Documentation
Multi-LLM Orchestration
Powered by GML. One memory layer that every model reads from and writes to — with routing that picks the right model for each task.
The five primitives
- Recall. Retrieve the right memory for the task at hand.
- Ingest. Store new memory back into the shared layer.
- Orchestrate. Route tasks between models in a workflow.
- Model Router. Automatically choose GPT, Claude, Gemini or an OSS model based on the task.
- Shared Brain. Every model reads and writes to one memory layer.
The flow
Recallretrieve memory
→
Ingeststore memory
→
Orchestrateroute tasks
→
Model RouterGPT · Claude · Gemini · OSS
→
Shared Brainone memory layer
Why it matters
No single model is best at everything. GML treats models as interchangeable workers over a common memory, so a task can flow from one to the next — research with one, draft with another, verify with a third — while context stays intact the whole way through. Combined with .aal and Context Fidelity™, that's how the handoff stays lossless.
This page describes the orchestration model. Routing rules, supported providers and the API surface will be documented here as they stabilise.