Skip to content

LIM · the interpretation layer

The Large Intent Model.

LIM is the proprietary interpretation layer inside the Intended platform. It interprets every agent action — software or physical — into a confidence-scored, governable intent, then maps it to your business processes.

02 · Capabilities

Auditable interpretation, not opaque inference.

01 · Output shape

Confidence-scored interpretation

Every LIM output carries a confidence score, a rationale, assumed context, and any missing fields the model surfaced. The interpretation is auditable, not opaque.

02 · Multi-hypothesis

Ranked candidate interpretations

When the input is ambiguous, LIM returns a ranked list of candidate interpretations so authority decisions and human reviewers can see what alternatives were considered.

03 · Process map

Translation to your workflow

Customer-defined business processes (Procure-to-Pay, Order-to-Cash, or anything bespoke) become first-class context. Every interpreted intent gets mapped to a process step with its own confidence.

04 · Conformance

Sequence scoring

LIM tracks the recent stream of agent actions and scores how well it conforms to the declared workflow. Drift surfaces as a continuous signal, not a post-hoc audit finding.

05 · Fail-closed

Patent-pending failure mode

When LIM cannot confidently interpret an action, the result fails closed — escalated or blocked rather than guessed. Invention Disclosure 002 covers the cryptographic chain that makes this verifiable.

06 · Audit-native

Full reasoning preserved

Every LIM compilation emits structured audit events with the classification, alternatives, process-step match, and conformance score. Auditors get the full reasoning chain, not just the decision.

03 · Backends

Three pluggable backends. Operator chooses.

rules
deterministic · 1ms · zero data egress
llm-api
multi-hypothesis · highest accuracy
local-model
air-gapped · GPU-resident
domain packs
fine-tuned per vertical

04 · Next

See LIM compile a real intent.

Watch the full ceremony — eight acts in which one autonomous action is interpreted, evaluated, and signed in your browser.

LIM — Large Intent Model | Intended