v1.0.0

Theme: embedded lifecycle memory for Python and LangChain agents.

• Added SCM, the URL-free Python runtime backed by isolated per-user engines.
• Added LangChain 1.x middleware and create_scm_agent with automatic recall,
  durable message capture, bounded wake context, and exactly five tools.
• Added lifecycle snapshot v2 with atomic local persistence, optional
  encryption, corruption recovery, idempotency, and PostgreSQL coordination.
• Added per-user export, import, hard deletion, process locks, and sleep leases.
• Kept REST, MCP, SCMEngine, and the JavaScript REST client compatible as
  advanced and remote surfaces.

Verification

• Embedded direct API and LangChain tests run without an SCM server or URL.
• Local recovery and real PostgreSQL multi-process tests cover lifecycle state,
  concurrent writes, and exclusive sleep leases.
• The published paper and reproducibility manifest are linked through DOI
  10.5281/zenodo.21323877.

SCM v0.9.1 Plug-And-Play Runtime

SCM v0.9.1 makes the open-source runtime a developer-first plug-and-play product.

Highlights:

• scm demo for the 60-second local lifecycle-memory experience.
• scm doctor for install, asset, data-dir, SDK, REST, and provider checks.
• Offline/hash first-run path with no OpenAI key required.
• Public Python API exports both SCMEngine and SCMClient.
• Python and JS package versions bumped to 0.9.1.
• README, Getting Started, Product thesis, examples, publish runbook, and CI smoke gates refreshed around the five-tool contract.

Assets:

• Python wheel: scm_memory-0.9.1-py3-none-any.whl
• Python sdist: scm_memory-0.9.1.tar.gz
• JS package tarball: scm-memory-0.9.1.tgz

Verified locally and in CI: Python 3.10/3.11/3.12 tests, distribution build, installed-wheel smoke, REST five-tool smoke, JS SDK test/pack, and tracked-file secret scan.

SCM v0.9.0 Product Runtime

SCM v0.9.0 Product Runtime

Draft release for the SCM product-grade open-source runtime.

This release pairs the installable runtime with the July 2026 paper evidence. The raw benchmark outputs are attached as release assets rather than committed to Git.

Assets

• scm-alb-v02-canonical-1500.tar.gz: 1500-run canonical ALB matrix.
• scm-alb-v02-reference-4000.tar.gz: 4000-run reference-adapter matrix.
• scm-ablation-1200.tar.gz: 1200-run component ablation matrix.
• scm-mem0-official-500.tar.gz: 500-row official upstream Mem0 external baseline.
• scm-real-llm-openai.tar.gz: OpenAI gpt-5.4-mini real-LLM brutal artifacts.
• SHA256SUMS.txt: checksums for all attached artifact bundles.

Verification

The release-prep branch was verified with focused Python SDK/REST/MCP tests, JS SDK tests, tracked-file secret scanning, wheel/sdist build, twine check, and clean installed-wheel smoke tests from /tmp.

Draft PR: #1