How Svalin works

Deployment

Deploy via MDM — zero developer action

Svalin deploys as a Go binary through Jamf, Microsoft Intune, or Kandji. It requires no developer action, no code changes, and no manual configuration. Once enrolled, the agent auto-detects supported AI coding tools on the machine and configures itself transparently.

MCP Proxy Layer

Every tool call logged and policy-checked

For supported agents, Svalin wraps the tool's MCP configuration to intercept tool calls before they execute. Every file read, shell command, and API query that flows through MCP is logged, attributed, and policy-checked before transmission. Approved operations proceed transparently. Blocked or redacted operations are logged as policy events.

LLM Proxy Layer

Prompts attributed before they leave the machine

Where supported, Svalin configures itself as the local LLM proxy. Every prompt sent to a model provider and every response received is attributed to a specific developer session before being forwarded. Sensitive data patterns are detected and redacted before leaving the machine.

This is the layer that gateway-dependent tools depend on developer cooperation to route traffic through. Svalin's agent configures this at the OS level via MDM — no developer action required, and configuration changes are detected as compliance events.

Session Reconstruction

A coherent timeline, not just a list of events

Svalin uses process tree analysis — PID and PPID attribution — to reconstruct the full coding flow from developer prompt to session close. This means the audit trail is not just a list of events — it is a coherent session timeline showing what the agent did, in what order, on behalf of which developer.

Each tool call is linked to the session that spawned it. Each session is linked to the developer identity and device. The result is a navigable audit record: any event can be traced to its full context — what was asked, what was executed, what data moved.

Configuration Integrity

Bypass attempts detected, not just logged

Svalin continuously monitors agent configuration files for tampering. If a developer or administrator changes the LLM proxy URL, adds an unapproved MCP server, or suppresses a hook, Svalin detects the change, automatically re-applies the correct configuration, and raises a compliance alert on the platform.

Telemetry & Storage

Tamper-evident logs, append-only backend

Events are hash-chained and signed on device before being shipped to Svalin's append-only backend — EU-hosted, with separate key management. The result is a tamper-evident log that proves events were not modified after capture.

Sequence numbers and heartbeat checks allow auditors to identify gaps in coverage, not just verify what was logged. The distinction matters: a tamper-evident log proves integrity of what was captured. The heartbeat system helps identify periods where nothing was captured — so gaps are visible rather than invisible.

Coverage Scope

What Svalin governs — and what it doesn't

Svalin governs enrolled devices running supported agents. Knowing the boundary of coverage is part of governance.

Svalin's governance scope is explicit. What falls outside it is outside its visibility — and the platform makes no claim otherwise.