AI

AI in 3forge Enterprise¶

3forge Enterprise integrates large language models (LLMs) directly into a live deployment. Developers can use their preferred LLM to query, build, and operate Dashboards in a running AmiCore instance.

This page introduces the 3 ways Enterprise exposes AI and how to enable them.

Overview¶

There are 3 complementary AI surfaces. They differ on 2 points: whether Enterprise calls the model or an outside client does, and whether the model is given Enterprise operations to call.

Operating 3forge over MCP

The canonical page an AI agent reads to operate a running instance over MCP: which surface exposes which tools, and the shared transport rules.

Operating over MCP
Ask AI Chatbot

An in-browser assistant inside the Web component. Build, query, and debug a Dashboard through natural-language conversation.

Ask AI Chatbot
AI Agent Access (MCP)

Let an external AI client, e.g: Claude, connect to a running AmiCore instance and operate it through tools.

MCP Server
AI in AmiScript

Call a model programmatically from a Web callback with session.promptAi to classify, summarize, or extract over live data.

Scripting with AI

Using an AI coding tool?

To drive a running instance from Claude Code, Codex, or GitHub Copilot with a ready-made set of skills, agents, and commands, install the 3forge MCP plugin for AI coding tools. It connects to the MCP server described here.

The Three Surfaces¶

Surface	Runs in	Operated by	Use Case
Ask AI Chatbot	The Web browser client	An interactive user in a chat window	Building, querying, and debugging a deployment through conversation
MCP Server	AmiCore JVM, as a REST endpoint (`/3forge_rest/mcp`)	An external AI client (any MCP host)	Letting an outside agent inspect and operate a running AmiCore instance
`session.promptAi`	Inside a Web AmiScript callback	Your own Layout logic	Calling a model programmatically over live, filtered data

The Ask AI chatbot and session.promptAi are configured on the Web component.

The MCP server is a plugin that can run on any component (Center, Web, or Relay).

Choosing a Model Provider¶

Every AI surface talks to a provider, the backend that hosts the model.

Enterprise natively ships with 4 built-in providers:

Provider id	Backend
`anthropic`	Anthropic (Claude)
`openai`	OpenAI
`gemini`	Google Gemini
`xai`	xAI (Grok)

You choose which providers are active in the Web component's configuration. Each provider reads its own credentials and model list from ami.web.ai.<id>.* properties. See the Ask AI Chatbot page for the full setup.

Note

For custom providers or to support a locally-hosted model, you can write your own provider. See Custom AI Providers.

Prerequisites¶

A Web component running (for the chatbot and session.promptAi), or any Enterprise component (for the MCP server).
Access to at least one model provider: either an API key for a built-in provider, or a custom provider that reaches your own gateway.
AI is off by default on every surface. Nothing calls out to a model until you explicitly enable it.

Observability¶

All chatbot and session.promptAi activity is written to AmiAi.log (one JSON object per line) so any AI interaction can be reconstructed and audited. See here for more information.