scope-recall v1.0.11

[1.0.11] - 2026-06-11

Added

• Added a MiniMaxEmbedder (provider: minimax) and a build_embedder route for the MiniMax embo-01 embedding endpoint. The endpoint is non-OpenAI-compatible (texts plural, type: "db" | "query", vectors reply), so the embedder talks to it directly via urllib.
• Added MiniMax document/query request-type separation: vector indexing/upserts use db, while vector search uses query through the embedder query path.
• Added optional MiniMax GroupId support for accounts that still require it, with group_id / group_id_env configuration.

Changed

• Bumped package, plugin, release-check metadata, README, and stability docs to 1.0.11.

scope-recall v1.0.10

[1.0.10] - 2026-06-10

Added

• Added deterministic external-artifact enrichment for direct memory writes and nightly digest candidates. GitHub issues, PRs, commits, releases, repositories, and URLs now get a human-readable Artifact anchors: block plus structured artifacts metadata, derived entities, and tags.
• Added scope_recall_store_secret_index, an explicit credential-index tool that stores searchable service/account/purpose/vault-reference metadata without storing plaintext secret values in SQLite, FTS, vector text, exports, logs, or chat replies.
• Added regression coverage for direct GitHub issue anchors, nightly digest artifact preservation, and secret-index export hygiene.

Changed

• Bumped package, plugin, README, stability contract, and release-check metadata to 1.0.10.
• Updated project URLs to the Hermes-specific repository slug 410979729/scope-recall-hermes while keeping the runtime package and plugin ID as scope-recall.
• Strengthened nightly digest extraction instructions so external artifacts retain repo/name, issue/PR/release/commit identifiers, exact URLs, and available status/date/author/next-step anchors.

Fixed

• Fixed vague memory records that mentioned external work without durable lookup anchors, forcing later sessions to rediscover issue/PR/release URLs from scratch.
• Fixed a secret-index false positive where multiline credential metadata such as a label ending in credential followed by Kind: api_key could be rejected as secret-like-content even though no plaintext secret was stored.

scope-recall v1.0.9

[1.0.9] - 2026-06-09

Added

• Added the sqlite-bruteforce vector backend for non-AVX or native-dependency-sensitive hosts. It stores rebuildable vector companion rows in $HERMES_HOME/scope-recall/vector.sqlite3 while keeping $HERMES_HOME/scope-recall/memory.sqlite3 as the truth source.
• Added docs/naming.md to define the public scope-recall spelling versus Python/tool/config identifiers that use scope_recall.
• Added docs/hermes-upstream-recommendation-plan.md with the standalone-provider checklist and Hermes upstream recommendation route.
• Added regression coverage for native-free vector imports, sqlite-bruteforce runtime sync/search, doctor reporting, and repair-script rebuilds.

Changed

• Moved lancedb/pyarrow to the lancedb optional dependency extra. Default package import no longer requires native vector dependencies, while CI and LanceDB installs use .[lancedb].
• Extended vector.backend configuration, runtime dispatch, doctor diagnostics, release checks, and repair tooling to cover both lancedb and sqlite-bruteforce companions.
• Updated installation docs to distinguish the recommended LanceDB path from the native-free SQLite fallback path.

Fixed

• Fixed the no-AVX/native-import failure mode where importing vector runtime modules could fail before the operator had a chance to select a safer backend.
• Fixed the #4 naming ambiguity by documenting where each spelling is authoritative instead of performing a risky whole-repository rename.

scope-recall v1.0.8

[1.0.8] - 2026-06-03

Added

• Added deterministic Chinese entity fallback hints so compound input-method terms such as 自然码 and 双拼 are extracted even when Jieba is unavailable or segments differently in CI/runtime environments.
• Added docs/external-shared-memory.md with deployment guidance for connecting a local scope-recall instance to an existing central shared backend such as PostgreSQL.

Changed

• Bumped package, plugin, release-check metadata, README, and stability docs to 1.0.8.
• Reworded the V1 scope documentation around the positive architecture: local-first recall, SQLite truth storage, LanceDB companion retrieval, explicit bridge boundaries for external shared backends, Hermes-native skill ownership for procedural knowledge, and deployment-driven observability.
• Included the external shared-memory integration document in release-gate source and wheel checks.

