Batch Venice AI code review for a whole project: each run processes up to N files, saves progress under <project>/.venice_review/, and appends to VENICE_CODE_REVIEW.md until every file is done. Use the CLI or a dark-themed Tk GUI. Optional macOS .app build via ./build_app.sh.
Source: github.com/chrisstampar/Autoreview
Requires Python 3.10+ (see pyproject.toml).
cd /path/to/autoreview
python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev,gui]"Extras: [dev] — pytest, coverage, ruff, PyInstaller. [gui] — Pillow for scripts/make_icon.py. Runtime (pip install -e .) does not require Pillow; the Tk GUI uses the stdlib only.
- Set
VENICE_API_KEYin the environment, or - Run once without it: you’ll be prompted and the key is stored in the system keychain once (service
autoreview-venice, reused for every project folder).
# From the project you want reviewed (or pass --root)
python -m autoreview --root /path/to/repo --batch-size 10
# Entire remaining folder in one run (same as GUI “All”):
python -m autoreview --root /path/to/repo --batch-size 0
# Preview files only (no API calls)
python -m autoreview --root /path/to/repo --dry-run
# Start over: clears .venice_review state AND deletes the existing markdown report
# so the next run does not duplicate sections (then re-reviews everything).
python -m autoreview --root /path/to/repo --reset-progressConsole scripts (after pip install -e .): autoreview, autoreview-gui.
- Continuing a review: New sections are appended to the report. The file is not overwritten. Only files still pending in
.venice_review/state.jsonare sent to the API. - Done: When nothing is left pending, another run only says “Nothing pending” and leaves the report as-is.
- Full redo: Use
--reset-progress, which clears state and deletes the existing report file before writing a fresh one (avoids duplicated## \path`` blocks).
Report shape: Each invocation (each batch you run) is preceded by a ## Autoreview run block with a timestamp, model, batch size label, and the list of paths scheduled for that run—so appended content does not turn into an unlabeled pile of passes. Under that, each reviewed file becomes a ## \relative/path`` section with optional subsections (Security, Code quality, Structure, Performance, Testing, Suggestions). Empty or generic filler (e.g. “no issues”, “N/A”, “OK”) is dropped so only substantive text appears; if nothing remains, the report shows a short no substantive feedback line. The Suggestions list is limited to at most five items per file (strongest severities kept first), and the model is instructed to tie each suggestion to a named symbol or concrete line-visible behavior in that file. A horizontal rule separates files. The top # Venice code review header (root + model snapshot) is written only when the report file is first created (later runs append new Autoreview run sections).
VENICE_CODE_REVIEW.md is never included in discovery, so the report is not reviewed as source.
Discovery caveats: With git, only tracked files are listed (git ls-files). Untracked files are skipped unless you add them. Symlinks to files are followed as normal paths; unusual layouts (nested repos, submodule checkouts) may require adjusting --include / --exclude or using non-git discovery (no .git → directory walk).
Skipped paths (token savings): Files under common tool/cache/build directories are never reviewed (e.g. .ruff_cache, .pytest_cache, .mypy_cache, __pycache__, node_modules, .venv, dist, build, .git, .cursor, .next, .turbo, vendor, Rust target, etc.), and the cache marker file CACHEDIR.TAG is skipped. Root .gitignore and real source are still included.
Documentation-heavy paths (default): To focus on code, Autoreview does not review Markdown / reStructuredText (.md, .mdx, .mdown, .markdown, .rst) or anything under a doc, docs, or documentation directory segment. To include those, use --review-markdown or set AUTOREVIEW_REVIEW_MARKDOWN=1, or enable Include markdown & docs folders in the GUI.
The GUI model list is curated from Venice text models (code- and reasoning-capable picks). Set VENICE_MODEL to any valid chat model id if yours is not in the dropdown.
autoreview-gui
# or
python -m autoreview.guiBrowse for a folder, pick a category (Recommended, Fast coding, Long context, etc.) and model, set batch size, optionally check Include markdown & docs folders, then Run review. Repeat until the log says all files are done.
Usage & spend (GUI): Below the subtitle, a line shows cumulative prompt/completion tokens for Autoreview runs on that folder (stored in .venice_review/state.json). With an API key, it also shows an estimated USD from GET /models pricing.
