Skip to content

Memmy v1.0.5

Choose a tag to compare

@github-actions github-actions released this 06 Aug 12:52
b50f616

Memmy v1.0.5

Highlights

  • Improved time-aware Memory recall. Requests with a concrete time range can now retrieve matching L1 trace summaries in chronological context.
  • Made episodic Memory capture more reliable by committing routing, raw turns, and L1 records only after a turn completes. Cancelled, interrupted, and structurally incomplete turns no longer reserve an episode or create durable L1 memories.
  • Added QR-based Feishu bot setup through the official Lark flow, while retaining manual App ID and App Secret configuration as a fallback.

Agent and integration improvements

  • Refined the first-report handoff for detected Cursor, Claude Code, Codex, OpenCode, OpenClaw, Hermes, and WorkBuddy installations, with prompt-copy fallback and Memory lookup verification when permission allows.
  • Codex integration setup now registers, trusts, and verifies only the two user-level Memmy hooks it installs.
  • Structured quota-exhaustion errors are now propagated consistently across supported providers and the desktop UI, without mistaking ordinary response text for a quota failure.
  • Invalid existing Agent configuration files now fail with a clear error instead of silently reverting to defaults; missing configuration files still retain first-run behavior.
  • When the Memmy Memory service is unavailable, Agent turns continue but show an explicit warning instead of making a silent fallback look like an empty recall.

Desktop and Memory improvements

  • Reworked the onboarding report around user preferences and the latest task, and now preserves the generated report as a real chat and handoff memory instead of rerunning the Agent after onboarding.
  • Expanded the guided product tour across Memory logs, connected Agents, sync settings, the Memory overview, and tools; the flow adapts when scan permission is declined.
  • Memory detail views now provide direct navigation between related traces, tasks, policies, world-model entries, and skills.
  • Strengthened L3 evidence validation and episode reward processing, including safer handling of negative outcomes and reusable avoidance guidance.
  • Agent-source scan permission is now installation-scoped, so the same local installation uses one consistent choice across account contexts.
  • Settings now show configured invitation reward amounts and clearer BYOK usage copy.

Packaging notes

  • Consolidated macOS and Windows packaging behind parameterized scripts for version, architecture, edition, and signing mode.
  • Desktop packages now exclude dependency tests, documentation, examples, and coverage files from the runtime payload.
  • Packaged builds now prepare and include the default local embedding model as an application resource.

Artifact verification

  • Before publication, all four signed installers were downloaded again from OSS and matched their published Content-MD5 and object size; both macOS disk images also passed hdiutil verify.

What's Changed

Full Changelog: v1.0.4...v1.0.5

Downloads

Installation

On Windows, download the appropriate signed and run the installer. On Apple Silicon Macs, download the appropriate signed , open it, and drag Memmy to Applications.

Checksums

Verify downloads with or attached to this release. The workflow also verifies every OSS object against its header before publishing.