Add Brain Memory integration example#44
Conversation
Signed-off-by: onur karali <onurkarali@outlook.com>
Signed-off-by: onur karali <onurkarali@outlook.com>
|
Verification update: the preset has now been exercised against a live NemoClaw sandbox (v0.0.73, OpenShell 0.0.71, default docker image) — the open TODO from the original submission is resolved in b855b45. Probe matrix (from inside the sandbox):
Also verified: the preset loads cleanly ("Policy version N loaded"), re-applies identically after a sandbox One honest remaining limit, now documented in the example's Verification Status: the in-sandbox OAuth login flow can't be exercised on stock NemoClaw yet because v0.0.73 pins OpenClaw 2026.5.27, which predates the |
|
@onurkarali Thanks for this contribution. Our goal with NemoClaw is to show examples of new agents being deployed in OpenShell. This PR is showing us a way to connect an existing Hermes or OpenClaw with BrainMemory, which is not our goal here. I really enjoyed reading through the library and I believe there is a place for this integration to happen, but it should be happening at the agent harness level. I would highly recommend an integration with Hermes or OpenClaw (or both!). That would have more impact and would allow users to choose this as an option for memory. Thank you again. |
Description
Adds
examples/brain-memory/— an integration example that gives a NemoClaw-managed OpenClaw agent persistent, cross-session memory via the hosted Brain Memory remote MCP server.NemoClaw sandboxes run with deny-by-default egress, so OpenClaw's zero-install remote-MCP path (
openclaw mcp add brain-memory --url https://mcp.brainmemory.ai/mcp --auth oauth) fails out of the box: both the MCP JSON-RPC calls and the OAuth login flow are denied. This example ships the OpenShell network-policy preset that allowlists exactly what the integration needs, plus a setup guide.What's included
examples/brain-memory/README.md— guide covering: apply the policy preset → register the MCP server + OAuth login → optional deeper integration via theopenclaw-brain-memorymemory-slot plugin, with verification steps, a durability note (sandbox is disposable; memory lives in Brain Cloud), and troubleshootingexamples/brain-memory/policies/brainmemory-policy.yaml— egress policy preset fornemoclaw <sandbox> policy-add --from-file, openingPOST /mcp(MCP JSON-RPC) andGET /**(OAuth discovery/login) onmcp.brainmemory.ai:443, with commented-out optional endpoints for syncing a sandbox-local brain via Brain Cloud or a private git remoteVerification status (up-front honesty)
binaries:paths carry explicitVERIFYcomments because they depend on the sandbox image, and the example README states this plainly in a "Verification Status" section. Happy to iterate with a maintainer who has a sandbox handy.Verification
python scripts/check_license_headers.py --check examples/brain-memory/policies/brainmemory-policy.yaml— the added YAML carries an SPDX header and passes. (Repo-wide--checkcurrently reports pre-existing misses underexamples/retail-assistant/, which this PR does not touch.)git diff --checkyaml.safe_load); no shell or Python scripts are added, so there are nobash -n/py_compiletargetsRelease And Compliance
.envfiles, private certificates, snapshots, or token caches are included.THIRD-PARTY-NOTICES— none needed: this PR vendors no third-party code; the optional npm plugin is user-installed and only documented, not distributed here.git commit -s).Authors
Onur Karalı — Omelas / Brain Memory