One wallet, many agents, one brain

Spawning sub-agents in OpenClaw is easy. Getting them to share what they learn is the annoying part. MemoClaw ties memory to your wallet, so any agent that signs with that wallet writes to the same space. Scouts can drop discoveries, fixers can log outcomes, and the lead agent recalls it all without syncing files or stuffing prompts.

Multi-agent wiring diagram

scout agent (wallet W) ---- store --> MemoClaw namespace:research
                       \
fixer agent (wallet W) ---- store --> MemoClaw namespace:fixes
                         \
 orchestrator (wallet W) -- recall --> research + fixes

• Scout: Crawls docs, summarizes findings, and stores them with tags like source:docs or topic:auth.
• Fixer: Closes incidents, then logs results with tags such as incident:123 or status:done.
• Lead agent: Runs memoclaw recall --tags topic:auth --namespace research before assigning more work, so it reuses what the scout already wrote.

Because the wallet is the identity, setup ends after you fund the same x402 address. Everyone shares the same semantic memory and you never bolt on a side database.

Implementation steps

1. Fund the wallet once. One Base wallet covers the lead agent and every child agent. No keys to rotate.
2. Name your spaces. Give each sub-agent a namespace such as research, fixes, or handoff so recalls stay targeted.
3. Agree on tags. Use tags like incident:<id> or customer:<slug> so every recall filter is predictable.
4. Batch the scout's notes. Let the scout buffer each doc, then call store --batch to cut paid calls.
5. Recall with filters. The lead agent can merge namespaces on the fly: memoclaw recall --query "latest auth changes" --namespaces research,fixes.

Example handoff flow

1. Scout finishes crawling the docs and writes:

memoclaw store "Auth callback now requires PKCE for partners" \
  --tags topic:auth,release:2026-03,source:docs \
  --namespace research \
  --importance 0.7

1. Lead agent spins up a fixer. Before writing a script, it fetches the scout's findings:

memoclaw recall --query "auth partner changes" --namespace research --tags topic:auth

1. Fixer ships the change and locks the outcome for future shifts:

memoclaw store "PKCE enforcement shipped to staging" \
  --tags incident:482,status:done \
  --namespace fixes \
  --importance 0.6

No message bus, no shared database, and no exploding prompt windows. The wallet binds every agent together. MemoClaw stores the collective brain, and OpenClaw's spawn API just works. Once you try it, you'll never go back to emailing JSON between agents.