OpenMeld is an MIT-licensed, TypeScript collaboration system that gives people and AI agents a shared Space for context, tasks, and parallel work, and it is built for teams and individuals who already run coding agents such as Codex, Claude Code, Grok, OpenCode, Antigravity, and Cursor CLI.

What it is

OpenMeld describes itself as the open human-agents collaboration system. A Space holds people, agents, conversations, tasks, files, and folders in one place, so work continues from wherever anyone left off. It ships with the Center Agent, a hosted AI teammate available at app.openmeld.ai with no local installation, and it also connects the agents that already run on a member's own computer through the OpenMeld CLI. The project lives in the AI agent tooling ecosystem, and its topics list marks that placement directly: mcp, agent-client-protocol, agent-skills, agent-orchestration, and ai-agents. Agent tools covered by name include Codex, Claude Code, Grok, Grok Bot, OpenCode, Antigravity, and Cursor.

The problem it addresses is context loss between people and their agents. Team chat sits in one product, agent sessions sit in another, and task state and files sit somewhere else, so handoffs depend on copying context by hand. OpenMeld replaces that split with a single shared Space that carries chat, tasks for people and agents, per-Space files and folders, a team knowledge base, and memory across chats. Members keep their existing agent accounts and subscriptions, and the README states plainly that a computer running a Local Agent must stay connected while that agent works.

Key capabilities

• Hosted Center Agent that a workspace can use immediately, with no local installation, as described in the quick start.
• Local Agent bound to a chosen computer, agent tool, and project folder, so existing agents and subscriptions stay in use.
• Built-in team chat alongside tasks for both people and agents.
• Required task approval, so work can be reviewed before it is accepted.
• Files and folders per Space, access to local files, previous file versions, a team knowledge base, and memory across chats.
• Public sharing links for a Space, plus hosted cloud agents and a free personal cloud computer, which the README notes does not include free model usage or unlimited compute.
• Schedules for running tasks on a schedule, saved instructions called Skills, and 60+ apps and plugins.

Who uses it and how

• Teams already paying for or running Codex, Claude Code, Grok, OpenCode, Antigravity, or Cursor CLI use OpenMeld to bring those agents into one shared Space rather than one member's terminal.
• New users start entirely in the browser at app.openmeld.ai, create a workspace or join a team, open the Center Agent, and hand it a small task with no installation step.
• Developers keep work on their own machine and register a Local Agent per computer, agent tool, and project folder, accepting that the machine stays connected while the agent runs.
• Teams that need review gates enable task approval so agent output is checked before it lands.
• Groups that need to share outward use public sharing links, and teams that onboard members lean on the team knowledge base.

Getting started

Open OpenMeld at app.openmeld.ai and create a workspace or join your team, or install the CLI with npm install -g openmeld@latest and run openmeld setup, using a supported Node.js version. Developing from source is documented in docs/development.md.

How it compares

The README carries a capability comparison against Raft, Multica, Paperclip, Dust, Viktor, and OpenHands, reviewed on October 13, 2026. On the axes that table records, OpenMeld marks an MIT-licensed core where Raft, Multica, and Viktor mark no, and marks built-in team chat and hosted cloud agents where Multica and Paperclip mark no. Cells the project could not verify are marked with a question mark, and the table states it compares documented capabilities rather than measured speed or task success.

When to use it — and when not to

OpenMeld is in active development, so interfaces and commands can change, and the self-hosting status document describes the current deployment limits, which anyone planning to self-host should read first. The MIT licence covers the core but not every third-party asset or hosted service, and the free cloud computer excludes free model usage and unlimited compute. Teams that need frozen, stable interfaces, or that want local agents to keep running while a laptop is closed, should look elsewhere.

npm install -g openmeld@latest
openmeld setup