Overview
Praeon turns AI write-access on safely by governing typed plans, enforcing policy via the Praeon Kernel, and producing audit-ready traces.
What is Praeon?
Praeon is compliance-grade action audit infrastructure for AI agents. It combines typed action plans (PlanLang), a reference kernel that enforces least-privilege execution, and replayable traces accepted by internal risk and external auditors.
Quickstart
Choose the path that matches your operating model.
SMB Pilot
- Connect orchestrator to emit PlanLang.
- Run Praeon Kernel in managed sandbox.
- Validate adapters with provided policies.
Enterprise Pilot
- Integrate identity (Okta/Entra) for approvals.
- Route traces to SIEM (Splunk/Datadog).
- Execute 50-plan audit drill. See audit checklist.
PlanLang v0.1
PlanLang is a human-auditable JSON schema capturing actors, capabilities, constraints, policies, and approvals. Download the canonical schema: planlang-v0.1.json.
Example plans live in docs/examples.
Schema (JSON)
See planlang-v0.1.json for the full specification. Core types cover actor, capabilities, constraints, steps, approvals, policyRef, and effects.
Types & Effects
Each action step enumerates effects for least-privilege evaluation. Use typed scopes (e.g. domain:example.com) and budgets for deterministic guardrails.
Policies
Policies can be authored in CEL or Rego. Example: approve outbound email when budget <= $200 and domain is allowlisted. Store policies under policy:// URIs referenced in PlanLang documents.
Kernel
The Praeon Kernel performs static analysis, dry-run execution, adapter mediation, and semantic replay. Deploy managed, on-prem, or sovereign.
Execution Model (Determinism = Semantic Equivalence)
Praeon targets semantic equivalence between live execution and replay. Adapter success criteria are codified per integration; optional bit-identical mode is available when the adapter supports it.
Approvals & Dry-Run
Combine automated policy approvals with human checkpoints. Dry-run mode exercises adapters with synthetic credentials to validate side effects before production approval.
Traces (Operational vs Audit)
Operational traces remain encrypted with customer KMS/HSM. Audit traces are redacted, signed, and exportable to auditors. Replay any incident with semantic success criteria.
Security
No tokens appear in traces. Secrets are fetched at execution time via KMS/HSM adapters. Runtime sandboxes enforce egress policies; vendor adapters ship as reproducible builds with Sigstore/OCI digests.
Adapters
Praeon ships adapters for Email, Slack/Teams, Salesforce, Jira/ServiceNow, GitHub/GitLab, and Payments (Stripe/Razorpay). Each adapter exposes typed capabilities with least-privilege scopes.
Enterprise Guides
- Okta/Entra SSO
- Splunk/Datadog Integration
- DPDP/GDPR Retention
- Big-4 Audit Drill Checklist
API Reference
Praeon JavaScript SDK: @praeon/sdk. REST and CLI endpoints expose plan submission, approval workflows, and trace export.
@praeon/sdk & CLI
Install via npm (npm install @praeon/sdk) and pip (pip install praeon). CLI supports plan validate, kernel preflight, and trace export.