Model Routing
OpenHuman routes each workload to the most appropriate model. You can also override the defaults manually.
Default routing table
| Workload | Default model | Override key |
|---|---|---|
| Chat | claude-sonnet-4-20250514 | OPENHUMAN_MODEL_CHAT |
| Reasoning | claude-sonnet-4-20250514 | OPENHUMAN_MODEL_REASONING |
| Vision | claude-sonnet-4-20250514 | OPENHUMAN_MODEL_VISION |
| Memory embeddings | all-minilm (local) | OPENHUMAN_MODEL_EMBEDDING |
| Memory synthesis | gemma3:1b-it-qat (local) | OPENHUMAN_MODEL_MEMORY |
| Subconscious | claude-sonnet-4-20250514 | OPENHUMAN_MODEL_SUBCONSCIOUS |
| Heartbeat | Small chat model | OPENHUMAN_MODEL_HEARTBEAT |
Manual override
[model]
chat = "claude-opus-4-20250514"
reasoning = "claude-opus-4-20250514"
Local model fallback
If a local provider (Ollama / LM Studio) is configured and reachable, local workloads route there automatically. Remote workloads still go to the cloud provider.
See also
- Local AI - How local routing is configured
- Privacy & Security - What data local processing keeps on-device