Skip to content

release: Memmy v1.1.4 - #382

Merged
syzsunshine219 merged 10 commits into
mainfrom
release/v1.1.4
Sep 8, 2026
Merged

syzsunshine219 merged 10 commits into
mainfrom
release/v1.1.4

Conversation

@syzsunshine219

@syzsunshine219 syzsunshine219 commented Sep 8, 2026 •

Copy link
Copy Markdown
Collaborator

Prepare Memmy 1.1.4, with bundled Memory, viewer and CLI 2.1.2, against the published v1.1.3 baseline 1807e3a5.

Agent-history scans preserve installation-scoped checkpoints and deduplication across account changes, respect incremental turn boundaries, and store each turn as one memory. Tool-heavy exchanges no longer fan out into many small memories. Oversized content is truncated at UTF-8 boundaries; both the 512 KiB raw-content limit and a 1 MiB JSON-string limit are enforced so escaped control characters cannot overflow Memory's 2 MiB HTTP body limit.

Account mode applies its dedicated Memory models and clears stale BYOK connections while preserving local embedding selection. The process.pkg SQLite path uses content-addressed, verified, atomically published native assets. macOS runtime staging resolves the local AgentSourceCore dependency into a real installed directory that remains loadable after the runtime is moved away from the source checkout. English and Chinese release notes describe these changes and the truncation behavior.

Validation is bound to candidate e967fc3bf7b2b3112c104aedbdc93a491d6030a0:

  • Full Harness passed on a clean, unchanged commit with baseline 1807e3a5 explicitly supplied. Nine checks executed; changed ESLint, the dedicated local-API matrix and renderer/interface assertions were conditionally not applicable.
  • Workspace suites: Backend 844 passed; Agent 5,134 passed / 3 skipped; Desktop 249 passed / 106 skipped; Memory 807 passed / 1 skipped. Platform and external-service configuration skips are recorded, not counted as passes. Smoke-plan tests: 6 passed.
  • Supplemental release workflow tests: 35 passed; packaging guards: 32 passed; shared Core/migration selection: 145 passed / 1 Windows-only skipped.
  • Core, Backend/Memory scan and real HTTP regression selection: 74 passed. The control-character request previously failed through the production HTTP client and server parser with request body is too large; the bounded request now succeeds. The production storage service is replaced by an isolated fixture in that test.
  • Two new macOS staging regressions pass against the current script and fail against the previous script. The offline fixture verifies manifest/lock resolution, real package installation and imports after relocating the runtime and removing the original workspace. No installer or native binary was built or loaded by this fixture.
  • The first Full run identified two stale Desktop command assertions missing the newly required --install-links. Those assertions now require the flag; the focused boundary suite and subsequent complete Full pass. Both the failed and successful runs are retained in the local evidence package.
  • Same-head Windows validation passed 65 unit/upgrade checks and 1 native scheduled-task test, with zero skips. Linux CLI validation passed 71 contract tests, the build, and both x64/arm64 installation, upgrade, Memory health and gateway lifecycle checks. The PR-event release-upload job is correctly skipped. The tested PR merge tree matches the candidate tree exactly. A post-delivery Harness freshness recomputation reports current=true.

Release handoff: the packaging machines must build and accept Windows x64 CN/INTL and macOS arm64 CN/INTL signed installers from this exact candidate. After the four versioned assets are ready and the independent release review passes, merging release/v1.1.4 into main triggers the existing full Draft Release preflight. Review the Draft notes, installers and checksums before publishing; publication triggers the Linux release upload. Local source tests, simulated pkg loading and Windows temporary-task checks do not substitute for signed-package installation acceptance.

@syzsunshine219

Copy link
Copy Markdown
Collaborator Author

Project Harness review handoff marker. Only reviews submitted strictly after this GitHub timestamp qualify.

@syzsunshine219 syzsunshine219 changed the title release/v1.1.4 release: Memmy v1.1.4 Sep 8, 2026
hijzy and others added 4 commits September 8, 2026 16:55
splitTurn cut a turn on every blank line without repacking, so a single
agentic turn fanned out into hundreds of near-empty tool-call fragments
(one observed turn produced 640 parts, 639 of them a 67-character
"Tool calls: - tool_1"). A codex scan of 1,000 selected turns wrote
19,413 memories.

Restore one turn == one memory. Oversized turns are now clipped on a
UTF-8 boundary via renderTurnClipped so the add-memory request stays
under the wire limit instead of inventing extra memories. Both ingest
call sites (memory runtime and desktop backend) use the same helper and
keep the pre-split legacyTurnRequestId / legacyTurnId idempotency keys,
so watermarks and dedup stay aligned with pre-4eea5f52 memories.

Also bundles AgentSourceCore/dist into the packaged memory runtime and
rewrites workspace deps to file: with --install-links, so the packaged
runtime can resolve @memmy/agent-source-core.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
fix(scan): keep one agent-source turn as one memory
@syzsunshine219

Copy link
Copy Markdown
Collaborator Author

Project Harness review handoff marker. Only reviews submitted strictly after this GitHub timestamp qualify.

@syzsunshine219
syzsunshine219 marked this pull request as ready for review September 8, 2026 13:43
@syzsunshine219
syzsunshine219 merged commit 9814671 into main Sep 8, 2026
5 checks passed
@syzsunshine219
syzsunshine219 deleted the release/v1.1.4 branch September 20, 2026 13:11

This branch was successfully deployed

1 active deployment
release — e967fc3b Deployed Sep 8, 2026 by syzsunshine219 via release #29
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants