Hexgate

Runtime authorization for AI agents. On every tool call, Hexgate decides whether this user, in this role, may run this tool with these arguments — allow, deny, or require approval. For OpenAI Agents, LangChain, Google ADK, Pydantic AI, or a native runtime.

Website · Docs

---

What is Hexgate?

Hexgate is two things that move together:

• hexgate — the SDK. A Python runtime that gates every tool call through a typed Decision (allow / deny / approval-required), resolving the caller's role at call time to apply that role's rules. Wrap an existing agent without rewriting it, or build one natively — every decision is traced and audited with the caller's identity. See supported frameworks →
• The Hexgate platform (optional) — a FastAPI control plane + React dashboard for editing policy in a browser, minting per-project tokens, watching live decisions stream from a serving agent, and shipping signed WASM policy bundles to production. Available as Hexgate Cloud (hosted — set one env var, no infra) or self-hosted.

You can use the SDK three ways: local (YAML/bundle on disk, no platform), Hexgate Cloud (remote enforcement + audit — just set HEXGATE_API_KEY), or self-hosted (run the control plane yourself). HEXGATE_API_URL defaults to https://app.hexgate.ai, so remote enforcement is one env var away.

Quickstart

pip install hexgate

See it enforce — no API keys. Save a policy that gives two roles different limits on the same refund_order tool:

# policy.yaml
version: 1
roles:
  support:                                     # small USD refunds only
    default_policy: { mode: deny }
    tools:
      refund_order:
        mode: allow
        constraints:
          - args.amount <= 50
          - args.currency == "USD"
  billing:                                     # larger refunds, major currencies
    default_policy: { mode: deny }
    tools:
      refund_order:
        mode: allow
        constraints:
          - args.amount <= 500
          - args.currency in ["USD", "EUR"]

hexgate policy test decides the same $400 refund for each role offline — no model, no keys:

hexgate policy test policy.yaml --role support \
    --tool refund_order --args '{"amount": 400, "currency": "USD"}'
# ✗ DENY · support → refund_order({"amount": 400, "currency": "USD"})
#   reason: Policy on "refund_order" denied: constraint failed — args.amount <= 50

hexgate policy test policy.yaml --role billing \
    --tool refund_order --args '{"amount": 400, "currency": "USD"}'
# ✓ ALLOW · billing → refund_order({"amount": 400, "currency": "USD"})

Same tool, same request — the caller's role and the arguments decide, enforced outside the model. The full quickstart → puts this in front of a live agent.

Documentation

Full documentation lives at docs.hexgate.ai.

Build an agent	Define tools directly with create_agent, or wrap an existing framework agent.
Framework adapters	OpenAI Agents, LangChain/LangGraph, Google ADK, Pydantic AI.
Policy	YAML shape, constraints, WASM bundles, signing, local override.
Caller context + roles	Per-request identity, role resolution, biscuit attenuation via HexgateContext.
CLI	chat, serve, register, policy.
MCP servers	Wrap any Model Context Protocol server as policy-enforced tools.
Hexgate Cloud (hosted)	Remote policy enforcement + audit with zero infra — get a key, set one env var.
Platform (self-hosted)	Run the control plane, dashboard, ClickHouse audit, and Resend email yourself.

Development

Contributor setup, make targets, and the test suites are documented in Development & testing. The short version:

make install-dev     # uv sync --extra dev (first time only)
make check           # lint + fmt-check + test (matches CI)

License

MIT — see LICENSE.

---

If Hexgate looks useful, give it a ⭐ on GitHub — it helps more than you'd think. Built by Hexamind.