AIntent Documentation

AIntent is an intent alignment gate for LLM applications. It ensures that AI-generated outputs align with actual user intent before delivery.

AIntent is available as a Hosted SaaS product and as a local-first OSS runtime. Both share the same core alignment engine.

Choose a Path

Pick the setup that fits your workflow.

Hosted SaaS

The managed AIntent product. No local build required.

• Hosted dashboard for session management and analytics
• Account management and API key provisioning
• Managed REST API and hosted MCP connectivity
• ChatGPT App and Claude Web integrations

Hosted SaaS supports two practical access patterns: web surfaces and CLI-connected agent workflows.

The fastest path to a local AIntent runtime. Includes CLI and stdio MCP transport.

This OSS path uses the aintent package and binary. The hosted SaaS CLI documented elsewhere on this site uses aints.

Install globally

npm install -g aintent

Or run directly with npx

npx aintent start

What you get

• Local CLI interface (9 commands)
• stdio MCP server (5 tools)
• Local SQLite persistence
• Session export to JSON

Clone the OSS repository and build locally. Includes the local dashboard and full source access.

Clone and build

git clone https://github.com/acord25/aintent.git

cd aintent && npm install && npm run build

Run the local dashboard

cd web && npm install && npm run dev

What you get

• Everything in the npm package
• Local dashboard web app
• Editable source code
• Ability to customize and extend

What's Included

Understand what each OSS path ships.

Set up AIntent with your MCP host, CLI, or environment variables.

MCP Host Setup: Claude Desktop

Add to your Claude Desktop MCP configuration:

{
  "mcpServers": {
    "aintent": {
      "command": "npx",
      "args": ["-y", "aintent"]
    }
  }
}

MCP Host Setup: Cursor

Add to your Cursor MCP settings:

{
  "mcpServers": {
    "aintent": {
      "command": "npx",
      "args": ["-y", "aintent"]
    }
  }
}

Agent Skill Bootstrap

Install the bundled skill bootstrap for a supported agent. This does not generate Claude Desktop or Cursor MCP JSON:

These OSS bootstrap commands use aintent. The hosted SaaS CLI uses aints.

aintent init claude

Environment Variables

Features

CLI commands, MCP tools, and alignment modes available in AIntent OSS.

How It Works

AIntent runs an iterative alignment loop to discover and validate user intent before the LLM produces a final output.

Session Lifecycle

Scoring and Calibration

Each signal submitted by the LLM updates the alignment score based on slot coverage, specificity, confidence, and risk assessment. The engine calibrates domain fluency (L1 through L7) to adjust question depth and question layer while mode and risk settings determine the applicable score thresholds.

Artifacts

When a session reaches the "done" stage, the engine produces a structured alignment summary including the final score, slot states, unresolved items, risk assessments, and a recommended action.

AIntent OSS is organized into focused modules, each with a single responsibility.

Data Flow

Extension Points

• Store: Swap the default SQLite store with a custom persistence adapter.
• Export: Add custom export formats beyond JSON.
• Local API: Extend the local contract layer for dashboard integrations.

Which Path?

A quick guide to choosing the right setup.