Redpoint Interaction v7.x Documentation
Auto Light Dark
Redpoint Interaction v7.x Documentation
Auto Light Dark
Breadcrumbs

RedpointAI: Open-source AI agent platform

Overview

Your marketing team asks dozens of small questions a day: what campaigns are running now, why did this audience count change from last week, which offers are most successful. Your data team asks its own: which feeds failed overnight, how is data quality trending, when does the next match run. RedpointAI turns those questions into a conversation. It's a new, open-source AI agent platform that connects any large language model (LLM) provider, including OpenAI, Anthropic, and Azure OpenAI, directly to your RPI and Data Readiness Hub (DRH)/Customer Data Platform (CDP) instance, so agents can look things up, reason about your setup, and explain what they find.

What it is

RedpointAI is an open-source AI agent platform released under the Apache License 2.0. It now covers two Redpoint products, RPI and DRH/CDP, each in its own workspace with its own tools and domain knowledge. See What’s next for info on what is to come.

At its core, RedpointAI does three things:

  • Gives you a chat interface for talking to agents, with a workspace for each product.

  • Exposes RPI and DRH/CDP capabilities as standard tools that any Model Context Protocol (MCP) client can call, through two standalone MCP servers.

  • Routes each request to the right slice of those tools instead of handing the model your whole toolbox at once.

Bring your own LLM provider: there's no lock-in to a single model vendor. You choose the provider and model in your environment configuration; RedpointAI picks Azure OpenAI, then Anthropic, then OpenAI, based on which API key is present.

Why you should care

Two problems show up constantly when you connect an LLM to a real platform. The model either doesn't have the tools it needs, or it has too many and wastes time (and tokens) figuring out which one to use. RedpointAI's Skill Router solves this by grouping tools into domain-specific skills. For RPI, those cover audiences, clients, files, folders, interactions, selection rules, auth, and admin. For DRH/CDP, they cover databases, sources, feeds, runs, schedules, data quality, subject areas, aggregations, automation logs, and more. Instead of showing the model every tool on every turn, the router picks the right skill first, then hands a focused sub-agent only the tools that skill needs.

The payoff is faster responses, lower token costs, and more accurate tool selection. The model chooses among a compact skill catalog of roughly 500 tokens instead of reading about 15,000 tokens of tool definitions on every turn.

RedpointAI also ships with built-in domain expertise. Each product has an always-on foundation expert for cross-cutting basics like tenant handling and terminology. RPI also has a deeper domain expert that answers how-to, design, and strategy questions about building campaigns, covering attributes, selection rules, audiences, and interactions. That expert loads only when a question needs it, so its depth costs nothing on turns that don't.

Read-only by default. Both MCP servers ship with their write tools turned off. Agents can explore, explain, and recommend against your live instances with a low-risk footprint. Write tools can't be switched back on through configuration; enabling them requires a code change and a rebuild. In future iterations, we plan to support design capabilities across RPI functionality (rules, audiences, interactions, offers, and more).

RedpointAI runs as independent services: an orchestration server, the web app, an RPI MCP server, and a DRH/CDP MCP server. Each one restarts, scales, and fails on its own, so a problem in one doesn't take down the others. Workspace cards on the landing page show the exact reason a workspace isn't working, such as missing configuration, an authorization failure, or an unreachable server.

Security and access

RedpointAI supports single sign-on through Keycloak using OpenID Connect (OIDC), as well as native RPI logins. API keys are read only from your environment and are never stored in the application database. For anything beyond local evaluation, set AUTH_REQUIRED=true.

Who it's for

RedpointAI is a source repository built for engineers. Everything it produces is built from source with Bun.

  • Agent developers who already run an MCP client, such as Claude Desktop or Cursor, and want a working RPI or DRH/CDP tool surface to point it at.

  • Contributors who want a full local environment to build, test, and extend.

If you're not comfortable cloning and building from source, contact Redpoint Support for help evaluating RedpointAI.

Getting started

The GitHub repository walks through setup in detail, so we won't repeat it step by step here. Both paths start from the same clone:

Bash
git clone https://github.com/RedpointGlobal/redpoint-ai.git
cd redpoint-ai
bun install

  • Run the full stack. Copy .env.example to .env, add at least one LLM API key and your RPI and DRH connection details, then run bun run dev. The web app opens at http://localhost:3001. Start here even if you only want the MCP servers, since it's the quickest way to see the whole platform working.

  • Build the standalone MCP servers. From packages/mcp-rpi or packages/mcp-drh, run bun run build:all && bun run release:zip. You get a self-contained zip for Linux x64 or Windows x64 that doesn’t need Bun or Node.js to run. Extract it, fill in its .env, run the launcher, and point your MCP client at http://localhost:3002/mcp (RPI) or http://localhost:3003/mcp (DRH/CDP).

Prerequisites are Bun 1.3.0 or later and Node.js 22.6 or later. On Windows, run the stack inside WSL2. For complete prerequisites, environment variables, and troubleshooting steps, refer to docs/getting-started.md in the repo.

Licensing

RedpointAI is licensed under Apache License 2.0. Read it, fork it, extend it.

What's next

  • MCP for Redpoint Identity Studio: Building out agentic access to Identity Studio, bringing AI-powered identity resolution and graph capabilities into the platform.

  • Expanding Redpoint Interaction capabilities: Moving beyond read-only access to enable design-time actions, including creating, updating, and deleting rules, audiences, and interactions directly through the agentic platform.

  • Deeper Data Readiness Hub expertise: A curated DRH domain expert to answer how-to and design questions about data preparation and quality.

  • Broader agentic ecosystem: As each product area reaches MCP coverage, customers gain a unified, conversational interface across the full Redpoint suite, from data readiness through identity resolution to campaign execution.

Last updated: