Open Multi-Agent Framework Released

Open Multi-Agent (OMA) is an AI agent orchestration framework where the Coordinator generates the task DAG at execution time simply by passing a Goal, without the user needing to design a workflow graph in advance.

Key features are as follows:

• Dynamic Planning and Control: While providing the flexibility of plans being created at execution time, it manages the plan so it doesn't go out of control through Approval, Constraints, and Checkpoint Resume features.
• Flexible Execution Model: Provides three methods — runTeam() for driving an entire team, runAgent() for running a single agent, and runTasks() for performing a defined pipeline.
• Support for Diverse Environments: Cloud and local models can be used in combination, and collaboration with external agents such as Claude Code and Gemini CLI is also possible within shared memory and budget.
• Observation and Evaluation: Through the Run Viewer, the task DAG, agent status, token usage, and more can be visually replayed, and it supports monitoring via OpenTelemetry as well as quality Evaluation (Eval) based on execution records.

Unlike simple LLM toolkits or static graph-based frameworks, this framework is optimized for the orchestration layer that requires complex dependency relationships, approvals, and recovery steps among multiple agents.