Policies and procedures

Supported LLM providers

Single-tenant

AI Hub supports specific large language model providers and APIs for use with AI Hub deployments. This list is subject to change as new models become available or as requirements evolve.

Customers who elect to provision their own LLM endpoints must ensure compliance with these specifications to maintain compatibility and receive full support under applicable Service Level Agreements.

Customers are responsible for ensuring their LLM endpoints remain compatible with the latest specifications. Use of unsupported models or endpoints that don’t maintain complete API compatibility isn’t supported and is excluded from any Service Level Agreements or performance warranties.

For questions or clarifications regarding supported providers and models, contact the support team.

General requirements

All customer LLM endpoints must provide official models from supported providers. In addition:

  • Endpoints must maintain complete API compatibility, including all available parameters, options, model variants, and authentication methods.

  • Custom or alternative models not listed in this documentation aren’t supported.

  • Models and required capabilities might change over time.

Supported providers in agent mode

AI runtime 2.x operates exclusively in agent mode and supports the following model providers and APIs:

OpenAI API / Azure OpenAI API

  • Responses API: Must be enabled on your OpenAI or Azure OpenAI endpoint.

  • LLM models:

    • gpt-5.2
    • gpt-5.2-pro

Google Gemini

  • LLM models:

    • gemini-2.5-flash

    • gemini-2.5-pro

  • Access methods:
    • Google Cloud Vertex AI

Supported providers in legacy mode

AI runtime 1.0.13 is the final release in the 1.x series and will expire according to the standard timeline, unless otherwise requested. Contact Instabase Support to request timeline adjustments.

AI runtime 1.0.13 continues to support the following model providers and APIs in legacy mode:

OpenAI API

  • LLM models:
    • gpt-3.5-turbo-16k (deprecated)
    • gpt-3.5-turbo-1106 (deprecated)
    • gpt-4o-2024-05-13
    • gpt-4o-2024-08-06
    • gpt-4o-2024-11-20
    • gpt-4o-mini-2024-07-18
  • Embedding models:
    • text-embedding-ada-002
  • Moderation:
    • omni-moderation-latest

Azure OpenAI API

AWS Bedrock (Anthropic Claude)

  • LLM models: Cross-region inference endpoints are supported
    • anthropic.claude-3-5-sonnet-20241022-v2:0
    • anthropic.claude-3-5-sonnet-20240620-v1:0
    • anthropic.claude-3-5-haiku-20241022-v1:0
    • anthropic.claude-3-7-sonnet-20250219-v1:0
  • Embedding models:
    • amazon.titan-embed-text-v2:0

Monitoring LLM consumption

When provisioning your own LLM endpoint, there is added support for monitoring and optimizing token consumption. These features must be enabled by your Instabase Support team.

  • On the settings page, use the Tokens dashboard to view token consumption data, including input tokens, output tokens, and top consuming apps. Graphs display consumption over time, with filters and grouping across apps, task types, and fields.

  • In agent mode, the Prompt caching optimization project setting reduces token consumption and LLM costs. Document content is passed ahead of extraction instructions, allowing the LLM to leverage cached data when responding.