Summary

Closes #590.

IndexEngine.index_path_stream and GET /api/index/stream were missing two contract elements that the non-stream index_path / POST /api/index pair already had:

1. namespace parameter — silently dropped. Folder-mode UI / CLI / MCP callers that streamed an index ignored the user's namespace selection.
2. Error aggregation — _index_file returns result["errors"] per file (engine.py:606, 628, 684, 752); the stream loop summed only chunk counters (754-757) and never propagated errors. The complete event lacked an errors field, so partial-failure runs appeared successful in the UI.

Changes

• Engine (packages/memtomem/src/memtomem/indexing/engine.py:707-784):
  • Add namespace: str | None = None to index_path_stream and forward it to _index_file(fp, force, namespace=namespace) (which already accepted the param).
  • Accumulate result["errors"] into all_errors and emit errors: list[str] in the complete event — same loose shape as IndexingStats.errors so non-stream UI handlers (5-cap "+N more" rendering, error-row toggle) reuse verbatim.
  • Empty-path branch (no files / not-a-path) now emits errors: [] for shape consistency.
  • Stream-level uncaught-exception branch (line 752) prefixes f"{fp.name}: {exc}" so consumers see the same shape regardless of which branch caught the error. Matches non-stream's asyncio.gather(return_exceptions=True) branch (line 394).
• Route (packages/memtomem/src/memtomem/web/routes/system.py:768-790): add namespace: str | None = None query param to GET /api/index/stream, forward to engine.
• Tests (packages/memtomem/tests/test_indexing_engine.py):
  • test_index_path_stream_namespace_propagates — pass namespace="ns590", verify indexed chunks have metadata.namespace == "ns590".
  • test_index_path_stream_complete_errors_no_silent_drop — write a file with a NUL byte (binary-detected branch at engine.py:621-628), assert complete.errors is non-empty and mentions the file name. Test name documents intent so future grep finds the regression guard.

Compatibility

• Adds an optional kwarg + an additive event field. Existing callers (web/routes/system.py:789 legacy call site already updated; cli/init_cmd.py:1546 reads only total_files/indexed_chunks/skipped_chunks from the complete event and is unaffected) continue to work without change.
• Frontend runIndexStream (app.js:3467) doesn't yet read complete.errors; that wiring + the namespace URL param land in feat(web): Index tab mid-priority follow-ups (each ships as its own PR) #582 PR-B (Prev chore: prepare for open-source release #1 collapse) and PR docs: slim READMEs, add configuration + embeddings guides #6 (4.11 toast). This PR makes those follow-ups possible without silent regressions.

Out of scope

• Per-file streaming of errors (option ii/iii from the issue's planning discussion). complete.errors is sufficient for current UX; revisit if user reports indicate large indexing runs (>100 files) where waiting for complete is impractical.
• Normalizing _index_file's mixed path-prefix convention — some branches prefix f"{file_path.name}: ..." (file-too-large, binary), others don't (embedding failures at engine.py:684). Separate non-stream bug; this PR preserves the existing loose contract and only path-prefixes the stream-level uncaught-exception branch (line 752) for consistency with the non-stream outer-gather branch.

Refs

• Issue: Streaming indexing endpoint missing namespace + errors parity #590
• Unblocks: feat(web): Index tab mid-priority follow-ups (each ships as its own PR) #582 PR-B (Prev chore: prepare for open-source release #1 — Index tab two-button collapse)
• Informs: feat(web): Index tab mid-priority follow-ups (each ships as its own PR) #582 PR docs: slim READMEs, add configuration + embeddings guides #6 (4.11 — indexing in-flight visibility, will surface complete.errors via toast)

Test plan

• uv run ruff check packages/memtomem/src && uv run ruff format --check packages/memtomem/src
• uv run pytest packages/memtomem/tests/test_indexing_engine.py -k "test_index_path_stream" -m "not ollama" — 4/4 pass (2 existing exclude-guard tests + 2 new parity guards)
• uv run pytest packages/memtomem/tests/test_indexing_engine.py packages/memtomem/tests/test_web_routes.py packages/memtomem/tests/test_web_exclude_guard.py packages/memtomem/tests/test_init_cmd.py -m "not ollama" — 441/441 pass (combined). Confirms no regression in route handlers, exclude guards, or the mm init wizard's seed flow.

🤖 Generated with Claude Code