AffixIO/affixio-agent-guard-py

AffixIO Agent Guard — Python mirror (Model repo)

Python mirror of the npm affixio-agent-guard policy: deterministic, fail-closed check of one tool call + local JSON receipt. Zero network, zero keys. Receipts are local simulations, not cryptographic attestations — same honest contract as the npm package.

npm install affixio-agent-guard   # Node version
python test_guard.py              # this Python mirror

from guard import check_tool_call
d = check_tool_call({"tool": "payments.create", "host": "api.example.com",
                     "amount": 120, "args": {"customer": "cust_42"}})
print(d["allowed"], d["reason"])  # True allowed by policy

Policy keys: allowedTools, allowedHosts, maxAmount, blockedSecrets, requireApprovalFor (fail-closed: denied unless you build your own approval step).

Version

• v1.0.0 — Python mirror of guard v0.1.0 logic. See CHANGELOG.md.

Links

• npm guard: https://www.npmjs.com/package/affixio-agent-guard
• Product: https://www.affix-io.com
• SDK: https://www.npmjs.com/package/affixio
• API docs: https://www.affix-io.com/docs/
• Agentic Pay Kit: https://www.affix-io.com/agent-trust/

Keywords: AI agents, guardrails, LLM security, MCP, agent verification, tool calling, local AI.