The v0.2.0 source tag, both Go module tags, all four npm packages, server/RPG Python packages, and all multiarch container images are published and verified.
Install memorylayer-client, memorylayer-langchain, and memorylayer-llamaindex at 0.2.1. Their initial 0.2.0 wheels contained metadata but no Python modules. The corrected packages passed 249 tests against installed replacement wheels, all hosted checks, and verification of the downloaded PyPI artifacts, including successful imports. The server and RPG packages remain at 0.2.0. The three defective Python 0.2.0 releases have been yanked. Python correction publication.
MemoryLayer v0.2.0 adds deterministic memory and context handling, an entity and perspective layer for recall, and new SDK and plugin capabilities.
- New Go SDK with an optional nested Aether transport. Both modules require Go 1.25.14; the Aether transport uses Aether API/SDK v0.2.3.
- Updated Aether clients and the embed-server sidecar to v0.2.3; CUDA images explicitly install CUDA 13 PyTorch on native AMD64 and ARM64 build runners.
- First PyPI release of
memorylayer-server-rpg, providing repository graph sync, traversal, overlays, conflicts, maintenance, and enrichment. The server container bundles the plugin. - Durable session checkpoints and context deltas, typed relations, knowledge-work and email ingestion, and SDK support across Python, TypeScript, and Go.
- Improved retrieval and knowledge-base generation, with a retrieval-quality gate in CI.
- Packaging and dependency security fixes, browser SDK compatibility, corrected Python adapters, and preserved authorization responses for RPG APIs.
Python packages: memorylayer-server, memorylayer-server-rpg, memorylayer-client, memorylayer-langchain, and memorylayer-llamaindex.
npm packages: @scitrera/memorylayer-sdk, @scitrera/memorylayer-mcp-server, @scitrera/memorylayer-cc-plugin, and @scitrera/memorylayer-opencode-plugin.
Coordinated Go tags:
memorylayer-sdk-go/v0.2.0memorylayer-sdk-go/aether/v0.2.0
Container images are published to GHCR and Docker Hub. The server, CPU embed server, and CUDA 13 embed server are built natively for AMD64 and ARM64 and published as multiarch images.
The unsuffixed embed-server tags, including latest, now select the CPU variant. GPU deployments should use 0.2.0-cuda13 or latest-cuda13.
The OpenClaw plugin is included in source and tested; it is not published to npm in this release. The Codex plugin is distributed as source. Explorer remains an unpublished application.
Validation includes the Python, Go, and npm workflows, native container builds for both architectures, and an offline server-container smoke test covering health, RPG route discovery, and Aether v0.2.3. All four npm tarballs passed content and GitHub OIDC provenance checks, and their public entry points imported successfully after a fresh registry install. npm verified all 98 installed packages’ registry signatures and 13 provenance attestations. CUDA builds verify the installed PyTorch CUDA variant; live GPU inference and external-service end-to-end tests were not run.
Known dependency limitation: the LlamaIndex adapter brings in NLTK 3.10.3, with unresolved advisory GHSA-8mgp-746c-j5xp affecting model-artifact filesystem restrictions. Avoid untrusted model artifacts; see the release review for validation limits.
See the changelog and release review for details.
The final publication record covers registry and deployment verification completed after the initial review.