OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Agent Frameworks

OMA (Open Multi-Agent)

OMA (Open Multi-Agent) is a self-hosted TypeScript runtime for running AI agents, explicit task pipelines, and multi-agent teams with durable approval gates and offline-verifiable run journals.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

OMA (Open Multi-Agent) — OpenViST project profile in Agent Frameworks.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use OMA when an agent workflow needs self-hosted operation, durable human approval, restartable execution, or offline verification of recorded run content.

Who it is for

  • Engineering teams building agent workflows in their own environment.
  • Organizations that need human approval before consequential agent actions.
  • Developers using cloud providers or local model servers such as Ollama, vLLM, or llama-server.
  • Teams that need run lineage and content verification outside a hosted control plane.

Documented uses

  • Orchestrate single agents, explicit task pipelines, and teams from a goal.
  • Pause plans, task dispatches, or consequential tool calls until an approval is recorded.
  • Persist checkpoints and pending approvals so runs can resume after a restart.
  • Record and verify model context, tool activity, results, and context rewrites offline.
  • Evaluate governance receipts against required roles and dependency order.
Before you choose

Requirements & limits

The repository is public and licensed under MIT. The documented setup requires Node.js 20 or newer; the repository recommends a currently maintained Node.js LTS release for production. A live provider example requires `OPENAI_API_KEY`, while the documented local demo uses no API key and makes no model request.

  • Self-hosting: OMA is described as having no hosted control plane or telemetry, with operators controlling keys, models, and runtime infrastructure.
  • Model choice: the repository names cloud providers, OpenAI-compatible endpoints, AI SDK providers, Ollama, vLLM, and llama-server as supported model environments.
  • Approval durability: suspended requests are stored beside checkpoints and bound to a SHA-256 hash of the reviewer-visible content.
  • Verification scope: offline journal verification checks byte-for-byte reproduction and reports an evicted context window as inconclusive.
  • Governance evaluation: required roles and execution order can be assessed through an execution receipt, and a run may succeed while still reporting unsatisfied.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Scaffold a local OMA starter project

Create an OMA application and run its documented deterministic local demo.

Before you start

  • Node.js 20 or newer.
  • An interactive terminal.
  • The npm package tooling used by the documented scaffold command.

Steps

  1. Run the command below.

    Documented command
    npm create oma-app@latest my-oma

  2. Select a starter and runtime when prompted.

  3. Allow the scaffold to install dependencies and run the deterministic local demo.

Expected outcome

A starter OMA application is scaffolded and its local demo exercises the scheduler, result aggregation, and offline dashboard without an API key or model request.

Guide limitations

  • The documented demo uses scripted model responses and does not demonstrate a live provider or model connection.
  • The source does not provide setup details for each supported local model environment.
Documentation 1 ↗
Source-based practical guide

Add OMA to an existing backend with checkpointed execution

Configure the documented core runtime example with a provider, an approval policy, and a durable file checkpoint.

Before you start

  • Node.js 20 or newer.
  • The @open-multi-agent/core package.
  • An OpenAI API key for the provider shown in the example.
  • An existing TypeScript backend.

Steps

  1. Install @open-multi-agent/core.

  2. Create an OpenMultiAgent instance with the documented provider and model settings.

  3. Set consequential tool calls to suspend for a human decision.

  4. Create a team and run it with a FileStore checkpoint path.

Expected outcome

The backend starts a checkpointed team run whose consequential tool calls can remain suspended pending reviewer decisions.

Guide limitations

  • The example uses an OpenAI provider and requires OPENAI_API_KEY; other providers and local servers are covered separately by the repository documentation.
  • The source does not provide a complete approval-review interface or deployment procedure in the supplied material.
Documentation 1 ↗
COMPARE OPTIONS

Compare options for Agent orchestration

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

ProjectDocumented purposePlatformAccess and requirements
OMA (Open Multi-Agent)Orchestrate single agents, explicit task pipelines, and teams from a goal.CLI, Self-hosted, ServerThe repository is public and licensed under MIT. The documented setup requires Node.js 20 or newer; the repository recommends a currently maintained Node.js LTS release for production. A live provider example requires `OPENAI_API_KEY`, while the documented local demo uses no API key and makes no model request.
AgeniteDefine typed tools with JSON Schema validation and error handling.CLI
VeldraCreate an agent with a policy, tools, skills, knowledge base and reasoning method.Self-hostedThe documented deployment is self-hosted and uses a sign-in gate after a first-run administrator account is created.
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
2 October 2026
Original website title
open-multi-agent/open-multi-agent · Official homepage title · retained in its source language
How we review sourcesSuggest a correction