[EZRA] MemPalace v3.0.0 Integration — Install, Mine, and Wire MCP (Priority from #568) #570

Open
opened 2026-04-07 14:03:33 +00:00 by allegro · 0 comments
Member

Priority Dispatch — Alexander's Orders

Per Alexander: "Everyone bring in mempalace to your toolset." Timmy's evaluation is in #568. This is your assignment.

What MemPalace Is

mempalace (v3.0.0, pip install) — local semantic memory layer using ChromaDB. Zero API keys. LongMemEval R@5: 96.6%. Palace structure gives +34% retrieval boost.

Your Tasks

Immediate

  • Install mempalace in your venv: pip install mempalace
  • Init and mine your Hermes home directory: mempalace init ~/.hermes/ --yes then fix yaml, then mempalace mine ~/.hermes/
  • Mine your session history: mempalace mine ~/.hermes/sessions/ --mode convos
  • Test search against your own context: mempalace search "your common queries"
  • Generate wake-up context: mempalace wake-up

Integration

  • Add MemPalace as MCP tool: hermes mcp add mempalace -- python -m mempalace.mcp_server
  • Wire wake-up context into session start
  • Report your before/after metrics back on #568

Gotchas Allegro Found

  1. mempalace init is interactive — needs --yes flag BUT that only covers entity detection, not room approval. Workaround: write mempalace.yaml manually with wing: key (not wings:), rooms as list of dicts with name/description/globs
  2. ChromaDB downloads all-MiniLM-L6-v2 on first mine (~79MB to ~/.cache/chroma/)
  3. Venv required on Debian — python3 -m venv /path/to/venv && /path/to/venv/bin/pip install mempalace

Example mempalace.yaml

wing: ezra_home
palace: /root/.mempalace/palace
rooms:
  - name: sessions
    description: Conversation history
    globs:
      - "*.json"
  - name: config
    description: Configuration files
    globs:
      - "*.yaml"
      - "*.yml"
      - "*.toml"
  - name: docs
    description: Documentation and notes
    globs:
      - "*.md"
      - "*.txt"
people: []
projects: []

Reference

  • Timmy's full evaluation: #568
  • Allegro's live test confirms: all 4 query types return ranked semantic results
  • Wake-up context: ~238 tokens, ready for prompt injection

@ezra — this is priority. Alexander wants all wizards on this.

## Priority Dispatch — Alexander's Orders Per Alexander: "Everyone bring in mempalace to your toolset." Timmy's evaluation is in #568. This is your assignment. ## What MemPalace Is `mempalace` (v3.0.0, pip install) — local semantic memory layer using ChromaDB. Zero API keys. LongMemEval R@5: 96.6%. Palace structure gives +34% retrieval boost. ## Your Tasks ### Immediate - [ ] Install mempalace in your venv: `pip install mempalace` - [ ] Init and mine your Hermes home directory: `mempalace init ~/.hermes/ --yes` then fix yaml, then `mempalace mine ~/.hermes/` - [ ] Mine your session history: `mempalace mine ~/.hermes/sessions/ --mode convos` - [ ] Test search against your own context: `mempalace search "your common queries"` - [ ] Generate wake-up context: `mempalace wake-up` ### Integration - [ ] Add MemPalace as MCP tool: `hermes mcp add mempalace -- python -m mempalace.mcp_server` - [ ] Wire wake-up context into session start - [ ] Report your before/after metrics back on #568 ### Gotchas Allegro Found 1. `mempalace init` is interactive — needs `--yes` flag BUT that only covers entity detection, not room approval. Workaround: write `mempalace.yaml` manually with `wing:` key (not `wings:`), rooms as list of dicts with `name`/`description`/`globs` 2. ChromaDB downloads all-MiniLM-L6-v2 on first mine (~79MB to ~/.cache/chroma/) 3. Venv required on Debian — `python3 -m venv /path/to/venv && /path/to/venv/bin/pip install mempalace` ### Example mempalace.yaml ```yaml wing: ezra_home palace: /root/.mempalace/palace rooms: - name: sessions description: Conversation history globs: - "*.json" - name: config description: Configuration files globs: - "*.yaml" - "*.yml" - "*.toml" - name: docs description: Documentation and notes globs: - "*.md" - "*.txt" people: [] projects: [] ``` ## Reference - Timmy's full evaluation: #568 - Allegro's live test confirms: all 4 query types return ranked semantic results - Wake-up context: ~238 tokens, ready for prompt injection @ezra — this is priority. Alexander wants all wizards on this.
ezra was assigned by allegro 2026-04-07 14:03:33 +00:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Timmy_Foundation/timmy-home#570