MemoryLane v0.26.1

Internal hardening plus a smoother enterprise activation flow when the backend already has consent on file.

What's Changed

• Enterprise: skip the in-app consent dialog when the backend reports already_approved; the client binds via external consent and proceeds straight to polling.
• Tightened how custom vendor baseURL values and backend-returned URLs are validated.
• Stricter validation of the database export directory (must be absolute and inside an allowed location).
• Customer checkout and subscription portal now open via short-lived signed URLs minted from the backend.
• Release workflow resolves the input ref to an immutable commit SHA once and reuses it across every job; all GitHub Actions are pinned to commit SHAs and tracked by Dependabot.

Known Issues & Limitations

• Vertex managed-mode bearer tokens aren't refreshed in-flight — long-running operations that outlive the token TTL may see 401s until the next refresh cycle (DEU-84).
• Windows OCR still depends on native OCR component availability.
• Intel macOS is not yet officially supported.

Installation

• macOS customer (Apple Silicon): install from the GitHub release page (auto-update enabled).
• macOS enterprise (Apple Silicon): MemoryLane Enterprise-arm64-mac.pkg — delivered privately per customer.
• Windows customer: MemoryLane-Setup.exe
• Windows enterprise: MemoryLane Enterprise-Setup.msi — delivered privately per customer.

Full Changelog

https://github.com/deusXmachina-dev/memorylane/compare/v0.26.0...v0.26.1

MemoryLane v0.26.0

Reworks the LLM client layer behind a single InferenceProvider, adds Google Vertex AI for enterprise, remembers model selections per vendor, and bakes per-customer backend URLs into enterprise builds at build time.

What's Changed

• Google Vertex AI support (enterprise) (#130): the enterprise edition can now route LLM calls through Vertex via the license-issued inference config.
• Per-vendor model memory (#130): swapping vendors no longer resets your model picks — each vendor remembers its own snapshot, video, and pattern-detection model.
• Custom endpoint extended to pattern detection and user-context (#130): an OpenAI-compatible local endpoint (LM Studio / Ollama) now drives all three LLM features, not just semantic summaries.
• Unified LLM client provisioning (#130): internal refactor onto Vercel AI SDK via a shared InferenceProvider. Removes the legacy api-key-manager / custom-endpoint-manager split in favour of a single vendor-credentials.json; existing configs migrate automatically on first launch.
• summary_model activity column (#130): each activity now records which model produced its summary.
• Electron 40.1.0 → 40.9.2 (#131).

Fixes

• User-context model now comes from per-vendor settings instead of the legacy global slot (#133).

Known Issues & Limitations

• Vertex managed-mode bearer tokens aren't refreshed in-flight — long-running operations that outlive the token TTL may see 401s until the next refresh cycle (DEU-84).
• Windows OCR still depends on native OCR component availability.
• Intel macOS is not yet officially supported.

Installation

• macOS customer (Apple Silicon): install from the GitHub release page (auto-update enabled).
• macOS enterprise (Apple Silicon): MemoryLane Enterprise-arm64-mac.pkg — delivered privately per customer.
• Windows customer: MemoryLane-Setup.exe
• Windows enterprise: MemoryLane Enterprise-Setup.msi — delivered privately per customer.

Full Changelog

https://github.com/deusXmachina-dev/memorylane/compare/v0.25.0...v0.26.0

MemoryLane v0.25.0

Reshapes the enterprise activation flow around a single user-facing activation code and a Bearer-auth backend contract.

What's Changed

• Single activation code for enterprise (#128): users now paste one tt_<token>.<email> activation code instead of separate fields. The device parses it locally, sends tenant_token and email to the backend, and uses Bearer auth on /license/* and /device/upload endpoints. The consent document is fetched up-front via a descriptor, hash-verified, and pinned to the configured backend origin to prevent off-host redirection. 401 on /status and /key is now treated as inactive rather than an error.
• Skip semantic client rebuild on no-op key updates (#127): updateApiKey no longer rebuilds the embedding client when the key hasn't changed.

Known Issues & Limitations

• Windows OCR still depends on native OCR component availability
• Intel macOS is not yet officially supported

Installation

• macOS customer (Apple Silicon): install from the latest GitHub release or via the project install script
• macOS enterprise (Apple Silicon): MemoryLane Enterprise-arm64-mac.pkg — delivered privately
• Windows customer: MemoryLane-Setup.exe
• Windows enterprise: MemoryLane Enterprise-Setup.msi — delivered privately

Full Changelog

https://github.com/deusXmachina-dev/memorylane/compare/v0.24.0...v0.25.0