KYE™ Governed Prompts™

Agentic Governance™ with Authority Finality™ at the prompt boundary.

A prompt that invokes a tool, generates code that will be committed, or produces a plan that drives the next act is an authority act — it spends authority a principal delegated, within a purpose and a scope. KYE™ Governed Prompts™ govern it as one: emitted before transmission, decided allow / tool-restrict / escalate / refuse, and sealed into a signed, replay-derivable evidence chain. Prompt-as-authority-act, not prompt-as-input.

See the GovernedUI surfaceIntegrationsvs prompt firewalls

KYE™ Governed Prompts™, point by point

Demo data

You learn: The page's points, one at a time. You can: Step through them.

  1. Step 1 of 5

    KYE Protocol™

    LangChain wrapper

    wraps LLM / ChatModel calls.

  2. Step 2 of 5

    KYE Protocol™

    OpenAI SDK middleware

    intercepts chat.completions.

  3. Step 3 of 5

    KYE Protocol™

    Anthropic SDK middleware

    intercepts messages.create.

  4. Step 4 of 5

    KYE Protocol™

    Claude Agent SDK hook

    a pre-prompt / PreToolUse hook.

  5. Step 5 of 5

    KYE Protocol™

    MCP intercept

    every tool-call prompt crossing the gateway.

How this widget is built
Demonstrates
KYE Protocol™
Components
  • The points this page makes
Flow
Each point in order, as the page sets it out. Architecture diagram
Used on

Two envelopes — before transmission, and after the decision.

Your agents emit two signed envelopes per prompt: one before the model sees it, one after the verdict. A developer wires both with one middleware call.

Before a prompt reaches the model, the emitting agent emits a governance envelope binding the agent, the purpose it serves, the scope it acts within, the target model, the prompt class, a prompt hash (the body never leaves the boundary), and the attached PII categories. The governance decision then emits an outcome envelope — allowed, refused, escalated, or tool-restricted — sealing the act into the attestation chain.

Integrations — wherever the prompt is transmitted.

  • LangChain wrapper — wraps LLM / ChatModel calls.
  • OpenAI SDK middleware — intercepts chat.completions.
  • Anthropic SDK middleware — intercepts messages.create.
  • Claude Agent SDK hook — a pre-prompt / PreToolUse hook.
  • MCP intercept — every tool-call prompt crossing the gateway.

Versus prompt firewalls — complementary, but a different altitude.

A firewall asks whether the prompt is safe. KYE™ asks whether the agent holds the authority to send it.

Prompt firewalls (Lakera, Robust Intelligence, Cisco AI Defense, NVIDIA NeMo Guardrails) govern the prompt as input — scanning for injection, jailbreaks, and toxicity. KYE™ Governed Prompts™ govern the prompt as an authority act — bound to a purpose, a scope, and a replay-derivable evidence chain. A firewall verdict can feed the governance decision; it does not replace it.

Govern every prompt your agents send.

KYE™ Governed Prompts™ is a governance projection of KYE Protocol™, surfaced through KYE™ GovernedUI™. Every operator decision emits a signed Evidence Pack™.

GovernedUIKYE™ MCP Server™