AI sales agent send boundary
Gate a customer-facing agent action before pricing, discount language, or next-step commitments leave the company.
Krafthaus workflow app
Let low-risk agent actions proceed and route sensitive customer, financial, or irreversible actions to review before software acts. The gate returns allow, review, or block with the evidence and next owner attached.
Buyer wedge
Use this when an agent can touch customer state but not every action deserves the same autonomy. Krafthaus captures the facts, applies the bounded action policy, and gives the owner an action packet before execution.
Gate a customer-facing agent action before pricing, discount language, or next-step commitments leave the company.
Route refunds, subscription changes, and sensitive customer updates when confidence, consent, or approval is missing.
Allow low-risk reversible updates, but require review for financial, security, or irreversible actions.
A running ai_agent_action_gate_packet with intake facts, verdict material, handoff, packet hash, and Decision Record references.
Direct declarative rules for confidence, consent, human approval, value at risk, reversibility, and customer impact.
Execute the agent action, route it to human review, or block execution until the risk clears.
Executable demo
This form posts closed action facts to /api/ai-agent-action-gate. The server validates the intake,
calls Decide with a pinned direct_declarative_rulebook, and only returns an
ai_agent_action_gate_packet when the Decision Record material is present.
Decision and record
Run an action to see the decision, owner, next action, and technical record.
Application contract
AI Agent Action Gate is a software-company operating workflow that shows how Krafthaus can install Decide between a proposed autonomous action and the system that would mutate customer, revenue, or operational state.
Application Manifest v1 pins the workflow, intake contract, direct rulebook binding, record fields, action boundary, and handoff.
Rulebook rulebook-v1.schema.jsonThe production core evaluates limited declarative JSON. Customer executable rulebooks are rejected.
Sprint Start from AI agent actionSend a workflow brief to scope this boundary around the company's agent actions, approval policy, and handoff owners.