Docs / Integration

Integration

LogiRoot is provider-agnostic. You wrap each AI tool call with a governance evaluation before execution and record the decision. The sidecar is the only thing your application talks to; LogiRoot does not intercept your LLM provider calls directly.

Pre-call pattern

# Pseudocode — applies to any LLM provider you use today
def call_tool(tool_name, args, context):
    decision = logiroot.evaluate(
        tool_name=tool_name,
        args=args,
        context=context,
    )

    if decision.decision == "REJECT":
        raise GovernanceRejected(decision.policy_ids, decision.receipt_id)
    if decision.decision == "ESCALATE":
        return await_human_approval(decision.receipt_id)

    return execute_tool(tool_name, args)

Compatible providers

LogiRoot governs tool-call decisions; it works regardless of which model produced them. Customers commonly use it alongside major LLM providers and self-hosted models.

AWS-native deployment

• AWS CloudTrail — governance receipts can be exported to your CloudTrail audit log alongside other AWS API calls.
• AWS IAM — role-based access for operators viewing governance dashboards.
• AWS Secrets Manager — recommended store for your LogiRoot API key.

Failure mode

LogiRoot fails closed by default. If the governance endpoint is unreachable, the SDK raises rather than silently approving. You can opt into a fail-open mode for non-critical paths via SDK configuration.

SDK

Official SDK quickstarts ship per-language. Until then, the API is small enough that a thin HTTP client works fine. See the API Reference.