How to Build Multi-Agent Workflows in OpenClaw

OpenClaw lets you split complex jobs across multiple specialized agents. For example, you might have one agent scraping data while another analyzes it and a third formats the report. These agents need a central place to hand off work, which is where shared storage comes in.

Since OpenClaw agents run locally, they use ClawHub skills to manage files. But without a shared workspace, they end up overwriting each other's work or creating duplicate files.

Fastio gives your agents a shared workspace where they can store files safely. We include file locking to stop conflicts and an Intelligence mode that indexes everything so agents can find what they need instantly.

You'll often see this used for research pipelines like scraping, summarizing, and reporting or complex code reviews where different agents check specific parts of a pull request.

This approach lets you tackle much bigger projects than a single agent could handle alone.

First, grab a Business Trial account at fast.io/storage-for-agents. Create an API key in Settings > Devices & Agents > API Keys, or with POST /current/user/auth/key/.

Point OpenClaw at the Fastio MCP server. Use Streamable HTTP at https://mcp.fast.io/mcp, or https://mcp.fast.io/mcp/key when the client sends a Bearer token. Legacy SSE is at https://mcp.fast.io/sse. You can also install the ClawHub skill in the next section.

Seed the shared workspace by importing a file the pipeline will use:

{"jsonrpc":"2.0","id":1,"method":"tools/call",
 "params":{"name":"upload","arguments":{"action":"web-import","url":"https://example.com/report.pdf",
 "profile_type":"workspace","profile_id":"1234567890123456789"}}}

You can also create an organization with POST /current/org/create/ and a workspace with POST /current/org/{org_id}/create/workspace/. Workspace intelligence indexes uploads so Ripley, the built-in RAG agent, can answer from those files.

Invite other agents as workspace members with POST /current/workspace/{workspace_id}/members/{email_or_user_id}/.

Start OpenClaw Multi-Agent Workflows

Agents collaborate in shared intelligent workspaces with file locks, Ripley, and branded shares.

Start 14-Day Trial

The Fastio ClawHub skill connects your agents to the Fastio MCP server so they can manage workspaces.

Install it with a single command:

npx clawhub@latest install dbalve/fast-io

There's no configuration file to mess with. The first time you use it, it'll pop open a browser to authenticate, and the session stays active after that.

Named mode tools cover uploading, searching, sharing, Coordination Rooms, activity, and Ripley. Agents call them the same way they call any other MCP tool.

Try it out by telling your agent: "Create a Fastio workspace called test-pipeline and upload this file." It will handle the authentication, create the space, and upload the file for you.

File locks are essential for preventing conflicts. One agent locks a file, does its work, and then releases it. Any other agent trying to write that file has to wait its turn.

Acquire a lock with an authenticated call:

POST https://api.fast.io/current/workspace/{workspace_id}/storage/{node_id}/lock/
Authorization: Bearer {api_key}

Keep the lock alive with POST /current/workspace/{workspace_id}/storage/{node_id}/lock/heartbeat/. When the work is done, release it:

DELETE https://api.fast.io/current/workspace/{workspace_id}/storage/{node_id}/lock/
Authorization: Bearer {api_key}

Here's the standard pattern:

1. Agent A: lock data.csv
2. Process data
3. Release the lock
4. Signal Agent B in a Coordination Room

{"jsonrpc":"2.0","id":1,"method":"tools/call",
 "params":{"name":"room","arguments":{"action":"post"}}}

Agent B can wait on that room, then lock the next file and continue the pipeline. You can also see which files are locked right in the web UI.

Intelligence mode indexes your files so agents can query the workspace through Ripley. Use the MCP ai tool (ask). It returns a cited, read-only answer and requires profile_type.

{"jsonrpc":"2.0","id":1,"method":"tools/call",
 "params":{"name":"ai","arguments":{"action":"ask","profile_type":"workspace","profile_id":"1234567890123456789"}}}

For example, asking "Summarize risks in Q1 reports" returns an answer with citations pointing to the exact pages.

This lets agents share knowledge without needing to duplicate context. Semantic search (GET /current/workspace/{workspace_id}/storage/search/) finds concepts like "contract indemnity" even if those exact words aren't used.

Agents can invite humans to workspaces, too. You can use the web UI to preview files and leave comments.

Comments can be anchored to specific parts of an image or timestamps in a video.

For handoffs, invite the person as a workspace member with POST /current/workspace/{workspace_id}/members/{email_or_user_id}/. The agent keeps working in the same workspace. For people outside the org, create a branded Send, Receive, or Exchange share with POST /current/workspace/{workspace_id}/create/share/.

Agents watch for new files with GET /current/activity/poll/{entityId}?wait=95&lastactivity={timestamp} or GET /current/events/search/. A Coordination Room wait call is the MCP version of that handoff.

Define clear tool contracts and fallback behavior so agents fail safely when dependencies are unavailable. This improves reliability in production workflows.