Configure Providers

HiveMind OS is provider-agnostic — connect one or many LLM backends, and the model router picks the right one for each request based on data classification, task type, and availability.

All provider configuration lives in your config.yaml under models → providers.

OpenAI

models:
  providers:
    - id: openai
      kind: open-ai-compatible
      name: OpenAI
      auth: env:OPENAI_API_KEY
      base_url: https://api.openai.com/v1
      channel_class: internal
      models:
        - gpt-4o

Anthropic

models:
  providers:
    - id: anthropic
      kind: anthropic
      name: Anthropic
      auth: env:ANTHROPIC_API_KEY
      base_url: https://api.anthropic.com
      channel_class: private
      models:
        - claude-sonnet-4-20250514

Set channel_class: private if your Anthropic agreement covers sensitive data — the model router will respect this when routing prompts.

GitHub Copilot

models:
  providers:
    - id: copilot
      kind: github-copilot
      name: GitHub Copilot
      auth: github-oauth
      channel_class: internal
      models: []

No API key needed. HiveMind OS launches the GitHub OAuth device flow — sign in with your GitHub account and you're done. The token is stored securely in your OS keychain.

Free to use

If you have a GitHub account with Copilot access (free, Pro, or Enterprise), this provider costs you nothing extra. Great way to get started.

Ollama (Local Models)

Run models entirely on your machine. No data leaves your device.

models:
  providers:
    - id: ollama
      kind: ollama-local
      name: Local Models
      auth: none
      channel_class: local-only
      models:
        - llama3.2

Make sure Ollama is running before starting HiveMind OS. The default base URL is http://localhost:11434/v1 — you can omit base_url if using the default.

OpenAI-Compatible

For any API that implements the OpenAI chat completions spec — self-hosted models, corporate proxies, or third-party services.

models:
  providers:
    - id: my-custom-api
      kind: open-ai-compatible
      name: My Custom API
      base_url: https://my-api.example.com/v1
      auth: env:MY_API_KEY
      channel_class: internal
      models:
        - my-model

This also works with Azure OpenAI (use the alias azure-open-ai for kind if you prefer).

Microsoft Foundry

With API Key

models:
  providers:
    - id: azure-foundry
      kind: microsoft-foundry
      name: Azure Foundry
      base_url: https://my-resource.services.ai.azure.com/api/projects/my-project
      auth: env:AZURE_API_KEY
      channel_class: private
      models:
        - gpt-4o

With Azure Identity (Managed Identity / CLI)

models:
  providers:
    - id: azure-foundry
      kind: microsoft-foundry
      name: Azure Foundry
      base_url: https://my-resource.services.ai.azure.com/api/projects/my-project
      auth: azure-default
      channel_class: private
      models:
        - gpt-4o

The azure-default auth method uses the Azure credential chain — no API key required. It tries the following sources in order:

1. Managed Identity — automatically available on Azure VMs, App Service, Azure Functions, AKS, and other Azure-hosted compute
2. Azure CLI — for local development, authenticate with az login
3. Azure Developer CLI — authenticate with azd auth login

Prerequisites for Azure Identity

Your identity (managed identity or user account) must have the appropriate role assignment on the Azure AI resource, such as Cognitive Services User or Cognitive Services Contributor.

Base URL Format

The base_url for Microsoft Foundry must include the project path: https://{resource-name}.services.ai.azure.com/api/projects/{project-name}

You can find your project name in the Azure AI Foundry portal. The "Fetch Models from API" button will list only your deployed models (not the entire catalog).

Multiple Providers & Fallback Chains

You can configure multiple providers. The model router selects among them based on data classification, task type, and priority. If the primary provider fails (rate limit, timeout, auth error), HiveMind OS automatically cascades to the next eligible provider.

models:
  providers:
    - id: anthropic
      kind: anthropic
      name: Claude (Primary)
      auth: env:ANTHROPIC_API_KEY
      base_url: https://api.anthropic.com
      channel_class: private
      models:
        - claude-sonnet-4-20250514

    - id: openai
      kind: open-ai-compatible
      name: OpenAI (Fallback)
      auth: env:OPENAI_API_KEY
      base_url: https://api.openai.com/v1
      channel_class: internal
      models:
        - gpt-4o

    - id: ollama
      kind: ollama-local
      name: Local (Offline Safety Net)
      auth: none
      channel_class: local-only
      models:
        - llama3.2

With this setup:

• Claude handles primary reasoning tasks and can receive CONFIDENTIAL data (via private channel class)
• OpenAI kicks in as a fallback for INTERNAL-classified prompts
• Ollama runs locally for scanning tasks — data never leaves your machine

Environment Variables for Secrets

Use the env:VAR_NAME syntax in the auth field to reference environment variables. HiveMind OS resolves these at startup.

auth: env:OPENAI_API_KEY         # ✅ reads from environment
auth: none                       # ✅ no auth needed (e.g. local models)
auth: github-oauth               # ✅ GitHub device flow
auth: api-key                    # ✅ API key from OS keychain
auth: azure-default              # ✅ Azure Managed Identity / CLI (Microsoft Foundry only)

Never hardcode API keys

Always use env:VAR references for secrets. Hardcoded keys in config files risk leaking credentials through backups, version control, or file sharing. Set your keys in your shell profile, .env file, or OS secret manager.

Channel Class Reference

Each provider needs a channel_class that tells the model router what data is safe to send:

Channel Class	Meaning
local-only	Most sensitive — only local providers should use this
private	Sensitive data — provider must have a data-handling agreement
internal	General-purpose — safe for org-internal data
public	Open data only — nothing private or proprietary