Fixed

• Fixed the GitHub Actions regression where the Chinese entity test could fail because 自然码 was not extracted when Jieba was not installed or did not split the compound phrase as expected.

scope-recall v1.0.7

scope-recall v1.0.7

Added

• Added scripts/doctor.py, a read-only source/runtime health report that checks release metadata alignment, SQLite truth availability, LanceDB companion readability, and repair recommendations.
• Added BM25 as an optional final-score component for hybrid retrieval, while preserving candidate-local SQLite FTS5 bm25() normalization and raw-score metadata for explainability.
• Added optional Jieba-backed Chinese entity extraction and broader code-ish entity extraction for mixed Chinese/English project memory.
• Added explicit temporal-decay scoring, deterministic source-trust priors, typed memory_relations, and conservative contradiction marking with feedback/metadata evidence.
• Added opt-in shared-pool scope stats plus scope_recall_inspect, scope_recall_explain, and scope_recall_benchmark observability tools.

Changed

• Bumped package, plugin, release-check metadata, README, and stability docs to 1.0.7.
• Extended the release gate stable-tool check to cover the full public V1 default tool surface and new observability tools.

Fixed

• Aligned the README public version text with package/plugin metadata and documented the Hermes venv + PYTHONPATH test command so plain pytest from an unrelated environment is not mistaken for release evidence.
• Preserved pure lexical recall in default hybrid mode when BM25 metadata exists but bm25_weight is still zero, avoiding accidental dampening of local/general matches.
• Reduced generic English entity noise so related-entity results keep explicit caller-provided entities such as yuheng visible.

scope-recall v1.0.6

[1.0.6] - 2026-06-01

Added

• Added capture_llm module: LLM-powered semantic extraction of user+assistant turns into classified durable memory (preference, workflow, pitfall, decision, etc.) with user-configurable model and endpoint.
• Added capture_llm configuration block (capture_llm.enabled, capture_llm.model, capture_llm.base_url, etc.) with safe defaults (disabled by default, requires API key).
• LLM extraction runs in sync_turn before legacy regex extraction; if LLM succeeds, regex and raw-user fallback are skipped to avoid noise.
• LLM extraction preserves entity and tag metadata on stored candidates for better recall targeting.

Changed

• sync_turn now has a four-tier capture pipeline: LLM semantic extraction → regex extraction → raw user capture → raw assistant capture (legacy).
• Bumped package, plugin, and release-check metadata to 1.0.6.
• Synced public README/stability/OpenClaw comparison wording with the v1.0.4/v1.0.5 entity, feedback, and nightly digest features.
• Extended the public scope_recall_store tool schema memory_type enum to include workflow-oriented digest types already accepted by the governance layer.

v1.0.5 — Nightly Digest & Workflow Memory Consolidation

Added

• Added scripts/nightly-digest.py, a profile-scoped daily conversation digest that reads Hermes state.db/legacy lcm.db, extracts durable memories, writes through the SQLite truth store, syncs the LanceDB companion when enabled, and records digest run/source ledgers.
• Added task-session workflow extraction so successful tool-heavy work can be retained as reusable workflow/tool-chain memory without storing raw tool or system output.
• Added digest safeguards for secret redaction, task-vs-normal session classification, dry-run planning, exact duplicate cleanup, and semantic skip/update/insert decisions against existing scope-recall rows.
• Added regression coverage for nightly digest session loading, sensitive-value redaction, workflow memory writes, digest ledgers, duplicate skips, and dry-run no-write behavior.

Changed

• Bumped package and plugin metadata to 1.0.5.
• Extended accepted memory_type values with workflow-oriented digest types such as workflow, tool_trace, summary, pitfall, and decision.

v1.0.4 — Entity Graph, Trust Feedback, BM25 Hybrid

Added

• Entity graph: SQLite memory_entities + memory_feedback tables, deterministic regex entity extraction, auto-backfill
• 4 new tools: scope_recall_context, scope_recall_probe, scope_recall_related, scope_recall_feedback
• BM25 hybrid retrieval: FTS5 candidate pool now uses ORDER BY bm25() before recency tie-breaking
• Trust scoring: feedback rating (+1/-1) adjusts trust by +0.08/-0.12, applied in recall ranking
• Entity overlap bonus: memories sharing entities with the query get ranking boost
• Quality weight: trust/importance/confidence metadata factored into recall scores

Changed

• storage_views.py: FTS5 candidates BM25-ordered, configurable candidate_pool
• recall.py: applies quality_weight and entity_overlap_bonus from graph layer
• Extended stats with scoped entity and feedback counts

Fixed

• Block [System note] gateway wrappers and context-compaction handoffs from automatic capture

Full changelog

scope-recall v1.0.3

Patch release with structured memory classification metadata, FTS hygiene repair coverage, release-gate coverage for scope_recall_hygiene, and dedicated SCOPE_RECALL_GEMINI_EMBEDDING_API_KEY isolation for the default Gemini embedding path.

scope-recall v1.0.2

Patch release for memory-quality governance: capture/write hygiene, receipts, curated-memory controls, stable import fingerprints, and SQLite/vector cleanup readiness.