Documentation
Start here. What this platform does, how to connect an agent to it, and how to run that agent anywhere — a laptop, your own datacentre, or any cloud.
Aegis is a governance layer that sits between an AI agent and everything it is allowed to touch — models, databases, MCP servers, third-party APIs, and the humans who must approve consequential actions. It does not replace your agent framework. You keep your agent; Aegis wraps it.
The one thing to understand first
Aegis runs in two halves, and the split is the whole product. The control plane (this console) holds policy, audit metadata and rolled-up metrics. The client zone runs beside your agent, on your infrastructure, and is where patient data, prompts, model output and tool payloads live. Those never reach Aegis.
What crosses the boundary, and what never does
- Crosses to Aegis
- Which tool was called, allowed or blocked and why, token counts, latency, guardrail hits, credential decisions, audit hashes.
- Never crosses
- Prompt text, model output, tool payloads, patient identifiers, the redaction token map, and anything a human reviewer reads.
What you get for wrapping an agent
- Registration
- Every agent has a named owner, a declared purpose and an autonomy tier.
- Credential brokering
- Short-lived, task-scoped credentials issued per call. No standing keys.
- Guardrails
- Redaction, an output gate, token ceilings and human-approval requirements, configured here and enforced in your zone.
- Human-in-the-loop
- High-risk actions pause for a signature, with 21 CFR Part 11 e-signature semantics.
- Audit
- An append-only, hash-chained record that is re-verified on every read.