Product13 rooms
How Sophon works, room by room.
A model call is stateless: it forgets, it dies with its process, and it does what the prompt says. Sophon adds the three layers a model doesn't come with. State is an entity graph you can read, correct, and delete. Execution is a Graph-mode run that checkpoints each transition, so it comes back at the step it stopped on. Authority is risk levels, approval gates, and a credential vault outside the model, where it cannot argue with them.
Around them sit the surfaces you reach it from, every way in held to the same gates, and the ways you extend it.
01The floor plan
Pick a room.
The layer a model runs inside
State
What it knows after the window closes.
Execution
What happens to the work when the process dies.
Authority
Who is allowed to act.
Surfaces
Every way in, the same gates.
Extend
What you add to it.
02Under every room
One pipeline, any model, your infrastructure.
- One pipeline for every message
- Every message runs through twelve composable middlewares: budgets, routing, context window management, tool filtering, planning, parallel execution, approvals, and the LLM call itself. The pipeline · Planning
- Model-agnostic
- Anthropic, OpenAI, Gemini, Mistral, xAI and 20+ other provider types, or local models via Ollama, LM Studio or any OpenAI-compatible server; none configured by default. Pin one model, or let Auto pick a model per role. Model routing
- Self-hosted
- Docker Compose or Kubernetes with Helm, on any infrastructure you control. How your data stays yours
Put it to work.
Install Sophon, connect a model, and give it a goal. The Personal tier is free for individual, non-commercial use.