Providers & Models

A provider is where your AI models live — cloud APIs, local servers, or both.

HiveMind OS doesn't lock you into a single vendor. You can wire up as many providers as you like — mixing cloud heavyweights with local models running on your own hardware — and the system will route requests to the right one automatically.

What Is a Provider?

A provider is an API endpoint that serves language models. It could be a cloud service like OpenAI, a self-hosted Ollama instance on your LAN, or even a tiny model running directly inside the HiveMind OS daemon. Each provider has:

• A connection (URL + credentials)
• A list of available models
• A data classification level that controls what information can be sent to it

Supported Providers

Provider	Kind (config value)	Notes
OpenAI-compatible	open-ai-compatible	Any API that speaks the OpenAI protocol — including OpenAI itself
Anthropic	anthropic	Claude Sonnet, Claude Opus
GitHub Copilot	github-copilot	Free for GitHub users — see tip below
Microsoft Foundry	microsoft-foundry	Azure-hosted models with auto-discovery
Ollama	ollama-local	Run open models on your own machine
Local Models	local-models	Models running directly on your hardware

GitHub Copilot — Free AI Models

If you have a GitHub account, you already have access to AI models through GitHub Copilot. HiveMind OS can authenticate via GitHub OAuth — no API keys to manage, no credit card required. It's the fastest way to get started.

Model Roles

Not every task needs the most powerful (and expensive) model. HiveMind OS assigns models to roles so the right model handles the right job:

Role	Purpose	Example Model
Primary	Main conversation & reasoning	Claude Sonnet, GPT-4o
Secondary	Fallback when primary is unavailable	GPT-4o-mini, Gemini Flash
Admin	High-frequency housekeeping (routing, summarisation, triage)	Llama 3, GPT-4o-mini
Coding	Optimised for code generation	Claude Sonnet via Copilot
Scanner	Prompt injection detection	Llama 3.2 (can be a cheap/fast model)
Vision	Image understanding	GPT-4o

When the system needs a model, it resolves in order: explicit role → admin → primary. You can override any role per conversation, per bot, or per workflow step.

Multi-Provider Setup

The real power is mixing providers. Use the best model for each job:

• Claude for complex reasoning (primary)
• Local Llama for classification and scanning (admin, scanner)
• GPT-4o for image understanding (vision)
• GitHub Copilot for code generation (coding)

This keeps costs down, data local where it matters, and gives you frontier-quality results where it counts.

Channel Classification

Every provider gets a data classification level that controls what data can be sent to it.

How Classification Works

Every piece of data in HiveMind OS carries a label — PUBLIC, INTERNAL, CONFIDENTIAL, or RESTRICTED. Every provider has a channel class that determines the maximum data level it can receive. If data is too sensitive for a provider's clearance, the request is blocked or rerouted automatically.

Channel Class	Accepts Data Up To
public	PUBLIC only
internal	INTERNAL
private	CONFIDENTIAL
local-only	All levels (data never leaves your machine)

This means you can safely mix cloud and local providers — sensitive data automatically stays on-device.

Putting It All Together

Here's a real-world configuration using two providers with different trust levels:

providers:
  - kind: anthropic
    name: Claude (Primary)
    channel_class: private
    models:
      primary: claude-sonnet-4-20250514
      coding: claude-sonnet-4-20250514

  - kind: ollama-local
    name: Local Llama
    channel_class: local-only
    models:
      scanner: llama3.2
      admin: llama3.2

Cloud Claude handles conversations and coding with private channel class (accepts up to CONFIDENTIAL data), while local Llama runs prompt scanning and housekeeping tasks with local-only channel class — meaning even your most sensitive data never leaves your machine for those tasks.

Learn More

• Configure Providers — Step-by-step setup for each provider
• Privacy & Security — Deep dive into the classification system
• How It Works — The big picture of HiveMind OS architecture