Vita Tree

Browse, edit, and adapt LaTeX CVs across Git branches. Compare versions side by side, keep a job description with each branch, and preview the compiled PDFs.

Install and run

Requires Python 3.10+, Git, and a Git repository containing a LaTeX CV. Install the CLI in an isolated environment with pipx:

pipx install "vita-tree[ai]"
vt run /path/to/your/cv-repository

Or use uv:

uv tool install "vita-tree[ai]"
vt run /path/to/your/cv-repository

vita-tree is also available as an alias for vt; python -m vita_tree works inside the environment where the package is installed. Use pipx upgrade vita-tree or uv tool upgrade vita-tree to update a published installation.

The optional [ai] extra installs the compatible VITA CLI and enables provider and reasoning integration automatically. Install plain vita-tree if you only need browsing, editing, and PDF previews. AI calls still require your VITA configuration and provider credentials or Codex login. See VITA for CV initialization and provider setup. VITA can also be used from a local checkout with --vita-source.

PDF previews: install MiKTeX (Windows) or TeX Live (Linux/macOS), including your template's packages and fonts, and put its executables on PATH. Git and TeX are system requirements and are not installed by pip. Browser assets, PDF.js, fonts, and character maps are bundled; no Node installation or CDN is needed at runtime.

To run from source:

git clone https://github.com/ABDELLAH-Hallou/vita-tree.git
cd vita-tree
python -m pip install -e ".[ai]"
vt run /path/to/your/cv-repository

On Windows use a path such as vt run "C:\path\to\cv-repository".

vt run [REPOSITORY] [--port 8765] [--file main.tex] [--no-browser] [--vita-source PATH]

The server accepts localhost connections only. It can modify the chosen repository, so do not expose it through a network proxy. Compile only trusted local LaTeX: disabling shell escape is not an operating-system sandbox.

Workflow

The LaTeX tab includes syntax colors, line numbers, bracket matching, undo/redo, and search (Ctrl/Cmd+F). Live syntax checks underline parser errors; click the status below the editor for details. Checks report the first syntax error and do not validate custom macro expansion or installed packages; compile the PDF for full validation. Unsaved and temporarily invalid source can still be saved. The editor is bundled locally and follows the selected light/dark theme.

1. Open a branch. Existing checked-out branches display working files, including uncommitted main.tex, job.md, and .vita/results/*.md. Other branches display committed files.
2. Choose Branch from this, name the branch, and paste the job description. Vita Tree creates a Git linked worktree below the repository's Git common directory at vita-tree/worktrees/. The original checkout is not switched. The viewed main document and the new job.md are written to the new worktree.
3. In Structure, select/copy text normally, or double-click text in the new version to edit it. The parser maps text to its exact source range and escapes LaTeX special characters. LaTeX provides the full editable source for other changes. Save writes the branch worktree's real file.
4. In Job & AI, save the job description and select a provider, model, reasoning level, and language. Use Run all for every available AI step, Run selected for any checked subset, or select a graph node and choose Run … only. Each enabled step runs in order in an isolated snapshot. Different branches still run concurrently, without a queue.
5. Follow the Application workflow graph: extract → analyze → adapt → AI review → cover letter → validate. Select a task for its status, duration, output, and independent run action, or choose an earlier run. Actions always use the current saved branch files, even when inspecting an older run. Skipped tasks make no AI calls; failed tasks block downstream work. The validation task checks source structure and conflicts, not PDF compilation. Review the Markdown reports, changes, and compiled PDFs before committing.
6. Use Commit & push to review the destination, commit message, and included files, then publish only the selected branch. A failed push leaves the commit safely in the worktree and offers a push-only retry.

Save changes remains local and uncommitted. Commit & push includes only the displayed main document, job.md, the generated cover-letter.tex, and reviewed VITA Markdown/JSON/TOON reports; unrelated working-tree changes are reported and excluded. It never force-pushes or publishes other branches. Worktrees containing local changes persist across server restarts; leaving a clean worktree releases its branch for CLI checkout.

For existing branches without a recorded baseline, Original means that branch's committed CV. New branches retain the viewed base CV; AI runs retain the pre-run CV. Reload imports changes made by a terminal/editor. Saving with a stale revision returns a conflict and preserves the user's editor text.

VITA's local config.json, extensions.json, cv.yaml, and task Markdown files under .vita/skills/<step>/ are supplied to isolated AI runs. API-key/token values from .vita/.env are passed in child-process memory. Existing reports are branch-local. The UI does not expose credentials or raw provider logs.

Provider/model dropdowns scan .vita/extensions.json on every local and remote branch ref and in linked working trees. Duplicate choices are combined; hovering over Model lists their sources. Reload rescans the repository. Choosing a model from another branch overrides only the isolated run, leaving your branch configuration intact. These are discovered configurations, not a guarantee that historical model names remain available to your account.

