Skip to content

planning codemonkeyninja extraction

Jump to bottom
CodeMonkeyNinja edited this page Jun 22, 2026 · 1 revision

Planning: Extract lenzu/ + lenzu_server/ to CodeMonkeyNinja

Status: planning — decisions locked, ready to execute on user signal Goal: move the shippable Lenzu product (Rust HUD client + Electron server + packaging) to github.com/CodeMonkeyNinja/lenzu, alongside manga-ocr-rs and the user's other published Rust crates. The prototypes/ tree stays at github.com/HidekiAI/lenzu-prototypes (renamed) as a dev-scratch archive.

Locked decisions (2026-05-09)

# Question Answer
1 New repo name under CodeMonkeyNinja lenzu
2 Rename lenzu/ → lenzu_client/ during the move No — keep as lenzu/
3 Preserve history (git filter-repo) vs. fresh init Preserve via filter-repo
4 Fate of HidekiAI/lenzu after the split Prune + rename to HidekiAI/lenzu-prototypes
5 AGPL DBNet ONNX in git history Yes, present — must be stripped from new repo (see §5/§7)

Correction on #5 (verified against current tree)

User recollection was that DBNet was tracked via LFS and pulled from HF at install. Actual state:

  • No LFS is configured (.gitattributes does not exist; git lfs ls-files is empty).
  • assets/stabrise-text_detection_dbnet_ml_v02_model.onnx (4.7 MB, AGPL-3.0) is committed as a regular blob and consumed at build time by scripts/build-dbnet-tarball.sh:24.
  • scripts/lenzu-appimage-installer.sh:75 downloads the AGPL sidecar from GitHub Releases, not HF. Only scripts/setup.sh uses HF — and only for Sarashina, manga-ocr, and PaddleOCR-VL (not DBNet).

Implication: a naive filter-repo carries the AGPL blob into the new MIT-licensed repo. The filter must explicitly invert-strip that path.

0. Naming clarification

The user-facing term "lenzu_client" maps to the directory lenzu/ in this repo (the Rust HUD desktop client). The historical lenzu_client/src-tauri/ crate was a deprecated Tauri prototype, replaced by the Electron-based lenzu_server/. There is no longer a lenzu_client/ directory on disk; Cargo.toml:5 only references it in a comment.

Decision (locked): keep as lenzu/. Rename is a separate, optional follow-up.

1. Why this is not a GitHub "Transfer ownership"

GitHub's repo-transfer feature moves the entire repository tree and history under a new owner. It cannot transfer a subset of paths. So "transferring just lenzu/ + lenzu_server/" requires:

  1. Creating a new empty repo under CodeMonkeyNinja (e.g. CodeMonkeyNinja/lenzu).
  2. Producing a filtered copy of the current repo containing only the paths that move.
  3. Pushing that filtered copy as the new repo's initial history.
  4. Pruning the moved paths from HidekiAI/lenzu (or leaving them as a frozen archive — see §6).

Side benefit: GitHub will still serve old release-asset URLs from HidekiAI/lenzu/releases/... because tag history isn't being deleted on the HidekiAI side. New releases publish under the new org.

2. Scope — what moves, what stays

Moves to CodeMonkeyNinja/lenzu

Path Notes
lenzu/ Rust HUD client crate (the "lenzu_client")
lenzu_server/ Electron server (Node/TypeScript)
packaging/ lenzu + lenzu-models-dbnet package recipes
scripts/ build/install/run scripts (most of them)
.github/workflows/ tag-driven AppImage release pipeline
Makefile top-level build entry points
README.md product README
RELEASE_NOTES.md release-page body
LICENSE MIT (Rust code)
SECURITY.md
assets/ demo gifs, icons referenced by README — except assets/stabrise-text_detection_dbnet_ml_v02_model.onnx (AGPL, stripped per §7)
models/ committed Sarashina ONNX (Apache 2.0) + LICENSE/NOTICE files. DBNet is not under models/ — it lives in assets/ and is stripped
docs/ (subset) technical-design., planning., scores.md, model-evaluation.md, todo-usability.md, sequence-flow.puml — anything that documents the shipped product
build.rs top-level build script (verify it's product-only, not prototype-glue)

Stays at HidekiAI/lenzu (or moves to a renamed archive — see §6)

Path Notes
prototypes/ 17 prototype crates (winit, gtk4, sarashina-onnx, dbnet-test, mecab, manga-ocr-test, etc.)
mirror/ personal mirror dir
notebooks/ Jupyter scratch
Cargo.toml (modified) drops lenzu from members, keeps prototypes
Cargo.lock (regenerated)
docs/ (subset) anything purely about prototypes

Per-file decisions still needed

  • docs/ — audit each docs/*.md and tag M / S / B before filter-repo. docs/scores.md is referenced by both product and prototype READMEs; canonical copy goes to CodeMonkeyNinja, a copy stays at the prototypes repo.
  • scripts/colab/ — likely prototype-related; default S (stay).
  • build.rs, .gitignore, .vscode/, project.code-workspace — duplicate or split as needed.
  • Cargo.lock — regenerated post-extraction; do not try to filter.

