Self-Hosted SNAP: Bring Your Own Model, Own Your Enterprise Memory
product
How SNAP runs inside your infrastructure with the model of your choice — hosted commercial engines or open-source in-house — under governed access.
Enterprise AI often forces a trade: use a hosted vendor and hand over your data, or self-host and lose the frontier models. SNAP takes a different line. The memory layer runs on your infrastructure, and the model plugs in behind it.
What "self-hosted" actually means in SNAP
SNAP is deployed inside your environment — your cloud tenancy or on-prem cluster. The knowledge graph, the source layer, the capture pipeline and the audit trail all live where your other regulated systems live. Nothing about your captures, connectors or graph state has to leave the boundary.
Bring your own model
The reasoning engine is decoupled from the memory. You choose:
- A hosted commercial LLM (via your existing enterprise contract and region)
- An open-source model running in-house on your own GPUs
- A mix — one model for drafting, another for sensitive workloads
Swap models later without rebuilding the graph. The captures, links and citations you've accumulated stay intact.
Governed access, grounded answers
Access follows your existing identity and permission model. Every answer cites the source nodes it was built from, so reviewers can trace a claim back to the document, screenshot or note it came from. That grounding is the same discipline behind our Responsible AI-by-Design Framework — auditable inputs, not just polished outputs.
Why it matters
You rent the model. You rent the storage. You own the memory. When contracts change or a better model ships, the institutional knowledge doesn't churn with them. For the wider platform view, see the Silverberry AI Platform.