LLM agents can act, but they cannot prove why an action was allowed. URGE is a deterministic governance gate: wrap any agent action in a formal policy check and get back an auditable verdict with a full reasoning trace.
▶ Try the live demo Runs entirely in your browser. No install, no sign-up.
Then add it to a project:
cargo add urge
Apache-2.0 · seven crates on crates.io
A verdict — permitted or denied — with a confidence score, the formal notation the decision was expressed in, and citations for every step that produced it. The same policy evaluated twice gives the same answer, which is the point: a governance decision you cannot reproduce is not a governance decision.
| Paradigm | Operators | Governance role |
|---|---|---|
| Boolean | ∧ ∨ ¬ → ↔ ⊕ | Base layer, all systems |
| Modal (S5) | □ ◇ | Access across system states |
| Epistemic | K B C | Agent knowledge verification |
| Deontic | O P F | Obligations and permissions |
| Temporal (LTL) | G F X U | Deadline enforcement |
| Fuzzy | μ ⊓ ⊔ | Thresholds and uncertainty |
| Paraconsistent | Belnap 4-valued | Contradiction without explosion |
| Probabilistic | P(·) | Bayesian confidence — planned, not implemented |
Seven are implemented and cross-validated against each other. The eighth is on the roadmap and is listed here so the table is not read as a promise.
cargo add urge — the facade crate, or take a narrower one