3. History strategy — git filter-repo (locked)

Preserves blame, commit history, and tag context for moved files. Rewrites history, so commit SHAs differ from HidekiAI/lenzu's.

# 1. clone a fresh working copy (filter-repo rewrites history destructively)
git clone --no-local git@github.com:HidekiAI/lenzu.git lenzu-extract
cd lenzu-extract

# 2a. KEEP only the paths that move
git filter-repo \
  --path lenzu/ \
  --path lenzu_server/ \
  --path packaging/ \
  --path scripts/ \
  --path .github/workflows/ \
  --path Makefile \
  --path README.md \
  --path RELEASE_NOTES.md \
  --path LICENSE \
  --path SECURITY.md \
  --path assets/ \
  --path models/ \
  --path docs/    # prune prototype-only docs in a follow-up commit

# 2b. STRIP the AGPL DBNet blob from the new repo's history
#     (file is committed at assets/stabrise-text_detection_dbnet_ml_v02_model.onnx)
git filter-repo --invert-paths \
  --path assets/stabrise-text_detection_dbnet_ml_v02_model.onnx \
  --force

# 3. point at the new remote and push
git remote add origin git@github.com:CodeMonkeyNinja/lenzu.git
git push -u origin trunk

# 4. cherry-pick or re-create release tags as needed

Notes:

  • Run on a separate clone, never on the working repo.
  • filter-repo also rewrites tags; review which tags carry over. v0.1.0 release-asset URLs already exist on HidekiAI/lenzu and stay there for back-compat redirect; new releases tag from the new repo.
  • After filter, the new repo's root Cargo.toml must be rewritten to drop prototype members entries (see §4).
  • After stripping the AGPL blob, scripts/build-dbnet-tarball.sh will fail until either (a) the model is fetched into assets/ at build time, or (b) the script is updated to take the path via env var. See §5.

4. Cargo workspace fixup

Current root Cargo.toml (/usr/src/github/mine/hidekiai/lenzu/Cargo.toml) lists lenzu plus 12 active prototype members.

New CodeMonkeyNinja/lenzu/Cargo.toml:

[workspace]
resolver = "2"
members = ["lenzu"]

(plus future siblings as they're added)

Updated HidekiAI/lenzu/Cargo.toml:

  • Remove "lenzu", from members.
  • Keep all prototypes/* entries.
  • Regenerate Cargo.lock (cargo check --workspace).

5. Hardcoded HidekiAI/lenzu URLs

Verified by grep -rn "HidekiAI/lenzu" against the repo (excluding node_modules/target/squashfs-root/lenzu-bundle). References that need updating in the new repo only (HidekiAI side keeps its own copies):

File Line What
RELEASE_NOTES.md 69 "Issues / source" link
README.md 91 link to docs/scores.md
.github/workflows/release.yml 16 comment only — gh calls follow $GITHUB_* env automatically
scripts/lenzu-appimage-installer.sh 37 RELEASE_BASE URL pattern
lenzu/src/client.rs 232, 282 HTTP-Referer header for OpenRouter API attribution
lenzu/README.md 438 benchmark link

Files that reference HidekiAI/lenzu but stay at HidekiAI (prototype READMEs) need no changes:

  • prototypes/dbnet-ocr-pipeline/README.md:169
  • prototypes/manga-ocr-test/README.md:45
  • prototypes/umi-ocr-eval/README.md:21
  • prototypes/paddleocr-vl-manga/README.md:71

6. HidekiAI/lenzu after the split — prune + rename (locked)

Plan: prune the moved paths from history and rename the repo to HidekiAI/lenzu-prototypes.

# fresh clone — never run on the working tree
git clone --no-local git@github.com:HidekiAI/lenzu.git lenzu-prune
cd lenzu-prune

# strip everything that moved (mirror of §3 step 2a)
git filter-repo --invert-paths \
  --path lenzu/ \
  --path lenzu_server/ \
  --path packaging/ \
  --path scripts/ \
  --path .github/workflows/ \
  --path Makefile \
  --path RELEASE_NOTES.md \
  --path SECURITY.md \
  --path assets/ \
  --path models/
# Note: README.md and LICENSE are kept (rewritten on the prototype side)
# and docs/ is kept whole — prototype docs reference docs/scores.md.

# regenerate Cargo.toml: drop "lenzu", keep prototypes/* members
# regenerate Cargo.lock
cargo check --workspace

git push --force-with-lease origin trunk    # destructive — confirm with user

Then via GitHub UI:

  • Rename HidekiAI/lenzu → HidekiAI/lenzu-prototypes. GitHub installs an automatic redirect from the old slug, so existing release-asset URLs (HidekiAI/lenzu/releases/download/v0.1.0/...) keep working.
  • Update the renamed repo's README.md to point at CodeMonkeyNinja/lenzu for the active product.

