agentguard-tech

Python compatibility SDK 0.15.0 for AgentGuard HTTP policy evaluation. This distribution is maintained for truthful compatibility and security fixes; it is not the normative v1 runtime contract and does not claim equivalent TypeScript feature scope, executor-owned capability separation, or a compliance outcome.

The distribution name is agentguard-tech; Python imports use agentguard. The TypeScript package is an independent release named @the-bot-club/agentguard.

Install

pip install agentguard-tech

Quick start

from agentguard import AgentGuard

guard = AgentGuard(api_key="ag_live_...")
decision = guard.evaluate(tool="shell_exec", params={"cmd": "rm -rf /"})
# → {"result": "block", "riskScore": 95, "reason": "...", "durationMs": 1.2}

evaluate() returns the camelCase keys that POST /api/v1/evaluate returns:

A kill-switch block additionally carries killSwitchActive: True. Other optional fields the API may add on a block or require_approval include agentId, approvalId, approvalUrl, suggestion, docs, alternatives, notice, warnings and pii — treat anything beyond the five keys above as optional.

With local_eval=True and a cached policy bundle, evaluate() runs in process and returns those same keys, so callers never have to branch on which path produced a decision.

Policy templates

list_templates() returns the template ids this deployment actually serves; get_template(name) and apply_template(name) take an id from that list. The set of ids is a property of the deployment, not of this package.

OpenClaw compatibility hook

OpenClawGuard is a Python-side HTTP adapter for a structural before_tool_call hook. Strict mode is required and returns a block decision on HTTP failure. strict=False is a hard startup error. The hook can be disabled or bypassed and owns no raw capability under a separate identity, so it is compatibility telemetry only.

from agentguard.integrations.openclaw import openclaw_guard

guard = openclaw_guard(
    api_key=os.environ["AGENTGUARD_API_KEY"],
    agent_id="ops-dispatch-1",
    strict=True,
)

OpenClaw-provided session, run, and tool-call IDs are correlation data only. Other integration modules in this distribution are compatibility adapters; their presence is not evidence of runtime coverage or equivalent v1 feature scope.

Telemetry

Telemetry is opt-in and off by default. Nothing is sent unless you construct the client with telemetry=True:

guard = AgentGuard(api_key=key, telemetry=True)   # opt in
guard = AgentGuard(api_key=key)                   # default: nothing sent

When you opt in, the SDK sends one anonymous ping to POST {base_url}/api/v1/telemetry per client instance, on the first evaluate() call. The payload is exactly four fields:

No API key, tool name, parameters, policy, decision, hostname, or IP-derived identifier is included, and the ping is fire-and-forget on a daemon thread — it never raises and never delays an evaluation.

Setting the environment variable AGENTGUARD_NO_TELEMETRY=1 force-disables the ping even when telemetry=True is passed.

This is separate from your own audit trail: in local_eval=True mode the client batches decision records to POST /api/v1/audit under your API key. That is your tenant's data, not usage telemetry, and it is not controlled by this setting.

Documentation

• Website — agentguard.tech
• Docs — agentguard.tech/docs/
• OpenClaw integration — agentguard.tech/openclaw/
• Source — agentguard.tech/docs/#source
• Support — hani@thebot.club

Licence

Business Source License 1.1. The LICENSE file shipped in this distribution is authoritative. © 2026 The Bot Club Pty Ltd (ABN 99 695 980 226).