@allow-ai/sdk

Network-layer and framework-layer authorization SDK for AI agents. ALLOW intercepts outbound HTTP requests and tool calls, evaluating them against OPA/Rego policies before execution. Blocked actions are denied at the network and framework layers, ensuring agents can only perform authorized operations.

Prefer @visiq/sdk -- the unified VisIQ SDK bundles both ALLOW (authorization) and RECALL (context firewall) into a single package with one wrapper function. Install @visiq/sdk unless you specifically need ALLOW standalone.

Installation

npm install @allow-ai/sdk

Requires Node.js >= 20.

Quick Start

Network-Layer Interception

Patches globalThis.fetch and Node.js http/https modules to intercept all outbound requests and evaluate them against ALLOW policies.

import { init, cleanup } from '@allow-ai/sdk';

// Initialize network-layer interception
await init({
  apiKey: process.env.ALLOW_API_KEY!,
  agentId: 'my-agent',
});

// All outbound HTTP requests are now evaluated against ALLOW policies.
// Unauthorized requests are blocked before they reach the network.

// When finished, remove patches and flush telemetry
cleanup();

Framework-Layer Wrapping

Wraps AI framework tool objects to enforce policies at the tool-call interface, before execution begins.

import { withAllow } from '@allow-ai/sdk';

// Wrap a LangChain tool
const tool = withAllow(new SearchTool(), {
  agentId: 'research-bot',
});

// Wrap a LlamaIndex tool
const llamaTool = withAllow(new QueryEngineTool(), {
  agentId: 'research-bot',
});

Configuration

init(options) -- Network-Layer

| Option | Type | Required | Default | Description | |--------|------|----------|---------|-------------| | apiKey | string | Yes | -- | ALLOW API key | | agentId | string | Yes | -- | Unique agent identifier | | endpoint | string | No | Production URL | ALLOW backend endpoint | | mode | 'enforce' \| 'audit' \| 'off' | No | 'enforce' | Enforcement mode | | timeoutMs | number | No | 5000 | Request timeout in ms | | failBehavior | 'closed' \| 'open' | No | 'closed' | Behavior on evaluation failure | | passthroughHosts | string[] | No | [] | Hosts to skip evaluation | | enableLocalEval | boolean | No | true | Enable client-side rule evaluation | | syncIntervalMs | number | No | 30000 | Bundle sync interval in ms |

withAllow(target, options) -- Framework-Layer

| Option | Type | Required | Default | Description | |--------|------|----------|---------|-------------| | agentId | string | Yes | -- | Unique agent identifier | | apiKey | string | No | ALLOW_API_KEY env var | ALLOW API key | | endpoint | string | No | ALLOW_ENDPOINT env var | ALLOW backend endpoint | | mode | 'enforce' \| 'audit' \| 'off' | No | 'enforce' | Enforcement mode |

Supported Frameworks

| Framework | Tool Detection | Adapter | |-----------|---------------|---------| | LangChain | _call() + name | LangChainAllowTool | | LlamaIndex | call() + metadata | LlamaIndexAllowTool | | Semantic Kernel | functionInvocationFilters | AllowKernelFilter | | Generic | execute() or run() | Proxy wrapper |

Exports

// Primary API
import { init, cleanup, withAllow } from '@allow-ai/sdk';

// Advanced: OPA engine, client, config
import { AllowEngine, AllowClient, resolveConfig, RuleEngine } from '@allow-ai/sdk';

// Advanced: request inspection
import { shouldIntercept, extractRequestInfo } from '@allow-ai/sdk';

// Advanced: framework detection
import { detectToolFramework, isAllowWrapped, ALLOW_WRAPPED } from '@allow-ai/sdk';

// Adapter classes (for direct use without auto-detection)
import {
  LangChainAllowTool,
  LlamaIndexAllowTool,
  installAllowFilter,
  wrapToolWithProxy,
} from '@allow-ai/sdk';

Environment Variables

| Variable | Description | |----------|-------------| | ALLOW_API_KEY | API key (fallback when not passed programmatically) | | ALLOW_ENDPOINT | Backend endpoint URL | | ALLOW_MODE | Enforcement mode (enforce, audit, off) |

License

MIT