open-multi-agent: a goal-driven multi-agent orchestration framework for TypeScript backends that dynamically generates task DAGs

What it solves

Open Multi-Agent (OMA) solves the problem of rigid, hand-wired agent workflows. Instead of requiring developers to manually define every node and edge in a graph, OMA allows users to provide a high-level goal, which the system then automatically decomposes into a dynamic task graph at runtime.

How it works

OMA uses a coordinator agent that takes a goal and breaks it down into a Directed Acyclic Graph (DAG) of tasks. A deterministic scheduler then executes these tasks, parallelizing independent steps to improve efficiency. The framework is built for TypeScript backends and can be embedded directly into Node.js applications. It supports a lightweight core with opt-in dependencies for various model providers (like Anthropic, OpenAI, Gemini, and Bedrock) and tools like the Vercel AI SDK.

Who it’s for

TypeScript developers and backend engineers who need to build multi-agent systems where the plan of execution needs to adapt to the specific goal provided, rather than following a pre-defined, static workflow.

Highlights

• Goal-Driven Orchestration: Automatically generates a task DAG from a goal at runtime.
• TypeScript-Native: Designed specifically for Node.js backends with a lean runtime.
• Observability: Includes a post-run dashboard for replaying runs, tracing spans, and analyzing token usage.
• Resilience: Supports checkpointing and resuming runs to survive crashes andes restarts.
• Flexible Model Routing: Allows matching models by phase, agent, role, or priority.
• Extensible Tooling: Supports custom tools, a filesystem sandbox, and the Model Context Protocol (MCP).