HalosHalosDocs

Model Providers

The runtime is model-agnostic. Any LLM that supports tool use can power an agent. The personality_model field on each agent record maps to a Vercel AI SDK provider instance.

Supported Models

Anthropic

| Model ID | Label | |----------|-------| | claude-sonnet-4-20250514 | Claude Sonnet 4 (default) | | claude-haiku-4-5 | Claude Haiku 4.5 | | claude-opus-4-6 | Claude Opus 4.6 |

Env var: ANTHROPIC_API_KEY

OpenAI

| Model ID | Label | |----------|-------| | gpt-4o | GPT-4o | | gpt-4o-mini | GPT-4o Mini | | gpt-4.1 | GPT-4.1 | | gpt-4.1-mini | GPT-4.1 Mini | | o3-mini | o3-mini |

Env var: OPENAI_API_KEY

Google

| Model ID | Label | |----------|-------| | gemini-2.0-flash | Gemini 2.0 Flash | | gemini-2.5-pro | Gemini 2.5 Pro | | gemini-2.5-flash | Gemini 2.5 Flash |

Env var: GOOGLE_GENERATIVE_AI_API_KEY

How Resolution Works

resolveModel() in providers.ts:

  1. Looks up the personality_model string in the model registry
  2. If found, returns the corresponding Vercel AI SDK model instance
  3. If the string contains / (e.g. anthropic/claude-sonnet-4-20250514), parses the provider prefix and creates the instance directly
  4. Falls back to Claude Sonnet 4 if the model is unknown
// Registry lookup
resolveModel("gpt-4o")                        // → openai("gpt-4o")
resolveModel("claude-sonnet-4-20250514")       // → anthropic("claude-sonnet-4-20250514")
resolveModel("gemini-2.0-flash")               // → google("gemini-2.0-flash")

// Provider/model format
resolveModel("anthropic/claude-opus-4-6")      // → anthropic("claude-opus-4-6")
resolveModel("openai/gpt-4.1")                 // → openai("gpt-4.1")

// Unknown → fallback
resolveModel("llama-3-70b")                    // → anthropic("claude-sonnet-4-20250514")

Adding a New Provider

  1. Install the Vercel AI SDK provider package (e.g. @ai-sdk/mistral)
  2. Add entries to the MODEL_REGISTRY in providers.ts
  3. Add a case to the provider/model parser
  4. Add entries to listSupportedModels() for UI dropdowns

Per-Agent Model Selection

Each agent has a personality_model column in the agents table. This can be set at creation time or changed in agent settings. Different agents can use different models — you could have a fast Haiku agent for triage and an Opus agent for complex research.