What buyers ask us

Alyria runs a signed agent called Beacon on the machines where coding agents work. It inventories the agent tooling it finds, evaluates your Lyra policy locally against what the agent tries to do, and records each decision in a hash-chained audit trail. A tenant console shows the fleet, the policy and the decisions.

Neither. A gateway governs traffic that goes through it; it cannot see a local MCP server or a shell command that never leaves the laptop. An endpoint security product sees the host but usually has no opinion about which model or MCP tool an agent may call. Beacon covers the local decision path, so the two run side by side rather than one replacing the other.

Not on its own. A gateway sees a request only once a client is routed through it, and local tools, MCP servers started by the IDE, and shell commands do not have to be. That work has to happen on the endpoint, which is the gap Beacon fills.

It adds value when you need per-agent decisions rather than process-level alerts: which model was called, which MCP server was reached, which policy allowed it, and a record an auditor can verify. If your existing tool already answers those questions for coding agents, you may not need this.

Partly. A cloud workspace moves the code, but the agent still runs somewhere, and the same questions apply to that host. Beacon covers the Windows, macOS and Linux machines your teams work on, including the ones behind a cloud workspace you control.

Yes. Beacon reads the MCP server entries in each tool's own configuration and reports the server names and transport it finds, alongside the version of the tool itself. One console view covers the fleet instead of one machine at a time.

Windows, macOS and Linux. It is a signed local daemon on all three, distributed through your own software channel or MDM rather than installed by hand. On macOS the deeper observation runs in a system extension, which your MDM approves as part of the same push.

The quickstart is the shortest path: fetch the signed Beacon from get.alyria.ai, run beacon verify to check the signature and version, then connect it to your tenant with an enrollment token. For a fleet, the same installer goes into your software channel and is pushed by MDM.

Yes. The Windows MSI takes two public properties, OBSERVATORY and ENROLLMENTTOKENFILE, so the enrollment token travels as a file and never appears on a command line. On macOS the agent ships with configuration profiles for its extension and privacy grants, so one push covers install and approval.

No. Beacon is user space on every platform. On macOS it uses the Endpoint Security and network extension frameworks, each with its own entitlement that arrives in a configuration profile, rather than a driver of its own.

One ring, monitor only. Read the inventory on two or three hosts and confirm the tools you actually run are listed. Leave it a week, read the decisions, then promote one category at a time and add the next ring.

Enforcement is off until you turn it on, and when it is on it starts in monitor mode. Decisions are evaluated and recorded, and nothing is blocked. Monitor is a visibility mode, so it is the right place to start and the wrong place to stay.

Beacon is $6 per device per month, with a $30 monthly minimum covering up to five devices. Platform is $299 per organization per month and includes five devices. Enterprise is scoped to your requirements.

No. Every tenant receives a managed baseline of rules, reconciled into monitor mode. That reconciliation never overrides a rule an admin set to enforce or off, and it never silently escalates enforcement.

Lyra is a capability-brokered policy language. A rule matches on the tool, the capability it asks for, the data classification and the destination, and returns a disposition with a rule id and severity. The same policy is evaluated locally on the endpoint and works offline.

Policy evaluation runs locally, so decisions keep being made and recorded without a network hop. Events queue and ship when the machine reconnects, so a flight does not become a gap in the record.

MCP server identity is pinned per tool, so a changed description or a new tool under an old name shows up as a detection with the pinned and observed digests attached, not as a silent update. From the console you can accept the new fingerprint, reset one pin, or repair a store that has been tampered with, and each action is recorded.

Three paths. Verify the decision chain on the endpoint with the chain verifier, page the tenant audit log through the API at 100 entries a page, or stream signed decisions to your own collector through Spectra. Audit history is included for 30 days within plan capacity.

Each chained event commits to the previous link, so a gap or an edit breaks verification and the verifier names the first broken index. It reads the events themselves, so the check does not depend on trusting the system that produced them.

Yes. Decisions are signed at the endpoint and forwarded to Spectra, which terminates TLS itself and takes a destination policy and key set, so your collector sits in front of a component that already speaks the format.

Enroll a host and compare beacon status with what the console shows. Trigger a deny and verify the chain. Uninstall and confirm the device stops checking in. Those three are observable on your own hardware without our involvement.

The MSI uninstaller on Windows, the package manager on Linux, or the MDM that installed it on macOS. Local state stays on purpose so a reinstall resumes the same device. Revoking the device is a separate console action.

No. Uninstalling removes the agent; the tenant record stays until it is revoked. A device shown as Active after its agent is gone will simply stop checking in.

No. The record is the decision: bundle version, disposition, rule id, severity, action, and digest-only detection evidence. Conversation content is not collected as content.

The locations its catalog knows for each tool: install paths and package names, the client configuration files that hold MCP server entries, and the environment variable names that indicate a provider is configured. It takes server names and transport from the MCP block. It does not index your repositories.

On Windows, a virtual service account named for the service, AlyriaBeacon by default, with protected program and state directories. On Linux, the system users _alyria and _alyria-gw with 0700 state and 2750 run directories. On macOS it is a launch daemon with the approved extension entitlements.

Yes. Fleet data is encrypted under an envelope key, with BYOK and HYOK for enterprise fleets, and secrets are issued to agents as short-lived leases rather than stored on the endpoint. Revoke your key and our ability to decrypt ends, visibly in your own KMS log.

Policy evaluation does, because it runs locally with no cloud hop. Enrollment and fleet reporting need a reachable Observatory, so plan a connected window for those and let evaluation carry on in between.