Release v3.5.0 — promote develop to main

[igorls]

Promote develop → main for the 3.5.0 release

Releases publish only from main (per docs/RELEASING.md). This promotes the 3.5.0 bump (merged via #1853) plus all develop work accumulated since v3.4.1 — 33 PRs.

Version

All six sources at 3.5.0 (version-guard.yml green on the bump commit). Tool count refreshed 34 → 35 (delete_by_source + checkpoint).

Headline changes since v3.4.1

Features

• Opt-in local write daemon for queued/serialized palace writes (feat: opt-in local daemon for queued MemPalace writes #1826)
• Opt-in HTTP transport for the MCP server — loopback-default, Host/Origin DNS-rebinding guard + optional MEMPALACE_MCP_HTTP_TOKEN (Long-lived stdio server stops answering tools/list after extended uptime (small frames still work) #1801, feat(mcp): add opt-in HTTP transport #1806)
• mempalace_checkpoint batch-save tool (feat: add mempalace_checkpoint batch save tool #1851)
• mempalace_delete_by_source bulk cleanup — dry-run-by-default, purges drawers and their AAAK index entries (Bug: Benchmark/test data loaded into user wing pollutes semantic search #1722, feat(mcp): add mempalace_delete_by_source bulk-cleanup tool (#1722) #1729)
• Optional source_file filter for mempalace_search (Expose source_file filtering in mempalace_search #1815, feat(search): add an optional source_file filter to mempalace_search (#1815) #1817)
• New transcript parsers: Continue.dev (feat(normalize): add Continue.dev session parser #731), Gemini CLI / AI Studio (feat: add Gemini CLI / AI Studio session import support #204), Pi agent (feat: add Pi agent JSONL session normalizer #169)
• Miner coverage: C#/.NET, PHP, Swift/Kotlin, Java (feat(miner): add PHP ecosystem file extensions #1819, feat(miner): add support for Swift and Kotlin file extensions #1368, fix: detect Java project manifests #1720)
• Final mine on Claude plugin SessionEnd (Add a SessionEnd hook for a deterministic mine on clean exit (plugin only wires Stop + PreCompact) #1814, fix(claude-plugin): run final mine on SessionEnd #1820)

Performance

• Overview/status + graph_stats answered from the SQLite aggregate to kill large-palace timeouts (mempalace_status times out on large palaces under Claude Desktop — full-metadata pagination instead of the existing sqlite aggregation path #1748, MCP overview tools time out on large palaces #1379); embedder ORT thread cap (Background mempalace mine pins 400–500 % CPU — ORT intra_op pool ignores OMP env vars #1068); SQL-pushdown pagination for sqlite_exact/pgvector + single-scroll Qdrant metadata (fix(backends): push sqlite_exact get(limit, offset) pagination into SQL (#1841) #1842, fix(pgvector): push get(limit, offset) pagination into SQL (#1830) #1840, fix(backends): single-scroll bulk metadata fetch for Qdrant; bump scroll page size (#1796) #1832)

Bug fixes

• pgvector tolerates surrogate/NUL transcript bytes (pgvector backend aborts the whole mine run on a lone surrogate (U+D800-U+DFFF) in transcript content #1833, pgvector backend aborts the whole mine run on a NUL (0x00) byte in transcript content #1829); percent-encoded SQLite RO URIs; stale-HNSW divergence → SQLite fallback (fix(chroma): route stale hnsw divergence to sqlite fallback #1822); diverged-index recovery points at repair --mode from-sqlite not a data-losing re-mine (repair (legacy) and MCP server cannot recover from a ChromaDB compactor failure, though --mode from-sqlite can #1843, fix(repair): point index-read failures to repair --mode from-sqlite (#1843) #1847, fix: point diverged-index recovery at from-sqlite, not re-mine (#1843) #1849); MCP refuses a second writer (fix(mcp): refuse second writer for same palace #1823); Windows hook miner uses CREATE_NO_WINDOW (fix: use CREATE_NO_WINDOW so Windows hook miner spawns don't flash a console (#1783) #1848)

CI

• test-windows now retries only the transient ChromaDB HNSW compaction flake (--only-rerun); Linux/macOS keep zero reruns so real regressions stay loud (ci(test-windows): retry the transient ChromaDB HNSW compaction flake #1854)

Full detail in CHANGELOG.md under ## [3.5.0].

After merge

Draft a GitHub Release targeting main, tag v3.5.0 → triggers publish.yml (PyPI Trusted Publishing, manual approval on the pypi environment).

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[gemini-code-assist]

Code Review

This pull request introduces version 3.5.0 of MemPalace, featuring an opt-in local daemon for queued writes, an opt-in HTTP transport for the MCP server, and new MCP tools like mempalace_checkpoint and mempalace_delete_by_source. It also adds a source_file filter for searches, a final mine on Claude plugin session end, performance optimizations using SQLite aggregates, and expanded language support. Feedback on these changes highlights two issues: first, _purge_source_closets incorrectly calls .get("ids") on a GetResult object, which will cause an AttributeError on modern backends; second, the hardcoded _CLI_MAX_CHUNKS_PER_FILE_DEFAULT ignores the MEMPALACE_MAX_CHUNKS_PER_FILE environment variable.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

The function _purge_source_closets calls .get("ids") directly on the result of closets_col.get(...). However, BaseCollection.get() returns a GetResult object (which is a dataclass/namedtuple) rather than a dict on modern backends (such as sqlite_exact, pgvector, and qdrant). This will raise an AttributeError when trying to delete by source on any of these backends, silently failing the closet purge because the exception is caught and returns 0 closets deleted.

Suggested change

	ids = closets_col.get(where={"source_file": source_file}, include=[]).get("ids") or []
	batch = closets_col.get(where={"source_file": source_file}, include=[])
	ids = batch.ids if hasattr(batch, "ids") else (batch.get("ids") or [])

[gemini-code-assist]

The constant _CLI_MAX_CHUNKS_PER_FILE_DEFAULT is hardcoded to 50_000. However, the help text for --max-chunks-per-file (line 1675) states that the default is _CLI_MAX_CHUNKS_PER_FILE_DEFAULT (or MEMPALACE_MAX_CHUNKS_PER_FILE). Because argparse uses the hardcoded default value, any environment variable MEMPALACE_MAX_CHUNKS_PER_FILE is completely ignored when running via the CLI. It should be initialized from the environment variable. When parsing this environment variable, ensure to return a safe default (such as 0 to disable) if the value is unparseable, rather than falling back to a default value.

def _get_max_chunks_default():
    val = os.environ.get("MEMPALACE_MAX_CHUNKS_PER_FILE")
    if not val:
        return 50000
    try:
        return int(val)
    except ValueError:
        return 0

_CLI_MAX_CHUNKS_PER_FILE_DEFAULT = _get_max_chunks_default()

References

1. When parsing environment variables for configuration, return a safe default (e.g., 0.0 to disable) if the value is unparseable, rather than falling back to a default value.