Build AI workflows
in plain english
A team of specialists, not a generic chatbot. Describe what you need and Modular9 designs the workflow, places the specialist nodes on a canvas, wires the edges, and runs it natively as a DAG.
The Execution Model
True DAG execution. Parallel Fan-out and Fan-in.
Modular9 is not a linear chain. The engine resolves the canvas as a directed acyclic graph. Independent branches run in parallel, and multiple parents fan in to a single child.
Fan-in keys by edge ID, ensuring when N parents converge on one node, all outputs arrive distinctly instead of clobbering one another. That single property enables real multi-agent tournament brackets and adversarial verification.
Core Capabilities
Logic-aware specialists.
Modifier Injection Stack
The core differentiator. Every specialist accepts stackable custom skills that change how it thinks without replacing its base expertise.
Skills inject as ordered behavioural blocks between the base plugin prompt and the incoming content. Apply "OWASP Top 10 Deep Focus" to a Code Review node, stack several, and drag to reorder.
AI Native Workflow
Automation platforms bolt AI onto rule-based pipelines. Modular9 is AI-native from the ground up and gives you full control over the execution graph.
Every step is a specialist, powered by a COVENANT.md contract that explicitly declares its reasoning strategy and typed input/output schema.
Under The Hood
Asynchronous Edge Architecture
Vercel Web Tier
Validates triggers, enqueues BullMQ jobs, and subscribes to SSE Redis channels, returning instantly without serverless timeouts.
Render Worker Node
Consumes jobs and executes the DAG with no timeout. Supports inline execution for local dev.
Provider Resilient
BYOK Gateway. AES-256-GCM encrypted keys. Deterministic error classification across OpenAI, Anthropic, and NIM.