Skip to content

CHANGELOG.md

Latest commit

 

History

History
139 lines (125 loc) · 6.8 KB

CHANGELOG.md

File metadata and controls

139 lines (125 loc) · 6.8 KB
Raw

Changelog

All notable changes to MultiMind SDK will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Targets 0.3.0 — first release after the packaging modernization, lazy-import, and test-stabilization passes.

Added

  • Guard proxy: multimind serve — an OpenAI-compatible reverse proxy that adds PII guarding, budgets, and audit logging to any existing app (see docs/guard-proxy.md)
  • AI inventory and chargeback: multimind audit CLI (see docs/ai-inventory.md)
  • multimind scan-text CLI for scanning text for PII/sensitive data
  • Compliance MCP server (python -m multimind.mcp_server) for Claude Desktop / Claude Code and other MCP clients (see docs/mcp-server.md)
  • Framework adapters for LangChain, LlamaIndex, CrewAI, and the OpenAI SDK (see docs/integrations.md)
  • New extras: mcp, langchain, llamaindex, crewai
  • pyproject.toml with modular extras: rag, vector-stores, agents, memory, documents, finetune, finetune-gpu, compliance, gateway, dev, all (#41)
  • PEP 562 lazy imports — from multimind import OpenAIModel no longer pulls in torch, transformers, chromadb, etc. on minimal installs
  • Helpful ImportError messages from subpackages naming the right extras group to install (e.g. pip install multimind-sdk[rag])
  • multimind/_lazy.py helper module (import_optional, lazy_attr)
  • tests/conftest.py with mock_openai_key / mock_anthropic_key fixtures, requires_torch / requires_faiss / requires_plotly etc. skip decorators, and an auto-skip collection hook for tests marked @pytest.mark.requires_api_key when no keys are in the environment
  • plotly, dash, and pandas added to [compliance] extra (powers the compliance visualization dashboards)
  • optuna added to [finetune] extra (was eagerly imported but undeclared)
  • Multi-job CI workflow (.github/workflows/ci.yml): lint, test-core (Python 3.9 / 3.10 / 3.11 / 3.12 / 3.13 matrix), test-rag, test-compliance, test-finetune, test-gateway, test-full (coverage + 95% pass-rate gate)
  • CHANGELOG.md (this file)
  • Re-exports for SlackIntegrationHandler and JiraIntegrationHandler on multimind.integrations

Changed

  • Minimum Python version raised to 3.9
  • Version bumped to 0.3.0
  • Complete README rewrite (660 → ~180 lines): single Quick Start, honest feature table, working code examples verified against the actual API surface
  • Comprehensive .gitignore (build artifacts, IDE files, coverage, database files, OS metadata, jupyter checkpoints, ruff/mypy caches)
  • multimind/client/model_client.py: LSTMModelClient, RNNModelClient, GRUModelClient now call torch.load(weights_only=False) for PyTorch ≥ 2.6 compatibility (trusted-load is appropriate since users load their own training checkpoints)
  • Narrowed enforced Ruff rule set to E, F, W, I in pyproject.toml; UP/N rules disabled until Pydantic v1 @validator methods can be migrated safely (see pyproject.toml for rationale)
  • Tooling configuration consolidated into pyproject.toml ([tool.pytest.ini_options], [tool.ruff], [tool.black], [tool.mypy]); pytest.ini and .flake8 deleted
  • Moved root-level doc files to docs/:
    • README-llm.md → docs/llm-integration.md
    • llm-crawler-guide.md → docs/llm-crawler-guide.md
    • llm.txt → docs/llm.txt
    • multimind-sdk-metadata.json → docs/multimind-sdk-metadata.json

Fixed

  • Circular import in multimind.compliance — removed redundant run_compliance re-export from multimind.compliance.__init__ (the function remains available at multimind.cli.compliance.run_compliance)
  • Test discovery: pytest tests/ now works regardless of invocation (pythonpath = ["."] added to pytest config); previously some tests required PYTHONPATH=$PWD as a CI workaround
  • multimind/vector_store/typesense.py — nested same-quote f-string syntax error that prevented the module from loading on Python ≥ 3.11
  • examples/mcp/__init__.py — removed imports of modules that no longer exist (.ci_cd_workflow, .code_review_workflow, etc.); package init is now lightweight, callers import specific examples directly
  • multimind/fine_tuning/qlora_trainer.py — removed unused import bitsandbytes as bnb so the module loads on macOS/ARM and any host without a GPU
  • multimind/ensemble/advanced.py — replaced 4 bare except: with explicit except Exception: (preserves keyboard-interrupt semantics)
  • tests/test_import.py — three tests used return True instead of assert; pytest 9 will error on this pattern
  • tests/test_retrieval.py — two async retrieval tests were missing await, so the coroutines were silently dropped and the tests were trivially "passing" without exercising any code
  • tests/test_model_client.py — DummyModel moved to module scope so torch.save / torch.load round-trips work; shape corrected to [batch, seq_len, vocab]; three previously-skipped tests now run
  • tests/test_document_loader.py — replaced Mock(spec=BaseLLM) with AsyncMock(spec=BaseLLM) so the language-detect path (response.strip().lower()) gets a real string
  • 6,394 lint issues auto-fixed via ruff check --fix + black (whitespace, import sorting, end-of-file newlines, redundant open() modes, etc.) across 347 files in multimind/

Removed

  • setup.py was kept as a no-op shim in 92294804 for backward compatibility; further removal will follow once tooling has had time to settle on pyproject.toml
  • pytest.ini (replaced by [tool.pytest.ini_options])
  • .flake8 (replaced by Ruff configuration)
  • requirements-base.txt, requirements-compliance.txt, requirements-dev.txt (replaced by [project.optional-dependencies])
  • TEST_FIXES_COMPLETE.md (internal tracking — no longer needed in the repo root)
  • [local] install extra (the ollama PyPI package was never used; multimind.models.ollama talks to Ollama directly over HTTP via aiohttp)

Security

  • cryptography.zkp ZeroKnowledgeProof falls back to a clearly-named _DummyZKP with a UserWarning when the optional dependency is not installed. This is not a production-grade ZKP — install the real dependency or use a different compliance pathway for any guarantee that needs to survive audit.

[0.2.2] - 2025-08-18

Fixed

  • Model router error (#51)
  • Streaming, memory, SQLAlchemy cleanup, and reliability issues across the SDK

Changed

  • Refactored Unified API for stable Mixture-of-Experts execution with multi-modal routing and dynamic expert support

[0.2.1] - 2025-07

Added

  • Initial public release: multi-model chat, RAG, agents, basic compliance framework