129 lines (95 loc) · 4.07 KB

changelog

[1.3.0] - 2025-12-20

🎉 major refactors

api simplification

python sdk: simplified to zero-config Memory() api matching javascript
- from openmemory.client import Memory → mem = Memory()
- works out of the box with sensible defaults (in-memory sqlite, fast tier, synthetic embeddings)
- optional configuration via environment variables or constructor
- breaking change: moved from OpenMemory class to Memory class

benchmark suite rewrite

implemented comprehensive benchmark suite in temp/benchmarks/
- typescript-based using tsx for execution
- supports longmemeval dataset evaluation
- multi-backend comparison (openmemory, mem0, zep, supermemory)
created src/main.ts consolidated benchmark runner
- environment validation
- backend instantiation checks
- sequential benchmark execution with detailed logging

✨ features

core improvements

Memory.wipe(): added database wipe functionality for testing
- clear_all implementation in db.ts for postgres and sqlite
- clears memories, vectors, waypoints, and users tables
- useful for benchmark isolation and test cleanup
environment variable overrides:
- OM_OLLAMA_MODEL: override ollama embedding model
- OM_OPENAI_MODEL: override openai embedding model
- OM_VEC_DIM: configure vector dimension (critical for embedding compatibility)
- OM_DB_PATH: sqlite database path (supports :memory:)

vector store enhancements

added comprehensive logging to PostgresVectorStore
- logs vector storage operations with id, sector, dimension
- logs search operations with sector and result count
- aids in debugging retrieval issues

🐛 bug fixes

embedding configuration:
- fixed models.ts to respect OM_OLLAMA_MODEL environment variable
- resolved dimension mismatch issues (768 vs 1536) for embeddinggemma
- ensured OM_TIER=deep uses semantic embeddings (not synthetic fallback)
benchmark data isolation:
- implemented proper database reset between benchmark runs
- fixed simhash collision issues causing cross-user contamination
- added resetUser() functionality calling Memory.wipe()
configuration loading:
- fixed dotenv timing issues in benchmark suite
- ensured environment variables load before openmemory-js initialization
- corrected dataset path resolution (longmemeval_s.json)

📚 documentation

comprehensive readme updates:
- root README.md: language-agnostic, showcases both python & javascript sdks
- packages/openmemory-js/README.md: complete api reference, mcp integration, examples
- packages/openmemory-py/README.md: zero-config usage, all embedding providers
api documentation:
- environment variables with descriptions
- cognitive sectors explanation
- performance tiers breakdown
- embedding provider configurations

🔧 internal improvements

type safety: added lint error handling in benchmark adapters
code organization: separated generator, judge, and backend interfaces
debug tooling: created dimension check script (check_dim.ts)
logging standardization: consistent [Component] prefix pattern

⚠️ breaking changes

python sdk now uses from openmemory.client import Memory instead of from openmemory import OpenMemory
Memory() constructor signature changed to accept optional parameters (was required)
benchmark suite moved to typescript (was python)

[1.2.2] - 2024-11-xx

bug fixes

memory consolidation edge cases
multi-user query isolation
vector dimension handling

[1.2.1] - 2024-11-xx

improvements

performance optimizations for large datasets
enhanced sector classification accuracy

[1.2.0] - 2024-10-xx

features

multi-sector memory architecture
cognitive decay system
reflection and consolidation

[1.1.0] - 2024-09-xx

features

initial typescript sdk release
sqlite vector store
basic query and add operations

[1.0.0] - 2024-08-xx

initial release

python sdk
local-first architecture
basic memory operations