The engine is yours.

sansin-core is the safety engine, MIT licensed. Run it standalone, or plug it into any MCP-speaking agent as an MCP safety proxy — the same engine that powers the Sansin control plane.

pip install sansin-core
terminal
# MCP safety proxy: drop-in firewall for any MCP server
sansin-mcp-proxy --upstream stdio://npx -y @modelcontextprotocol/server-filesystem /tmp

Three functions: check(), override(), status(). Ships with pre-trained priors for 15 common tools. Add [mcp] for the safety proxy.

Safety you can verify

Isolation by tenant, coherence by fleet

Policies, priors, logs, and model context are tenant-scoped from day one — never shared, never leaked across tenants. Within a tenant, every agent sees one policy surface and one shared prior, so your fleet stays coherent instead of drifting apart.

The best guardrail knows when to block

A control plane that flags everything is just noise. Sansin scores risk, weighs context, and only intervenes when it matters — so your agents stay fast and your team stays focused.

Every verdict is auditable

See exactly why an action was allowed or blocked — which agent proposed it, the policy that matched, the risk score, the model's confidence. Full trace, no black boxes, one audit log across every agent.

Ask for a seat at the gate.

Sansin is onboarding teams one at a time while the gate earns its priors. Drop your email and we'll reach out when it's your turn.