Memory Layer v0.9.0

Memory Layer v0.9.0

Changes

• Build: prepare v0.9.0 release
• Docs: add contributor PR guidance
• Fix: gate eval shell execution
• Feat: add aggregate CLI status
• Refactor: extract web review tab
• Refactor: split TUI rendering helpers
• Refactor: move service reads into repository module
• Refactor: make mem-api lib an index
• Refactor: split service routes
• Refactor: split CLI bundle and proposals commands
• Fix: send watcher API token
• Test: add DB integration harness
• Feat: add memory provenance verifier
• Fix: harden service API auth
• Docs: add Claude review refactor plans

Package

• Debian package: memory-layer_0.9.0_amd64.deb
• SHA256: 46ad524a22831d89572979c876acd270e25b4c3493aced04a46c874b3b471ede

Memory Layer v0.8.7

Memory Layer v0.8.7

Changes

• Build: prepare v0.8.7 release
• Fix: preserve applied pgvector migration
• Feat: add refactor memory type
• Merge feat/memory-github-init into main
• Fix: stabilize MemoryAgentBench full run
• Docs: add 2026-05-16 extensive Claude Opus review
• Fix: stabilize MemoryAgentBench live pilot
• Feat: add MemoryAgentBench adapter pilot
• Docs: document OpenCode integration direction
• Feat: add memory proposal review skill
• Feat: add built-in MCP server
• Fix: shorten dev service unix socket path
• Fix: tolerate unreadable legacy memory env during bootstrap
• Docs: add architecture Mermaid diagrams
• Docs: link agent install prompt from quick start
• Feat: add GitHub initialization skill
• Merge pull request #24 from 3vilM33pl3/release/v0.8.1-refresh
• Fix: address PR review feedback
• Fix: satisfy clippy on config path test
• Fix: remove unsupported Codex approval flag
• Fix: authenticate Codex agents in Actions
• Fix: use Codex review built-in prompt
• Fix: pass Codex review model via config
• Refactor: move project runtime config to user paths
• Feat: bring web UI up to TUI parity
• Feat: add TUI LLM audit toggle
• Document LLM audit mode
• Show LLM audit events in CLI
• Add LLM audit activity support
• Feat: add TUI tab help pages
• Fix: skip graph retrieval for broad query terms
• Feat: add documentation memory type

Package

• Debian package: memory-layer_0.8.7_amd64.deb
• SHA256: 52e9a64a09b1c2cefd872bdef26d4171e6c05f57d079116691bf9e6613613f4b

Memory Layer v0.8.6

Memory Layer v0.8.6

Changes

• Fix: satisfy release clippy on stable
• Build: bump version to v0.8.6
• Feat: add shell completion support
• Docs: refresh navigation and skill documentation
• Fix: preserve embeddings across backend reindex
• Feat: restore cached query history results
• Feat: show query roundtrip timing breakdown

Package

• Debian package: memory-layer_0.8.6_amd64.deb
• SHA256: 421f1cdddaf546a7f34fd6d173b435ad6e4f0606682dbec122c779efecaf6f24

Memory Layer v0.8.5

Memory Layer v0.8.5

Changes

• Build: bump version to v0.8.5
• Feat: show unified skill bundle version
• Feat: add project skill version upgrades
• Fix: provide CI eval config
• Fix: stabilize CI runner checks
• Build: add GitHub automation workflows

Package

• Debian package: memory-layer_0.8.5_amd64.deb
• SHA256: e63de36ac2f3d11f7c539e168fe8716f2c0e0b703e328e7c3bc38340a6dd1858

Memory Layer v0.8.4

Memory Layer v0.8.4

Patch release focused on operational diagnostics and installable release metadata.

Highlights

• Added structured API diagnostics while preserving the existing error response field for compatibility.
• Added a TUI Errors tab that combines persisted backend diagnostics with session-local TUI errors and fix hints.
• Persisted diagnostic activity events for provider/query/watcher failures so errors are visible after the fact.
• Made automatic embedding creation after curation best-effort: provider quota/auth failures now return warnings instead of failing an already-saved memory write.
• Kept explicit embedding maintenance commands strict, but upgraded their failures with diagnostic codes, explanations, and suggested commands.
• Added documentation for diagnostics and the TUI Errors tab.
• Added a repo-local project initialization skill and fixed dev TUI manager status detection.

