ModulexAI

Context

Most teams stall before AI does any real work, not on the model, but on the setup: an account and an API key for every tool, glue code to wire them, and one credential per tool that's never enough. ModulexAI makes connectivity the product. It ships pre-connected to 200+ integrations, brings managed credentials for premium tools, and runs many accounts of the same tool at once, exposed as a chat assistant that acts, a visual workflow canvas, and an API, all on one engine.

What I built

• Batteries-included connectivity: 200+ pre-wired integrations available to every assistant and workflow out of the box, with OAuth 2.0 flows so teams skip the per-tool account-and-key setup entirely.
• ModuleX Key, a managed-credential layer that brokers access to premium tools (Tavily, Firecrawl, ElevenLabs, and a list growing weekly) through a metered credit system, so users get the premium stack with no signups or keys to hunt.
• Multi-credential support: many accounts of the same integration, usable simultaneously inside one assistant or workflow, for example several Gmail accounts through a single assistant.
• Runtime compiler that translates workflow definitions into executable LangGraph statecharts, with dynamic TypedDict state generation, a `{{node.path}}` DSL for cross-node reference resolution, and parallel FOREACH fan-out via the Send API.
• Composer, a multi-agent AI copilot (DeepAgents) that plans, generates, and mutates the workflow graph from natural language on a visual canvas (xyflow / React Flow), streamed via SSE over Redis Pub/Sub.
• Execution layer supporting 12+ node types (LLM, RAG, Conditional, Loop, Guardrails, Interrupt) with per-node retry, credit tracking, and PostgreSQL-backed checkpointing; hybrid keyword and semantic RAG over pgvector grounds assistants in a team's data.
• Distributed real-time collaboration engine (Socket.io + Redis Pub/Sub) with node-level locking and JSON-patch state sync (fast-json-patch), so multiple people edit one graph live.
• Credential vault with per-credential encryption via HKDF-SHA256 key derivation; one engine exposes every workflow as a chat assistant, a canvas, and a callable REST endpoint, with zero-dependency TypeScript and Python SDKs for embedding ModuleX in third-party products.

Outcome

Stage

Live in production

Recognition

Product Hunt #6 Product of the Day (2026)

Pre-connected integrations

200+

Surfaces, one engine

Chat · Canvas · API