The agent control plane, at the endpoint.
A control plane is the layer that decides what an agent is allowed to do before it acts: identity, policy, inspection, and audit over every model, tool, and MCP call. Alyria puts that layer where agents actually run, on the endpoint, across every vendor, under your keys.
The governance and decision layer: identity, policy enforcement, real-time inspection, and audit that determine what an agent may do before it acts. Alyria is that plane for your agent fleet.
Where the work happens: the model inference, tool calls, and API requests an agent makes. Alyria sits in front of it at the endpoint, so even agents that never touch a cloud gateway are still governed.
Callers, control plane, destinations
Every call an agent makes flows through the control plane before it reaches a model, a tool, or your data. Left to right on desktop; top to bottom on mobile.
Where agents run
What an agent may do, before it acts
Models, tools, data
The enforcement point is Beacon, on the endpoint. No cloud round-trip on the hot path, and no agent slips past by avoiding a gateway.
One module per job in the plane
The celestial family maps one-to-one onto the architecture: where each piece sits, and what it does there.
Beacon
Endpoint enforcementA signed, user-space agent on every machine. Inventories the AI layer and evaluates policy on every MCP call locally, where the call originates. Monitor first, enforce when ready.
Lyra
Policy-as-codeOne capability-brokered, information-flow-aware policy language. Written once, enforced deterministically at the Beacon, with one signed audit chain across every decision.
Constellation
Mesh + shared memoryOrg facts and ingested docs recalled over MCP, scoped by division and role from your IdP, with a prompt-injection safety layer on every read and write.
Spectra
Telemetry & auditOpenTelemetry from every Beacon: policy decisions, MCP activity, inventory. Routed to Observatory or straight into the SIEM your SOC already watches.
Umbra
Agent secretsFleet data and secrets encrypted under an envelope key we delete on request; enterprises bring their own (BYOK / HYOK). Static blobs are opaque to us at rest. A2A and client-held keys are on the roadmap, behind an external audit.
Observatory
Cloud consoleAlyria Cloud: enterprise SSO, fleet and policy dashboards, findings and posture, exportable tamper-evident audit. The managed surface over the whole fleet.
Statio
Agent runtimeThe governed runtime agents run inside, on a Beacon-governed node in your cloud, so every model and tool call is mediated by construction. Engine- and cloud-agnostic, Bedrock-first. In active development with a launch design partner.
What makes it a control plane, not a proxy
Cross-vendor and endpoint-resident
Enforcement lives at the endpoint, in Beacon, not only in a cloud gateway. That governs the agents a gateway never sees: framework agents and direct-provider CLIs that call models through their own libraries.
Safe enablement, not a veto
The same layer that routes agent traffic enforces policy on it, so governance and enablement stop being in tension. You can say yes to the Claude Code and Cursor rollout because you can see and bound it.
Policy as code
Rules are versioned, testable, and executable, applied at the point of use, not a wiki page. Run any policy in monitor mode first to see what it would decide before it blocks anything.
Under your keys
Fleet data is encrypted under a key you can delete or bring yourself. Cut us off and verify it in your own logs. Every agent decision lands in one signed, tamper-evident chain you can export.
Open standards, no lock-in
OAuth 2.1 and OIDC for identity, MCP for tools, OpenTelemetry for signal. Standard formats your team already routes, retains, and trusts.
Per-identity correlation
Identity threads through SSO to model call to tool call to response, so the audit trail answers who, not just what. Access is scoped by division and role resolved from your IdP.
See the plane over your own fleet.
A read-only deploy in about 30 minutes returns a board-ready Agent Exposure Report: every agent, tool, and MCP connection, and every policy violation, across the fleet.