What’s New in MemHouse v0.4.0
MemHouse v0.4.0 brings dedicated architectural support for personal agents and ephemeral multi-agent swarms.
1. External State Memory for Ephemeral Swarms
When building multi-agent systems with LangGraph, CrewAI, or AutoGen, worker agents spawn for short subtasks and terminate. MemHouse serves as the external state engine that survives subagent process death:
- Subagents post observations to MemHouse before shutting down.
- Subsequent agents instantly retrieve distilled, governed facts without context window bloat.
2. Experimental Model Context Protocol (MCP) Endpoint
Packaged Claude Desktop, Cursor, and OpenClaw integration is coming soon. The current server exposes an experimental /mcp endpoint; use the documented HTTP API for supported integrations.
3. Personal Agent Connectors (Coming Soon)
- OpenClaw CLI: Planned connector for terminal workflows and project configuration memory.
- Hermes Agent: Planned Python memory provider for local companion state.
4. Automated Release Artifacts
This post is served directly from the Astro-compatible source file src/content/blog/release-v0-4-0-ephemeral-swarms-and-mcp.mdx in the official repository.
What Is Supported Today
The supported integration surface described by this release is the governed HTTP API, including observation ingest and projection-backed context retrieval. The /mcp endpoint is experimental. Packaged framework adapters and personal-agent connectors remain roadmap items until their installation and compatibility instructions are published.
The 10,000-ingests-per-second figure discussed elsewhere on this site is also an engineering target, not a v0.4.0 result. See the benchmark methodology article for the evidence required before that target can become a measured claim.