jp
0c38deaab5
Add a helper that renames HNSW segment directories whose `data_level0.bin` is significantly older than `chroma.sqlite3`. Drift between the on-disk HNSW graph and the live embeddings table is the root cause of a segfault class where the Rust graph-walk dereferences dangling neighbor pointers for entries in the metadata segment that no longer exist in the HNSW index, crashing in a background thread on `count()` or `query()`. Issue #823 describes the same drift as a silent-staleness symptom (semantic search returns stale results after `add_drawer` because `data_level0.bin` lags the sqlite metadata under the default `sync_threshold=1000`). Under heavier load or after an interrupted write, the same drift can escalate from "silent stale results" to "SIGSEGV on next open," which is the failure mode observed at neo-cortex-mcp#2 (chromadb 1.5.5, Python 3.12) and acknowledged at chroma-core/chroma#2594. On one 135K-drawer palace where `index_metadata.pickle` claimed 137,813 elements against 135,464 rows in sqlite (2,349-entry drift), fresh Python processes crashed in `col.count()` 17/20 times; after renaming the segment dir out of the way and letting ChromaDB rebuild lazily, the same 20-run check went to 0 crashes. The recovery path #823 suggests (export / recreate / reimport) is heavy — it re-embeds every drawer. This helper is lighter: rename the segment dir so ChromaDB reopens without it, and the indexer rebuilds lazily on the next write. The original directory is renamed (not deleted) so the operator can recover if the heuristic misfires. If `chroma.sqlite3` is more than `stale_seconds` (default 3600) newer than the segment's `data_level0.bin`, the segment is considered suspect. One hour is deliberately conservative — normal HNSW flush cadence is seconds to minutes, so an hour of drift implies a crashed mid-write, not routine lag. - Additive: exposes `quarantine_stale_hnsw(palace_path, stale_seconds)` as a helper. Not wired into `_client()` / startup on this PR — the goal is to land the primitive first so operators and higher layers can opt in. A follow-up could call it automatically on palace open behind an env var or config flag. - Closes #823 by giving operators a first-class recovery path without having to install `chromadb-ops` or re-mine. Four new tests in `tests/test_backends.py`: - renames drifted segment, preserves original files under `.drift-TS` suffix - leaves fresh segments alone - no-op on missing palace path / missing `chroma.sqlite3` - skips already-quarantined (`.drift-` suffixed) directories `pytest tests/test_backends.py` → 11 passed. `ruff check` / `ruff format --check` — clean.
Caution
Scam alert. The only official sources for MemPalace are this GitHub repository, the PyPI package, and the docs site at mempalaceofficial.com. Any other domain — including
mempalace.tech— is an impostor and may distribute malware. Details and timeline: docs/HISTORY.md.
MemPalace
Local-first AI memory. Verbatim storage, pluggable backend, 96.6% R@5 raw on LongMemEval — zero API calls.
What it is
MemPalace stores your conversation history as verbatim text and retrieves it with semantic search. It does not summarize, extract, or paraphrase. The index is structured — people and projects become wings, topics become rooms, and original content lives in drawers — so searches can be scoped rather than run against a flat corpus.
The retrieval layer is pluggable. The current default is ChromaDB; the
interface is defined in mempalace/backends/base.py
and alternative backends can be dropped in without touching the rest of
the system.
Nothing leaves your machine unless you opt in.
Architecture, concepts, and mining flows: mempalaceofficial.com/concepts/the-palace.
Install
pip install mempalace
mempalace init ~/projects/myapp
Quickstart
# Mine content into the palace
mempalace mine ~/projects/myapp # project files
mempalace mine ~/.claude/projects/ --mode convos # Claude Code sessions (scope with --wing per project)
# Search
mempalace search "why did we switch to GraphQL"
# Load context for a new session
mempalace wake-up
For Claude Code, Gemini CLI, MCP-compatible tools, and local models, see mempalaceofficial.com/guide/getting-started.
Benchmarks
All numbers below are reproducible from this repository with the commands
in benchmarks/BENCHMARKS.md. Full
per-question result files are committed under benchmarks/results_*.
LongMemEval — retrieval recall (R@5, 500 questions):
| Mode | R@5 | LLM required |
|---|---|---|
| Raw (semantic search, no heuristics, no LLM) | 96.6% | None |
| Hybrid v4, held-out 450q (tuned on 50 dev, not seen during training) | 98.4% | None |
| Hybrid v4 + LLM rerank (full 500) | ≥99% | Any capable model |
The raw 96.6% requires no API key, no cloud, and no LLM at any stage. The hybrid pipeline adds keyword boosting, temporal-proximity boosting, and preference-pattern extraction; the held-out 98.4% is the honest generalisable figure.
The rerank pipeline promotes the best candidate out of the top-20
retrieved sessions using an LLM reader. It works with any reasonably
capable model — we have reproduced it with Claude Haiku, Claude Sonnet,
and minimax-m2.7 via Ollama Cloud (no Anthropic dependency). The gap
between raw and reranked is model-agnostic; we do not headline a "100%"
number because the last 0.6% was reached by inspecting specific wrong
answers, which benchmarks/BENCHMARKS.md flags as teaching to the test.
Other benchmarks (full results in benchmarks/BENCHMARKS.md):
| Benchmark | Metric | Score | Notes |
|---|---|---|---|
| LoCoMo (session, top-10, no rerank) | R@10 | 60.3% | 1,986 questions |
| LoCoMo (hybrid v5, top-10, no rerank) | R@10 | 88.9% | Same set |
| ConvoMem (all categories, 250 items) | Avg recall | 92.9% | 50 per category |
| MemBench (ACL 2025, 8,500 items) | R@5 | 80.3% | All categories |
We deliberately do not include a side-by-side comparison against Mem0, Mastra, Hindsight, Supermemory, or Zep. Those projects publish different metrics on different splits, and placing retrieval recall next to end-to-end QA accuracy is not an honest comparison. See each project's own research page for their published numbers.
Reproducing every result:
git clone https://github.com/MemPalace/mempalace.git
cd mempalace
pip install -e ".[dev]"
# see benchmarks/README.md for dataset download commands
python benchmarks/longmemeval_bench.py /path/to/longmemeval_s_cleaned.json
Knowledge graph
MemPalace includes a temporal entity-relationship graph with validity windows — add, query, invalidate, timeline — backed by local SQLite. Usage and tool reference: mempalaceofficial.com/concepts/knowledge-graph.
MCP server
29 MCP tools cover palace reads/writes, knowledge-graph operations, cross-wing navigation, drawer management, and agent diaries. Installation and the full tool list: mempalaceofficial.com/reference/mcp-tools.
Agents
Each specialist agent gets its own wing and diary in the palace.
Discoverable at runtime via mempalace_list_agents — no bloat in your
system prompt:
mempalaceofficial.com/concepts/agents.
Auto-save hooks
Two Claude Code hooks save periodically and before context compression: mempalaceofficial.com/guide/hooks.
Requirements
- Python 3.9+
- A vector-store backend (ChromaDB by default)
- ~300 MB disk for the default embedding model
No API key is required for the core benchmark path.
Docs
- Getting started → mempalaceofficial.com/guide/getting-started
- CLI reference → mempalaceofficial.com/reference/cli
- Python API → mempalaceofficial.com/reference/python-api
- Full benchmark methodology → benchmarks/BENCHMARKS.md
- Release notes → CHANGELOG.md
- Corrections and public notices → docs/HISTORY.md
Contributing
PRs welcome. See CONTRIBUTING.md.
License
MIT — see LICENSE.