The second brain gets depth release. v1.3 keeps private memory grounded in the current user’s own writing while making long sources, retrieval scope, progressive answers, deliberate automation inputs, and documentation gaps inspectable as that memory grows.
Added
- Chunk-aware Second Brain memory — long private notes, journal entries, portfolio theses, holding notes, and transaction notes are split into deterministic, lightly overlapping chunks. Stable internal keys preserve original citations; incremental reindexing embeds only changed chunks and safely prunes stale rows.
- Inspectable evidence scopes — asks can restrict retrieval to validated private source types. The UI exposes source filters, per-source index counts, and the evidence scope used for each answer.
- Incremental synthesis — compatible OpenAI-style providers stream grounded answers over authenticated NDJSON. Stable non-streaming and retrieval-only fallbacks remain complete, cited, and honestly labelled.
- API-key external note ingestion —
read_writekeys can idempotently upsert owner-scoped private notes atPUT /api/v1/notes/external, supporting deliberate inputs such as Hermes YouTube summaries. - On-demand journal completeness review — the Trade Journal identifies absent rationale, outcomes, emotions, setup labels, and stale open-trade thesis updates, then links each result to its editor without inventing facts or advice.
Verification
- Backend: 817 tests passed, 70.15% coverage
- Frontend: 296 tests across 92 files
- Production build, compileall, fabricated-data guard, Docker Compose configuration, and host smoke approved