Ambit Product System
Ambit Systems governs autonomous AI action at the point it becomes consequence. Authority decides before execution, using policy, delegation, and resolved consequence context. The decision ledger records the proof. Observatory turns that proof into assurance and policy-readable evidence.
One product system for the action boundary
Ambit is not an agent runtime, an IAM layer, or an observability platform. It is the authority and evidence layer between autonomous intent and downstream consequence: a proposed action is evaluated against policy, delegation, resolved consequence topology, authenticated time, and revocation evidence; the outcome is committed to a tamper-evident ledger; and the evidence can be replayed later.
Ambit Authority
Renders ALLOW, DENY, or ESCALATE before a governed consequential action executes. Decisions are deterministic, policy-bound, delegation-aware, consequence-aware, fail-closed, and committed to the decision ledger.
AuthorityAmbit Observatory
Verifies and interprets Authority's decision records. It explains decisions, prepares evidence bundles, detects ledger gaps, and publishes ledger-cited behavioural facts policy can require later.
ObservatoryIntegration
Places Authority on the execution path through SDK, gateway, or sidecar patterns. The placement test is simple: if a governed action can reach consequence without a decision, the integration is not authoritative.
IntegrationWhat the system proves
Ambit is valuable because it produces enforceable control and durable evidence from the same mechanism.
Authority before consequence
The governed action is evaluated before it reaches the tool, API, database, runtime, memory store, or external endpoint.
Consequence-aware admissibility
The system can distinguish a draft from a deployment, a local effect from external propagation, and a reversible action from an irreversible one before policy decides whether to allow, deny, or escalate.
Exact policy binding
The decision record binds the policy hash, ontology hash, delegation state, request fingerprint, consequence context, authenticated time, revocation evidence, outcome, reasons, timing, and ledger position.
Fresh authority at runtime
Constitutional mode fails closed when delegated authority cannot be proven fresh: unauthenticated time, stale revocation status, missing revocation bounds, or unverifiable revocation evidence deny the action.
Tamper-evident ledger
Decision records are append-only and hash-chained with seq, prev_hash, and record_hash, so mutation, deletion, and reordering are detectable.
Behavioural evidence without hidden enforcement
Observatory can publish behavioural facts, but Authority only enforces them when policy explicitly requires them as resolved context.
Authority can run alone. Observatory compounds the evidence.
Authority alone
A customer can deploy Authority without Observatory and still get pre-execution decisions plus tamper-evident, hash-verifiable decision evidence.
Authority with Observatory
Adding Observatory turns the ledger into operational assurance: explanations, replay, completeness checks, evidence packs, and behavioural preconditions for later policy.
Start with the action boundary that matters most: the action an autonomous system must not take without explicit authority.
Evaluate your action boundary