Multi-agent shared projects: team brain support

[bm-clawd]

Context

Multiple OpenClaw agents should be able to share a BM project as a "team brain" while also having their own private projects. This is the killer feature for agent orchestration — shared context without agent-to-agent messaging.

Inspired by real demand: https://x.com/ericosiu/status/2021249104710598785 (61K views — guy rebuilt this with symlinks and markdown files)

What this enables

• Agent A writes an insight → Agent B finds it via semantic search
• Shared priorities, KPIs, feedback loops across agents
• Each agent also has private memory (their own project)
• Humans participate too — same graph, Obsidian or cloud app

Implementation

• Support multiple projects in config (primary + shared)
• project: agent's own memory (read/write)
• sharedProjects: list of team projects (read/write or read-only)
• Search spans all configured projects
• Write operations go to the primary project by default
• bm_write gets optional project param to target shared project

Config example

{
  "basic-memory": {
    "config": {
      "project": "claw-private",
      "sharedProjects": [
        { "name": "team-brain", "access": "read-write" },
        { "name": "company-wiki", "access": "read-only" }
      ]
    }
  }
}

Depends on

• Cloud mode: support remote BM Cloud projects via API key auth #3 (cloud mode for cross-machine access)
• Basic Memory supporting cross-project search