests continue to use Bubblo’s managed embedding model.
After a downgrade to Free, saved provider configurations remain visible and can be removed, but custom-provider operations and routing are inactive until the organization returns to an eligible plan. See Billing and usage.