./build_app.sh # builds dist/Autoreview.app in this repo
./build_app.sh --install # same, then copies to /Applications/Autoreview.appOr copy manually: cp -R dist/Autoreview.app /Applications/
First launch may require Right-click → Open if Gatekeeper complains about an unsigned developer build.
Add:
.venice_review/Optionally ignore VENICE_CODE_REVIEW.md if it contains sensitive snippets. .env (basename) is never sent to Venice during discovery.
- Never commit Venice API keys,
.envfiles, or review reports that contain secrets. This repository’s.gitignoreexcludes common cases (e.g..env,VENICE_CODE_REVIEW.md); keep using env vars or your OS keychain for keys. - Code you review is sent to Venice’s API under your account; scope and network policies are your responsibility.
Sending source files to Venice is your decision; treat it like pasting code into any cloud API.
Regenerate assets/app_icon_256.png and Autoreview.icns with:
python scripts/make_icon.pypytest -q
# Coverage (optional)
pytest -q --cov=autoreview --cov-report=term-missingLint (optional): ruff check autoreview tests
After merging this repo, configure the remote (requires GitHub CLI gh auth login once, or use the web UI).
| Goal | What to do |
|---|---|
| Repository description | Repo → ⚙ Settings (or About → ✎). Paste a short description (≤350 chars), e.g.: Batch Venice AI code review — resumable batches, one markdown report, CLI + dark GUI, optional macOS app. Skips caches and env files by default. MIT. |
| Topics | On the repo main page: About → ⚙ → add topics such as venice-ai, code-review, python, tkinter, cli, pyinstaller, developer-tools. |
| Dependabot | .github/dependabot.yml enables weekly PRs for pip and GitHub Actions. Merge it, then watch the Pull requests tab. |
| Release | Tag v0.1.0 is published on main. In Releases → Draft a new release, choose tag v0.1.0, title Autoreview 0.1.0, add notes, publish. Or: gh release create v0.1.0 --verify-tag --generate-notes |
gh one-liners |
gh repo edit chrisstampar/Autoreview --description "Batch Venice AI code review — resumable batches, markdown report, CLI + GUI, optional macOS app. MIT." and gh repo edit chrisstampar/Autoreview --add-topic venice-ai --add-topic code-review --add-topic python --add-topic tkinter |
| Branch protection | Settings → Rules → Rulesets (or Branches → Branch protection): protect main, enable Require status checks to pass, and select the CI workflow jobs (e.g. test on Python 3.10 and 3.12) after at least one successful run. Skip “required reviewers” if you’re the only committer. |
MIT — see LICENSE.
| Issue | What to try |
|---|---|
| 429 / rate limits | The engine retries with backoff; wait and run again with a smaller --batch-size or add --delay-ms. |
| Keychain prompt fails | Install keyring (pip install keyring) or set VENICE_API_KEY in the environment. |
| GUI fonts look wrong | The app picks SF Pro / Segoe UI / DejaVu by platform; install a standard UI font or run from a terminal to see errors. |
| Stop in GUI | Stop requests cancellation before the next file; the current API call always finishes. Progress is saved in .venice_review/. |
| CLI Ctrl+C / SIGTERM | Same cooperative cancel as the GUI: first interrupt stops before the next file (in-flight HTTP may still finish). Exit code 130 when cancelled. |
| State save frequency | Default: flush every 5 completed paths. Set AUTOREVIEW_STATE_SAVE_EVERY=1 for every file (more IO). |
| Slow / stuck file reads | Per-file read uses a thread with timeout (default 120s). Set AUTOREVIEW_READ_TIMEOUT_SEC to adjust (1–3600). |
| PyInstaller build | Run ./build_app.sh on macOS only; use ./build_app.sh --fast to skip wiping dist/ between iterations. |
VENICE_CODE_REVIEW.md unchanged on another run |
Expected if nothing is pending (every discovered file is already in state.json) — the log explains this; the report is only appended when new files are reviewed. Also: a batch where every path was skipped (binary / too large / timeout) adds no sections. Reload the file in your editor. If you once used --output, later CLI runs must use the same --output, or omit it and let Autoreview use output_name from .venice_review/state.json. Full redo: --reset-progress. |