feat: add MiniMax as a supported embedding and LLM provider#181
Open
octo-patch wants to merge 76 commits intoCortexReach:mainfrom
Open
feat: add MiniMax as a supported embedding and LLM provider#181octo-patch wants to merge 76 commits intoCortexReach:mainfrom
octo-patch wants to merge 76 commits intoCortexReach:mainfrom
Conversation
… smart format with enriched metadata.
…bustness, CLI, and documentation.
… for decayable memory parsing.
feat: Implement lifecycle-aware memory decay and enhance retriever robustness, CLI, and documentation.
…ing` configuration, clarify default behaviors, and remove `AccessTracker` flush.
feat: Add plugin manifest regression test, implement `embedding.chunk…
…pture logging, and improve smart extraction fallback.
…d configuration, and delete AGENTS.md.
feat: Reduce smart extraction message threshold to 2, enhance auto-capture logging, and improve smart extraction fallback.
…ntegration, updating test scripts.
…model to gpt-4o-mini in example configurations.
… with built-in memory, and simplify the quick start guide.
Merge branch 'main' into dev
…, and feature development, and enhance LLM client with improved logging and error handling.
feat: Add new memory plugins for host validation, release consistency…
…o-capture debug logging, and update .gitignore with new plugin ignore rules.
… text processing skip logic with debug logging, and refine memory extraction prompts to improve memory classification.
…extractor for advanced noise filtering and learning, with accompanying documentation.
* "Claude PR Assistant workflow" * "Claude Code Review workflow"
…urce typing
- store.ts: add .distanceType('cosine') to vectorSearch (critical: L2 default drops valid results)
- store.ts: add getFtsStatus(), rebuildFtsIndex() for BM25 health diagnostics
- retriever.ts: extend source typing with 'cli' for CLI trace distinction
- cli.ts: mark CLI retrievals with source='cli', add reindex-fts command
- test: add vector-search-cosine.test.mjs (4 tests)
…ce tracking Extends OpenViking's smart memory architecture with context-aware support: - smart-metadata.ts: add SupportInfoV2/ContextualSupport types, normalizeContext, parseSupportInfo (V1→V2 migration), updateSupportStats; fix LegacyStoreCategory missing 'reflection' - memory-categories.ts: extend DedupDecision with support/contextualize/contradict, add contextLabel to DedupResult, supported count to ExtractionStats - extraction-prompts.ts: extend dedup prompt with 3 new decisions + context_label - smart-extractor.ts: add handleSupport/handleContextualize/handleContradict handlers in processCandidate pipeline, extract contextLabel in llmDedupDecision - test: add smart-metadata-v2.mjs (6 tests, all passing)
- smart-extractor.ts: handleMerge now accepts contextLabel and updates support stats after successful merge (aligns with support/contextualize/ contradict handlers) - smart-metadata.ts: stringifySmartMetadata caps arrays to prevent JSON bloat (sources≤20, history≤50, relations≤16) - test/context-support-e2e.mjs: 3 E2E scenarios testing support, contextualize, and contradict decisions end-to-end
Blocking: 1. Add smart-metadata-v2, vector-search-cosine, context-support-e2e to npm test chain (package.json) 2. Rewrite smart-metadata-v2.mjs to import production code via jiti (normalizeContext, parseSupportInfo, updateSupportStats, etc.) 3. Rewrite vector-search-cosine.test.mjs to use real MemoryStore against temp LanceDB (no more fakeStore) Suggestions: 4. Fix '下午' mapping: evening → afternoon (add to vocabulary) 5. parseSupportInfo: validate slice numeric fields (confirmations, contradictions, strength, last_observed_at) 6. Document slice truncation drift as accepted trade-off 7. dropIndex: log warning instead of silently swallowing errors
…ive noise filtering with non-discriminative embeddings When the embedding model produces identical vectors for all inputs (e.g. deterministic mock embeddings in tests), every text matches every noise prototype with cosine similarity 1.0, causing the noise filter to reject all content. Smart extraction is skipped entirely, falling back to regex which also captures nothing. Add a self-diagnostic after init(): if the first two prototype vectors have cosine similarity > 0.98, the bank recognizes the embedding model is degenerate and disables itself. This fixes the flaky smart-extractor-branches.mjs test on master (all 9 scenarios now pass reliably).
