HeartBeatAgents

Models

View as Markdown

Any model, any provider: how HeartBeatAgents routes between a dozen LLM providers, including fully local inference via Ollama.

HeartBeatAgents doesn't lock you into one model vendor. Each agent picks a primary provider and model independently, and you can mix providers across your fleet.

Supported providers

  • Anthropic
  • OpenAI
  • Google
  • Amazon Bedrock
  • DeepSeek
  • Mistral
  • Groq
  • xAI
  • OpenRouter
  • Qwen
  • Moonshot AI
  • Ollama (local)
  • Custom OpenAI-compatible endpoints

Add providers from Settings → Providers (see First Setup and Configuration).

Why mix providers

A high-volume triage agent benefits from a fast, inexpensive model; an agent doing deep research or complex multi-step reasoning benefits from a stronger, typically slower, pricier one. Because provider and model are set per agent rather than globally, you can optimize each agent independently.

Running fully local with Ollama

Selecting Ollama as a provider runs inference on your own machine with no API key and no per-token cost: the same local-first principle that governs the rest of the platform. This is the only path to running HeartBeatAgents completely free end-to-end, at the cost of needing enough local CPU/GPU and RAM to run the model you choose (see System Requirements).

Context window and tool access

Maximum context window varies by model and provider, up to roughly 1 million tokens on the largest supported models. Regardless of provider, an agent can call any of the 625 tools made available through its connected Integrations; tool access is a property of the agent's configuration, not of the model itself.