@noirstack/openaiguardrails

Official JavaScript/TypeScript client for the Open AI Guardrails runtime API.

Install

npm install @noirstack/openaiguardrails

Quick start

import { createClient } from '@noirstack/openaiguardrails';

const guardrails = createClient({
  baseUrl: 'https://openaiguardrails.org',
  token: process.env.OPENAIGUARDRAILS_API_TOKEN,
});

const policy = await guardrails.getPolicy('prod-safety-alpha');
const skills = await guardrails.listSkills();

Pass a bearer token for protected routes (policy manager, blog review, admin publish). Public catalog and health endpoints work without a token against the hosted API.

API surface

Core

| Method | HTTP | Auth | |--------|------|------| | health() | GET /health | — | | getOpenApi() | GET /openapi.json | — | | listEntries() | GET /api/entries | — | | listDiscussions() | GET /api/discussions | — |

Policy manager & OPA

| Method | HTTP | Auth | |--------|------|------| | listPolicies() | GET /api/policies | optional* | | getPolicy(policyId) | GET /v1/policy/:id | — | | togglePolicyControl(policyId, key, value) | POST /api/policies/:id/toggles | policy manager | | runPolicyAction(policyId, action, payload) | POST /api/policies/:id/actions | policy manager | | exportAudit(policyId) | GET /api/policies/:id/audit-export | policy manager | | listOpaImports() | GET /api/opa/imports | policy manager | | importOpaPolicy(payload) | POST /api/opa/imports | policy manager | | simulateOpaImport(importId, input) | POST /api/opa/imports/:id/simulate | policy manager | | exportOpaImport(importId, target) | GET .../export/:target | policy manager | | publishOpaImport(importId, payload) | POST .../publish | policy manager |

*Hosted production may require Auth0 when POLICY_MANAGER_REQUIRE_AUTH=true.

Agent Skills platform

| Method | HTTP | |--------|------| | listSkills() | GET /api/skills | | getSkill(skillId) | GET /api/skills/:id | | listProfessions() | GET /api/professions | | getProfession(professionId) | GET /api/professions/:id | | getMcpRegistry() | GET /api/mcp/registry | | getMcpTool(skillId) | GET /api/mcp/tools/:skillId |

Blog workflow

| Method | HTTP | Auth | |--------|------|------| | listBlogPendingPosts() | GET /api/blog/pending-posts | read:blog_submissions or admin | | submitBlogPost(payload) | POST /api/blog/pending-posts | submit:blog or user/admin | | getBlogReview(reviewId) | GET /api/blog/reviews/:id | read:blog_submissions or admin | | approveBlogReview(reviewId, payload) | POST .../approve | manage:blog or admin | | rejectBlogReview(reviewId, payload) | POST .../reject | manage:blog or admin | | publishBlogReview(reviewId, payload) | POST .../publish | manage:blog or admin | | publishBlogPost(payload) | POST /api/blog/publish | manage:blog or admin |

Operations

| Method | HTTP | Auth | |--------|------|------| | getSmtpStatus() | GET /api/ops/smtp-status | manage:system or admin |

Use request(path, options) for forward-compatible access to new OpenAPI routes.

Errors

Failed responses throw OpenAIGuardrailsError with status, statusText, and parsed body.

Requirements

• Node.js 18+ (native fetch), or pass fetch in options
• Zero runtime dependencies

Links

• Documentation
• API reference
• Packages

Changelog

See CHANGELOG.md.