Why Watchlight
One architecture, stage by stage
Governing one checkpoint is a feature. Governing a run from the intent an agent declared to the record of what it did is an architecture. Here is what Watchlight AI Beacon does at each stage, and why each stage needs the one before it.
One architecture, stage by stage
These are not six features that happen to ship together. Each one is what Watchlight AI Beacon does at a stage of the execution lifecycle, and each depends on the ones before it.
Task-scoped, intent-aware authority
Authority is granted for the intent the agent declared and the task in front of it, and nothing more.
Deterministic pre-action authorization
Every action is checked against formal, versioned policy before it runs, and scoped, time-bound authority is validated across every hop, agent to agent to tool. No language model sits in the trust path.
Two enforcement layers, one decision
The same decision is enforced in-process through the framework plugins and on the wire through the proxy, across agent frameworks, evaluated locally in your environment.
Drift measured against the agent’s own baseline
Behavior is scored against the baseline that agent established and the plan it declared, against a threshold rather than a model’s judgment, so the same behavior yields the same result.
Containment while the run is still going
Stop the run, quarantine the agent, sever the delegation subtree, or revoke the authority across the fleet, while the run is still going.
Signed execution lineage
A tamper-evident record of who authorized each action, and why, as audit-grade evidence.
Common questions
What is Agent Execution Governance?
Agent Execution Governance is what Watchlight practices within the Agent Runtime Governance category: governing an agent’s execution from the intent it declared, through authorization, execution, drift and containment, to a signed record of what it did, rather than deciding a single tool call in isolation.
How is Watchlight different from a tool-call gateway?
A per-call check decides one request and is finished. Watchlight AI Beacon governs the whole run: authority is scoped to the intent the agent declared, validated across every delegation hop, enforced in-process and on the wire, and a run that drifts from its plan can be contained while it is still going.