Motus Limen · Enterprise AI gateway
In developmentControl enterprise access to AI.
Keep governed requests verifiable.
Limen provides one OpenAI-compatible endpoint for local and external models. Before each model call, it identifies the requester, applies company policy and selects an approved route. Motus records the decision and the provider outcome.
The route exists. Passage depends on the rule at the gate.
Leonardo da Vinci, Codex Atlanticus, late 15th century · plate
What happens before an AI model is called.
Limen turns access policy into an executable Motus graph. The gateway evaluates the request before the provider call, then either blocks it, flags it for review or routes it to the approved local or external model.
Requesters
Motus Limen
Approved destinations
Identity is supplied by the organisation's OIDC/JWT system or LiteLLM virtual keys. The prompt content does not enter the decision trace; Limen records its hash.
Administrators decide who can use what.
Policies and gateway configuration define which principals may use which models, how request classes are routed, when access is denied or held for review, and which budgets and rate limits apply.
Identity
Uses the principal asserted by OIDC/JWT or a LiteLLM virtual key.
Policy
Evaluates groups, request class and requested model through a versioned Motus graph.
Data
Classifies request metadata and records a prompt hash, not the prompt text.
Routing
Selects the approved logical model, provider and local or external destination.
Review
Stops a request marked for review before a provider is called and queues it — there is no automatic release.
Costs
Uses LiteLLM keys, budgets and rate limits as the enforcement layer.
Control access.
Record governed use.
Limen is the policy decision point inside the LiteLLM enforcement layer. The policy is versioned by its Motus graph fingerprint, and the decision trace is created before the provider call rather than reconstructed later from gateway logs.
Conventional gateway
Records that a request was routed.
Motus Limen
Records the inputs read, policy clause, selected route and observed provider effect.
Adoption
Change the endpoint. Keep the applications.
Existing OpenAI-compatible clients can point to Limen instead of a provider, gaining a governed route without rewriting their AI integration.
Direct provider access
https://api.provider.com/v1Governed through Limen
https://limen.company.ai/v1Scope
What Limen does — and what it does not.
Limen controls and records the AI interactions routed through it. It supports governance and compliance processes, but it does not replace organisational policies, risk assessment or legal responsibility.
Beyond Limen
Limen cannot record AI tools that bypass it. Evidence is verified with the published Motus validator and, for the anchors, with OpenTimestamps' own tooling — not ours, so whoever verifies does not have to trust Limen or Vitruvyan. The current pre-MVP operates in local assurance mode: independent execution continuity is a declared gap until a future external witness lands.
What Limen does not do
Limen governs the API calls that pass through it. A browser opening a consumer AI site on a personal account does not pass through it, and no API gateway can see that traffic — closing that door takes three separate layers.
The network
Closing the wrong door is the network's job: corporate proxy or DNS blocks consumer AI domains, browser policy on managed devices enforces site lists, and DLP/CASB stops a classified upload where it is deployed.
Limen
Opening the right one is Limen's job: the company's own chat, routed through Limen, is the sanctioned alternative — same experience, evidence on every answer.
Rules and people
Written policy, training, and a real alternative close the rest. Whoever bypasses a governed path does so against a clear rule, not for lack of one.
Limen is the governed access point; blocking consumer sites is the network's job; together they close the case of the document uploaded to a personal ChatGPT.
One controlled access point
for enterprise AI.
Give people, applications and agents access to approved AI without losing control of models, data, costs or evidence.