jp
74ff5e6b98
Previous: quarantine fired whenever sqlite_mtime - hnsw_mtime exceeded the (lowered, in #1173) 300s threshold. ChromaDB 1.5.x flushes HNSW asynchronously and a clean shutdown does not force-flush, so the on- disk HNSW is *always* meaningfully older than chroma.sqlite3 — that's the steady state, not corruption. Quarantine renamed valid HNSW segments on every cold-start, chromadb created empty replacements, vector recall went to 0/N until rebuild. Confirmed in production on the disks daemon journal, 2026-04-26 06:56:45: three of three HNSW segments quarantined on cold-start with 538-557s mtime gaps (post-clean-shutdown flush lag), leaving a 151,478-drawer palace with vector_ranked=0. Drift directories at *.drift-20260426-065645/ each contained a complete 253MB data_level0.bin plus 18MB index_metadata.pickle — clearly healthy indexes, renamed by the false-positive heuristic. Fix: two-stage gate. 1. mtime gate (existing) — gap > stale_seconds is necessary. 2. integrity gate (new) — sniff index_metadata.pickle for chromadb's expected protocol/terminator bytes (PROTO 0x80 head, STOP 0x2e tail) and a non-trivial size, WITHOUT deserializing the file. Healthy segment with mtime drift → keep in place; truncated / zero-filled / partial-flush → quarantine. Format-sniff is deliberately non-deserializing — pickle deserialization can execute arbitrary code, and the PROTO+STOP byte presence + size floor is sufficient to distinguish a complete chromadb write from truncation, zero-fill, or a partial flush during process kill. Real load failures (the rare case where the bytes look right but chromadb fails to load) still surface to palace-daemon's _auto_repair, which calls quarantine_stale_hnsw directly on observed HNSW errors and bypasses this gate. The cold-start gate from 70c4bc6 (row 24) remains as a perf optimization — even with the integrity check, repeating the sniff on every reconnect is unnecessary work — but its load-bearing role is now covered by this deeper fix. 4 new tests in test_backends.py: - test_quarantine_stale_hnsw_renames_corrupt_segment (drift + bad meta) - test_quarantine_stale_hnsw_leaves_healthy_segment_with_drift_alone (drift + valid meta — the production case at 06:24) - test_quarantine_stale_hnsw_leaves_segment_without_metadata_alone (fresh / never-flushed, no meta file) - test_quarantine_stale_hnsw_renames_truncated_metadata (under-floor size, partial-flush shape) Existing test_quarantine_stale_hnsw_renames_drifted_segment renamed to renames_corrupt_segment with explicit corrupt meta_bytes — the old "renames any drift" contract is gone. Suite 1366/1366 pass. Coordinated cross-repo with palace-daemon's auto-repair-on-startup workaround (separate agent's commit ed3a892). With this fork-side fix the auto-repair becomes belt-and-suspenders; the structural cause of empty-HNSW-on-restart is addressed at the quarantine layer. CLAUDE.md row 26 + README fork-change-queue row + test count 1363→1366. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
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.