Package

• Debian package: memory-layer_0.8.4_amd64.deb
• SHA256: 43de0d0afeebbd3045c5cbcf2ad9171dbda0849b47763194ae01f5479130e9ac

Validation

• cargo test -p mem-api -p mem-service -p mem-cli --locked
• ./packaging/build-deb.sh
• git diff --check

Release v0.8.3

Highlights

• Refreshes the first-run documentation visuals with the Memories tab screenshot on the front page and the infographic surfaced in the feature section.
• Improves the TUI Memories tab so the initially selected memory loads its detail pane immediately.
• Adds the end-user install infographic assets to the documentation index pages.
• Aligns Rust, web, Debian, and Homebrew release metadata on version 0.8.3.

Validation

• cargo check --workspace
• cargo test -p mem-cli --locked
• packaging/build-deb.sh

Artifacts

• memory-0.8.3.tar.gz
• memory-layer_0.8.3_amd64.deb

Memory Layer v0.8.2

Memory Layer v0.8.2

Patch release focused on install/upgrade reliability and watcher efficiency.

Highlights:

• Adds memory service restart-all with --dry-run, --json, and --mark-tui-restart for installer-safe restarts.
• Debian, macOS package, and Homebrew installs now restart active Memory Layer services without starting intentionally stopped services.
• The TUI bottom bar now shows a red restart state when an install/upgrade writes a restart marker for the running TUI.
• Reduces watcher CPU usage with event-driven manager/session handling.
• Keeps release metadata aligned across Rust, web, Debian, and Homebrew packaging.

Verification:

• cargo fmt --all --check
• cargo check --workspace --locked
• cargo test -p mem-cli --locked
• npm --prefix web run build
• ./packaging/build-deb.sh

v0.8.1

Full Changelog: v0.8.0...v0.8.1

Memory Layer v0.8.0

Summary

Memory Layer v0.8.0 is a measurement-focused release. The headline work is a much stronger automated testing and evaluation framework for proving when Memory helps agents, where it costs tokens, and where claims still need more evidence.

Evaluation And Testing Framework

• Adds the memory-improvement-v1 benchmark suite for paired no-memory vs full-memory ablations with hidden memory-only facts, graph-backed retrieval checks, grounded-answer tasks, multi-step continuity scenarios, token accounting, latency tracking, and optional LLM judging.
• Adds grouped comparison metadata so evaluation reports can separate reasoning modes and quality dimensions instead of collapsing everything into one opaque score.
• Adds an LLM judge option for diagnostic answer quality checks while keeping deterministic scoring as the source of truth for pass/fail outcomes.
• Adds Dockerized Codex build-sequence evaluations, including verified Memory access, per-step token capture, persistent workspaces, and reproducible app-building scenarios.
• Adds stricter Codex eval isolation so benchmark containers copy only the minimal auth/version files they need, avoiding host AGENTS.md, plugins, logs, histories, and other non-benchmark context.

Published Benchmark Evidence

The README now highlights the latest valid memory-improvement-v1 Docker run:

• Success rate: 0.0% -> 18.1%
• Recall@K, MRR, and nDCG: 0.000 -> 1.000
• Assertion recall: 0.000 -> 0.725
• Token usage: 22,069,461 -> 12,970,186, a 41.2% reduction

The report is intentionally careful: it shows strong improvement for retrieval and grounded-answer tasks, while noting that long-running autonomous coding continuity still needs verified Memory-query evidence before making stronger claims.

Documentation And Release Hygiene

• Adds and promotes evaluation documentation, including beginner guidance, CLI references, and the full benchmark report.
• Removes invalidated May 2 evaluation reports and stale references so the front page points at the current valid evidence.
• Updates repo agent instructions to use the dev Memory environment and keep contained work committed incrementally.

Fixes

• Fixes UTF-8 snippet boundary handling in search so box-drawing or other multi-byte characters do not panic query processing.
• Fixes Docker eval execution to use the image-built memory binary and the correct production profile inside benchmark containers.
• Fixes app-build evaluation scoring around documented Escape-key behavior.

Validation

• cargo check --workspace --offline
• cargo check --workspace --locked
• cargo test --bin memory all_public_commands_have_help_metadata
• npm --prefix web run build
• packaging/build-deb.sh

Memory Layer v0.7.3

Summary

• Adds Web UI parity for the TUI embedding backend controls, including search on/off, per-backend automatic creation, and selected-backend embedding maintenance.
• Fixes selected-backend re-embed so it creates missing chunks before embedding, preventing zero coverage when chunk rows do not exist yet.
• Includes the v0.7.2 embedding controls and eval baseline improvements already on main.

Validation

• cargo fmt --check
• cargo test --workspace --all-targets
• cargo clippy --workspace --all-targets -- -D warnings
• npm --prefix web run build
• packaging/build-deb.sh