Old release-asset URLs continue to resolve because (a) release artifacts live on tags, and (b) the rename redirect handles the path change.

7. License audit — verified state and required actions

Verified state (2026-05-09):

Path License In git? Notes
LICENSE (root) MIT yes covers Rust code
models/sarashina2.2-mini-fp16/, …-fp32/ Apache 2.0 yes (committed blobs) upstream Sarashina; OK to keep in MIT repo (Apache 2.0 ⊂ MIT-compatible)
models/LICENSE, models/NOTICE, models/README.md — yes upstream attribution files
assets/stabrise-text_detection_dbnet_ml_v02_model.onnx AGPL-3.0 yes (4.7 MB blob) must be stripped during filter-repo into new repo
lenzu_server/LICENSE (separate) yes verify it doesn't conflict

Required actions before the new repo goes public:

  1. Filter-repo step §3 strips assets/stabrise-text_detection_dbnet_ml_v02_model.onnx from the new repo's history. Verify post-filter with: 
    git log --all --name-only --pretty=format: | grep stabrise || echo "OK: no AGPL blob in history"
  2. scripts/build-dbnet-tarball.sh currently reads the model from assets/<that file> (line 24). Update it to:
    • read LENZU_DBNET_ONNX env var first; OR
    • fetch the model from the existing HidekiAI release on demand. This way the build script works in CodeMonkeyNinja without re-introducing the AGPL blob.
  3. The first CodeMonkeyNinja release that needs lenzu-models-dbnet-*.tar.xz either re-publishes from a CI job that downloads the model into a runner workspace (not committed), or simply delegates to the existing HidekiAI release URL (back-compat redirect after rename still works).
  4. Confirm lenzu_server/LICENSE and models/LICENSE don't introduce restrictions incompatible with the root MIT license.

8. Release pipeline migration

.github/workflows/release.yml is tag-driven and uses $GITHUB_REPOSITORY env vars — it follows the move automatically. The only manual edits:

  1. Comment on line 16 (cosmetic).
  2. Verify the workflow has the secrets/permissions it needs in the new org (GITHUB_TOKEN is automatic; any custom secrets must be recreated under CodeMonkeyNinja org settings).
  3. Cut the first new tag from CodeMonkeyNinja/lenzu (e.g. v0.1.1) to confirm the AppImage build + publish path works end-to-end.

9. Step-by-step execution checklist

When the user signals "go":

  • Pre-flight
    • Audit each file under docs/ and tag M (move) / S (stay) / B (both)
    • Verify lenzu_server/LICENSE and models/LICENSE are MIT-compatible (Apache 2.0 is fine; anything else needs review)
    • Decide whether scripts/colab/, notebooks/, mirror/ should also stay at HidekiAI side (probably yes)
    • Update scripts/build-dbnet-tarball.sh to read the DBNet path from LENZU_DBNET_ONNX env var (so it works post-strip)
  • New repo: CodeMonkeyNinja/lenzu
    • Create empty repo via GitHub UI
    • On a fresh clone of HidekiAI/lenzu, run the two filter-repo commands from §3 (keep + invert-strip AGPL blob)
    • Verify the AGPL blob is gone: git log --all --name-only --pretty=format: | grep stabrise should return empty
    • Rewrite root Cargo.toml → members = ["lenzu"], regenerate Cargo.lock with cargo check --workspace
    • Update the 6 hardcoded URLs from §5
    • Push to CodeMonkeyNinja/lenzu
    • On a clean clone, verify cargo build --release and pnpm --dir lenzu_server install && pnpm --dir lenzu_server build both succeed
    • Recreate any custom Actions secrets at the org level (GITHUB_TOKEN is automatic)
    • Cut a sanity-check release tag (e.g. v0.1.1-rc.0) and verify the AppImage publishes under the new org
  • HidekiAI side: prune + rename to lenzu-prototypes
    • On a fresh clone, run the invert-paths filter-repo from §6
    • Update Cargo.toml to drop "lenzu" from members; regenerate Cargo.lock
    • Confirm with user before git push --force-with-lease (history rewrite is destructive)
    • Rename HidekiAI/lenzu → HidekiAI/lenzu-prototypes via GitHub UI (verify redirect works for old release URLs)
    • Update the renamed repo's README.md to point at CodeMonkeyNinja/lenzu for the product
  • Comms / external
    • Crates.io (if lenzu is published): update repository field on next publish
    • Any external docs/links the user controls (HF model card, blog posts, gists, etc.)
    • OpenRouter API attribution still uses HTTP-Referer: https://github.com/CodeMonkeyNinja/lenzu post-edit — sanity-check that the model providers see the new URL.

CodeMonkeyNinja/lenzu · MIT

Home

README

User docs

Technical design

Desktop issues

Evaluations

Planning & history

Development

Prototypes reference

Clone this wiki locally