AI-agent control-plane pilot
Evaluate AI actions before your integration executes them.
MAX-AI Guard demonstrates deterministic evaluation for document access, tool calls, external communications, workflow changes, and sensitive outputs. The public dashboard does not intercept outside systems; an agent must integrate the API and enforce its result before acting.
- Proposed action
- SEND_EXTERNAL_EMAIL
- Data involved
- Sensitive donor update
- Policy matched
- External communication + donor data
- Decision
- REQUIRES APPROVAL
- Reviewer
- Program Director
- Receipt
- receipt_demo_approval generated after approval
DENY
The pilot policy denies shell access; a real caller must honor that result before execution.
REQUIRES APPROVAL
The pilot policy routes external communication containing donor, grant, or program data to review.
ALLOW
The pilot policy allows a low-risk internal read and returns the canonical request context.
Competitive wedge
Authorization belongs in the execution path.
The pilot returns a machine-readable evaluation, matched policies, safer alternatives, approval routes, and hash-recorded receipts. Signed single-use grants remain disabled on this public demo until isolated production identities and durable replay protection are explicitly activated.
Pre-action evaluation
The pilot models ALLOW, DENY, REQUIRES_APPROVAL, REDACT, and QUARANTINE decisions before a caller executes a tool.
Policy-backed approvals
Demo scenarios show how sensitive emails, exports, and public posts can pause with the matched rule attached.
Hash-recorded receipts
Pilot decisions produce Markdown and JSON receipt artifacts with hashes, evidence metadata, and audit context.
Agent integration pattern
The API and SDK demonstrate the control point an OpenAI, Anthropic, LangChain, Vercel AI SDK, or custom agent must call.
Secret and data controls
Credential-like content, regulated personal data, and sensitive data classes trigger deterministic guardrails.
Pilot traceability
The demo workspace makes actions, denials, approvals, policy edits, receipts, and integration events inspectable.
Built for a narrow, urgent category
The product focus is clear: pre-action policy, approvals, and audit receipts for autonomous AI systems. That gives security, operations, and executive buyers one concrete control plane to evaluate.