Managed Tools Infrastructure

The core identity of Privane is serving as the Tool Execution Infrastructure for local AI. We abstract away the pain of building connectors, handling OAuth, and managing Model Context Protocols (MCP).

The 1-Click Philosophy

Instead of setting up webhooks, registering OAuth apps, and fighting with scopes, Privane allows you to inject fully-authenticated tools directly into your local LLM with a single line of code.

import { PrivaneTools } from '@privane/tools';
 
const tools = new PrivaneTools({ apiKey: process.env.PRIVANE_API_KEY });
 
// 1-Click Usable
const slack = tools.get("slack");
const github = tools.get("github");

The Tool Registry Matrix

Bridging local agents to cloud ecosystems requires a normalized interface. Privane’s Managed Tool Gateway offers native, pre-authenticated, rate-limit-aware tools ready to run instantly:

Cloud Tools (Managed Execution)

These tools execute on the highly available api.privane.dev execution infrastructure. We handle session storage, access token refreshes, rate-limiting, and connector maintenance.

Here is a 1-click execution loop utilizing the gateway to post updates directly:

import { PrivaneTools } from '@privane/tools';
 
const tools = new PrivaneTools({ apiKey: process.env.PRIVANE_API_KEY });
const slack = tools.get("slack");
 
// Normalize execution: Auth, retries, and network routing are fully managed
await slack.sendMessage({
  channel: "#engineering-updates",
  text: "Deployment completed successfully. Reasoning computed locally on device, dispatched securely via Privane Tool Gateway."
});

Local Tools (Sandboxed Execution)

These tools execute directly on the user’s host machine via the Privane CLI runtime, avoiding the cloud entirely:

• filesystem: Sandboxed read/write access to restricted work directories.
• sqlite: Local database query execution.
• shell: Security-restricted terminal command runner.

The Privane Tool Schema

All tools, whether local or cloud, conform to the unified Privane Tool Schema (which acts as an abstraction layer over OpenAI Tools and MCP).

type ToolDefinition = {
  name: string;
  description: string;
  permissions: string[];
  inputSchema: JSONSchema;
  outputSchema: JSONSchema;
  pricing?: {
    executions: number;
  };
};