​
Deterministic Security for AI Agents

AI can request actions. Dmint deterministically decides whether those actions may execute.
dmint is an in-process and gateway-level security boundary between AI models and privileged execution environments. It intercepts AI tool calls and deterministically evaluates them against declarative policies: ALLOW, DENY, or APPROVAL_REQUIRED.

​
Key Pillars

Deterministic Enforcement

Runtime authorization is governed strictly by declarative code and policy. An LLM never makes runtime authorization decisions.

Exact Request Binding

Approvals are bound to the cryptographic SHA-256 fingerprint of the canonicalized RFC 8785 payload. Approving update(id=1) cannot authorize update(id=2).

Transparent MCP Gateway

Intercept Model Context Protocol (stdio & streamable HTTP) tools in Cursor, Claude Code, and Antigravity with zero agent code changes.

Human Approval UI

Review, inspect argument diffs, and approve sensitive tool calls using the local web dashboard backed by Core-owned SQLite.

​
Quick Start

You can evaluate Dmint on any tool in under 5 minutes:
1

Install Dmint Suite

2

Author Policy

Describe rules in access.md and compile to verified policy.json:
3

Enforce Tool Calls

Wrap an MCP server with dmint-mcp or decorate Python functions with @dmint.protected.
4

Review Approvals

Launch the local dashboard to review and approve sensitive actions:
Ready to run a live test? Jump directly to the complete 5-Minute Quickstart.

​
The Dmint Ecosystem

dmint (Core)

v1.0.1 — Frozen deterministic authorization engine, SQLite approval store, and Ed25519 signing authority.

dmint-cli

v1.1.0 — CLI tooling for interactive policy authoring, offline schema validation, and skill installation.

dmint-mcp

v1.0.0 — Multi-integration gateway, SSRF defenses, downstream stdio/HTTP proxies, and retry loops.

dmint-dashboard

v1.0.0 — Local human approval web UI with 5-second polling and zero direct SQL access.

dmint-skills

v1.0.0 — Coding agent skills that guide LLMs to draft least-privilege security policies.

dmint-examples

Ready-to-run scenarios covering database mutations, shell operations, and multi-agent governance.