Repository navigation
lore
A history of Memongo as told by its own commit log — 219 commits, one author (Rom Iluz), 2026-05-06 to 2026-08-24. See Background for the ADR and benchmark story behind some of these events, and By the numbers for the activity data these eras are built from.
The repository opens with 65d193d chore: initial Memongo release on 2026-05-06. The very next commits, all on 2026-05-12, are dozens of scope-1:, scope-2:, scope-3:, scope-4: prefixed commits carrying explicit task numbers (Task 0.1, Task 0.5, Task 1.-1 through Task 1.9) and "remfix" (remediation-fix) follow-ups against a "Phase 1 Gate 1" checkpoint. This pattern — parallel scoped workstreams merged back with Merge pull request / Merge branch commits — appears nowhere else in the history at this density, and reads like a structured, agent-driven build-out against a written plan rather than organic day-to-day development.
From mid-May onward the commit messages shift toward benchmark integrity: ab8be35 mongodb: add managed atlas runtime checks, ef227e5 benchmarks: enforce strict artifact gates, bf4c9b4 benchmarks: lock full runs behind canary size, 86a6aa6 docs: add mongodb flaws ledger, culminating in 5036ec0 license: switch Memongo to BSL 1.1 and 47b9869 docs: replace unproven benchmark claims on 2026-05-27. Mem0-comparison evidence work continues into June (82bf4c5 benchmark: add mem0 evidence proof pack, 630ad27 benchmarks: disclose mem0 evidence pack, both 2026-06-13/14). The project appears to have spent this era making sure its benchmark claims could survive scrutiny before going public.
A tight two-day burst: ef1d9e9 launch: prepare open source release, 6d6ac17 web: add animated launch landing page, 0fdd7b0 web: add Cloudflare deployment config (2026-06-24), then on 2026-06-25 a sequence of release: commits (093477d release: publish packages with npm, 7a681cf release: make npm publish idempotent, a99d325 release: tolerate published npm versions) and a162a3d launch: clean public release surface / dcf9180 chore: final OSS polish — logger, NOTICE, README, drop internal tags. CHANGELOG.md records this as 1.1.0 (2026-06-24): "Prepared the public Apache-2.0 open-source release."
The largest sustained push in the history. Commit messages carry consistent fix(security): / feat(security): tags: 0b13acc feat(security): enforce scope/scopeRef as a hard tenant-isolation boundary, 6754ed6 feat(security): baseline network hardening — SSRF guard, body cap, rate limit (#28), 356e8da feat(security): retrieval-path prompt-injection defense (#29), alongside engine-correctness work tagged with numbered issues (#30–#71) covering bi-temporal recall, contradiction-driven invalidation, LLM fact extraction, and change-stream robustness. This era closes with bdad0fb release: bump all public packages to 2.0.0 on 2026-07-31. CHANGELOG.md's 2.0.0 entry matches the commit messages closely: "Major release: security hardening, tenant isolation, and engine robustness," including the explicitly breaking tenant-write identity fix.
August opens with 737b9fe pi-extension: add @memongo/pi-extension — Pi coding-agent memory bridge (2026-08-01) and a short pi-extension polish run, then a builder-queue batch (fad3a3f feat: builder-queue contract and surface parity, a5b0473 fix: builder-queue correctness batch, ab68243 chore: builder-queue architecture and deploy, 2026-08-03 to 2026-08-04) — another plan-driven, lettered-batch (B1–B15) workstream in the same style as Era 1's scoped tasks. The month tapers into scattered feat(memory):/fix(memory): hardening commits through mid-August, then closes with the CI/wiki infrastructure that produced this page: 5cfadda ci: add Droid Wiki refresh action (2026-08-23) through 500ac6d ci: use Grove custom model for wiki refresh (2026-08-24, the latest commit in the repo at time of writing).
packages/memory-engine/src/mongodb-manager.ts and the core search/write path trace back to the initial commit (65d193d, 2026-05-06) and have been the single most-touched file in the entire history since — 40 commits in the last 90 days alone (see By the numbers). The manager has effectively been under continuous revision for the whole 4-month life of the project rather than being written once and left alone.
CONTRIBUTING.md states plainly: "Historical, experimental, or comparison material should not be expanded into first-class product scope unless there is a clear ownership decision first." docs/platform/PLATFORM-README.md names the concrete surfaces this applies to: apps/browser-extension, apps/memory-graph-playground, and packages/memory-graph are called out as "not part of the supported product core" — optional or historical surfaces that exist in the codebase's history or periphery but were deliberately kept out of the maintained product.
No commit message at the scale of "rewrite" or "migration" turns up in git log --oneline | grep -iE "rewrite|migrat|refactor" across the 219 commits. The history instead shows continuous, incrementally numbered hardening batches (the #30–#71 issue-tagged fixes of Era 4, the B1–B15 builder-queue batches of Era 5) rather than any single big-bang rewrite of a subsystem.
A 4-month, single-branch, single-contributor history doesn't show much workspace-member growth to trace: the package/app layout visible today was largely present from the early scope-checkpoint commits, with @memongo/pi-extension (2026-08-01) the one clearly new package added after initial release. This section stays short because there isn't more honest signal to report at this project's age.