Positron Assistant now has preview support for Custom Providers (OpenAI compatible)

[jthomasmock]

As of Positron 2025.10.0, we have added initial preview support for Custom Providers for Positron Assistant! This will also soon be available in preview for Posit Workbench: Positron Pro in the next Q4-2025 release.

Custom Providers allow you to use arbitrary AI model providers that adhere to the OpenAI API specification. Custom Providers could include tools such as litellm, vllm, OpenRouter, Portkey.ai, or other arbitrary centralized routing, proxying, or LLM hosting frameworks that otherwise sit in front of direct model provider access.

Note

In the future, we intend to provide additional guardrail checks for models that have the core capabilities necessary to provide a positive Positron Assistant experience, which at a minimum, would be tool-calling support. In future iterations of Custom Provider support, we intend to at least warn users if models do not meet sufficient requirements to provide the expected user experience.

Caution

While Databot can also use models via a Custom Provider, we currently only suggest using Anthropic models with Databot per our testing during its Research Preview.

Why Custom Providers?

Data Science teams

Custom Providers may be interesting for teams that require:

• On-premise or private cloud deployment: All code and prompts remain within your private or even offline infrastructure, meeting regulatory requirements (HIPAA, GDPR) without transmitting data to external services or countries.
• Governance or auditability: Additional governance tools on top of model routing to provide complete audit trails and centralized access controls. This could enable compliance demonstration and usage monitoring across the organization
• Flexibility and/or Vendor independence: Switching between model providers and model versions without a change in user workflows
• Cost and performance optimization: Deploy models to multiple users near development environments for reduced latency, predictable costs, and custom resource allocation based on organizational needs.

Individual Data Scientists

Individuals may be interested in using Custom Providers to access “free” foundational models that run in-session via tools like Ollama, which often means sharing compute with their data science environment. Our current recommendations, based on model quality and Positron Assistant’s need for tool-calling and agentic workflows, are that models should be run in a separate computational environment.

This recommendation is based on:

• Sufficient model quality: Larger models (70B+ parameters) typically deliver better code completion, reasoning, and problem-solving results compared to smaller models, at the cost of much larger RAM/GPU requirements.
• Required model capabilities: Models must support tool-calling and agentic workflows to enable the full functionality of Positron Assistant
• Resource isolation: Running models separately prevents resource conflicts with data science workloads and allows independent scaling of AI tools and analytical computation.

Enable Custom Providers

To make use of Custom Providers, Positron Assistant must first be enabled as outlined in the Positron Assistant Getting Started Guide. Custom Providers can then be enabled via the positron.assistant.enabledProviders setting in your settings.json.

While Custom Providers could be used to access OpenAI endpoints, you should instead make use of the “openai” option within the same setting. The direct OpenAI option will change to “openai-api” in the 2025.11.0 release to avoid naming conflicts we have recently observed with other tools and model providers.

"positron.assistant.enabledProviders": [
  "openai-compatible",  //## <-- Custom Providers that adhere to OpenAI API standards
  //"openai",    //## <-- For using OpenAI directly in 2025.10.0
  //"openai-api" //## <-- Planned renaming for OpenAI in 2025.11.0
]

Once the Custom Provider option is enabled, you can sign in to your custom provider.

Sign-in to Custom Provider

1. Run the Positron Assistant: Configure Language Model Providers command via the Command Palette to open the language model provider dialog. You can also click the Add Model Provider… button in the top left of Positron Assistant’s chat area.
2. Select “Custom Provider”
3. Authenticate with the model provider by supplying a:
   1. Base URL: API endpoint specific to this provider, router or proxy.
   2. API Key: A long-lived authentication token that provides access to the model service.
4. Once authenticated successfully, the Sign in button will change to Sign out, and available models from the provider will be registered. Some custom providers such as OpenRouter may take additional time to authenticate.
5. Click the Done button once you have added all the desired model providers. You can always come back to this dialog later to add or remove providers.

We want your feedback!

We've heard the interest for Custom Providers in #8592 and #8326 and are happy to provide this initial support.

Please give it a try, and leave any ideas or feedback in this discussion or open an issue if you run into any problems!

[Upipa]

Hi i have problems configuring models provided by azure foundry. I configured liked said above

"positron.assistant.enabledProviders": ["openai-compatible"]

In the configuration panel i setted the API key and the base URL (the same that i use with ellmer) but it fails with the error

Failed to add language model Custom Provider: Could not fetch models Resource Not Found

Am I doing something wrong or there's simply no support for azure foundry right now?

[sharon-wang]

Hey @Upipa, thanks for giving this a try! I think you're running into #9971 (comment) -- we try to fetch models from the /models endpoint, but from the error message you provided, it seems like Azure Foundry may not support that endpoint. Once we have a solution for that issue, it should unblock this particular error you're seeing!

These other issues may be of interest to follow, although they are focused on Azure OpenAI, which may different implementation requirements than Azure Foundry.

• Spike: Test Azure OpenAI via OpenAI different base URL + API key #10171
• Positron Assistant: Support Azure OpenAI with SSO/OAuth via EntraID #10172

[Upipa]

Very sad, i was so excited to try the assistant. I'll wait for this problem to be fixed. Thanks for the answer.