…it/clean-recall-text-final Validated locally. - Host functional test passed - Agent E2E passed - Web dashboard / WebChat check passed - Visible recall text and auto-recall injected text are now cleaned up as expected - No blocking regression found from this PR I did observe one non-blocking model response inconsistency in a JSON prompt during manual chat testing, but the underlying recall output was correct and it does not appear to be caused by this PR.
…al-and-context # Conflicts: # package.json
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…coring Remove redundant vectorScore + 15% bonus branch from fuseResults Math.max. Weighted fusion (vectorScore * vectorWeight + bm25Score * bm25Weight) is now the primary scoring formula. BM25 high-score floor (>= 0.75) preserved for exact keyword matches. Closes CortexReach#130 (Layer 3) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
updateSupportStats returns a new SupportInfoV2 object — the original is
not mutated. The old code discarded the return value and persisted the
empty original, causing merge+contextLabel to always write empty
support_info ({global_strength:0.5, total_observations:0, slices:[]}).
* chore: add version consistency check and auto-sync Add CI job to verify package.json and openclaw.plugin.json versions match. Add npm version hook to auto-sync openclaw.plugin.json on version bump. * fix: stabilize plugin version sync --------- Co-authored-by: Heng Xia <pope@Hengs-Mac-mini.local>
…al-and-context # Conflicts: # package.json
…al-and-context feat: 检索正确性修复 + 情境化 SupportInfo — 适配 master 架构
Add support info regression coverage
…generate-embedding-guard fix: NoisePrototypeBank degeneracy guard — fix flaky smart-extractor-branches test
…-130 fix(retriever): make vectorWeight/bm25Weight actually affect fusion scoring
…sion-weighting-130 Revert "fix(retriever): make vectorWeight/bm25Weight actually affect fusion scoring"
…coring Remove redundant vectorScore + 15% bonus branch from fuseResults Math.max. Weighted fusion (vectorScore * vectorWeight + bm25Score * bm25Weight) is now the primary scoring formula. BM25 high-score floor (>= 0.75) preserved for exact keyword matches. Closes CortexReach#130 (Layer 3) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…-130-v2 fix(retriever): make vectorWeight/bm25Weight actually affect fusion scoring
- Add MiniMax embo-01 (1536 dims) to EMBEDDING_DIMENSIONS map - Add MiniMax API detection in getProviderLabel() for better error messages - Add MiniMax embedding + LLM config examples in both README.md and README_CN.md - Document MiniMax-M2.5 (204K context) as an LLM option for smart extraction MiniMax provides OpenAI-compatible APIs at https://api.minimax.io/v1 for both embeddings and chat completions, making it a drop-in provider choice.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Add MiniMax as an officially documented embedding and LLM provider for memory-lancedb-pro.
Changes
src/embedder.ts: Add MiniMaxembo-01model (1536 dimensions) toEMBEDDING_DIMENSIONSmap; add MiniMax API URL and model name detection ingetProviderLabel()for better error messagesREADME.md: Add MiniMax to the embedding providers table; add MiniMax config examples for both embedding and LLM (smart extraction) usageREADME_CN.md: Same documentation updates in ChineseWhy MiniMax?
MiniMax provides OpenAI-compatible APIs at
https://api.minimax.io/v1for both embeddings and chat completions. Key highlights:Test Plan
npm test— 100% green)embo-01correctly resolves to 1536 dimensions viagetVectorDimensions()getProviderLabel()detects MiniMax from both URL and model name patterns