Background AI runs: use one vt run server and port for all branches. Every workflow action starts an independent background process immediately, with its own provider, model, reasoning, language, and isolated files. Codex runs use separate CLI invocations; API providers make independent requests through VITA. There is no Vita Tree queue or concurrency cap. Each run executes its enabled tasks sequentially; different branches run at the same time. Only the running branch stays locked for edits and publication. Provider rate limits and machine resources still apply.

Leaving a branch (back to the list, opening another branch, or closing the page) releases its clean, idle Vita Tree worktree so the branch can be checked out from the CLI. No force removal is used. Running operations, uncommitted changes, untracked files, and ignored files keep the worktree intact; the UI explains when local files prevent release. Commit or move those files, then open and leave the branch again. Worktrees created outside Vita Tree and the original repository are never removed.

Use the sidebar Running, Needs review, and Needs attention filters to find active work and results. Open a branch for its workflow progress and run history. Branch badges and persistent in-app notifications show results waiting for review. Selecting a result opens its branch's Job & AI view with that run's retained reports and source files, including its original CV. Reviewing a result clears only that run's notification. The opportunity label is the branch name; job description text and provider output are not included in the global status list.

Run history and review state survive browser refreshes and server restarts. A server restart marks unfinished work as interrupted; it does not automatically resume or resend AI requests. Start a new run after reviewing the branch. Keep one Vita Tree server per CV repository. No desktop-notification permission is needed.

With the [ai] extra or --vita-source, the child-process adapter makes environment credentials available to VITA and forwards Codex reasoning as an explicit -c model_reasoning_effort override for all selected AI steps. Available levels come from the selected model's entry in $CODEX_HOME/models_cache.json (default ~/.codex). Models absent from that catalog and non-Codex providers offer Provider default only; unsupported settings are rejected by the server. No parent VITA source files or global Codex settings are changed.

Extraction and structured evaluation reports require Vita CV 2.5+. The UI detects task prompts in the connected installation and disables unavailable options; older integrations can still run Analyze and Adapt together or independently. The selected provider/model applies to every enabled step, overriding per-stage routing only in that run. Analyze, Adapt, and available AI review are selected initially; every AI step can be unchecked. Run all includes all supported AI steps regardless of checkbox selection. Validation runs automatically before applying outputs and can also run alone, without a job description or AI provider. Single steps never launch other AI steps. Analyze, Review, Extract, and Cover leave the CV and comparison baseline untouched. Review uses an existing matching structured analysis for score deltas, or evaluates the CV on its own when none exists. Matching extraction and analysis are captured at launch; analysis Markdown without job provenance is not reused. Running Analyze again clears the previous score delta; unrelated reports remain available. The current UI requires tex_entry=main.tex and, for cover generation, cover_tex_entry=cover-letter.tex. Cover-letter source is available in retained run files; the PDF tab previews the CV only.

Vita CV 2.5 writes analysis.md/analysis.json, review.md/review.json, and comparable delta.md/delta.json reports. Vita Tree retains these along with job.json and job.toon. Reports render as readable Markdown with tables; raw HTML is treated as text. Historical runs without recorded task progress display Not recorded, never inferred success.

The Appearance selector in the header offers Light, Dark, and System. Your preference is saved in this browser; System follows OS appearance changes. Reading pages and compiled PDFs remain white in dark mode.

For a pipx editable installation, source changes take effect when you restart vt; no reinstall is needed. To reinstall its environment explicitly, run pipx reinstall vita-tree, then restart the command above.

What is supported and what is limited

Capability	Status
Centered, responsive branch list and original/new comparison	Implemented
Actual LaTeX syntax parsing with positions	pylatexenc AST, not regex stripping
Common resume.cls content: names, addresses, rSection/rSubsection, formatting, bullets, tabularx	Structured reading view supported
Text selection and mapped editing	HTML reading view and PDF text layer; edits made in Structure/LaTeX update the source
Exact fonts, spacing, line breaks, page breaks	Compiled PDF is authoritative
Arbitrary macro expansion / exact HTML reproduction of every LaTeX package	Not supported; use PDF and source editing
Editing directly inside the compiled PDF	Not supported; PDF is read-only, then recompiled after source edits
\\input / \\include, custom classes, images	Included in compilation when present in the repository snapshot; included files are not visually editable in Structure
Multi-file uncommitted changes when creating a branch	Only the displayed main document is copied; other files come from the base commit
Bibliography, makeindex, minted/shell-escape builds	Not orchestrated; current compiler runs the selected TeX engine twice
Guaranteed perfect PDF reading order for arbitrary templates/fonts	Not possible; depends on the PDF's text encoding and layout
Successful AI generation for every provider	Depends on VITA credentials/provider availability; failed runs do not replace the CV

Structure preserves unsupported source and explicitly reports unknown commands/environments. It is a reading/editing aid, not a TeX engine. Source edits can temporarily be invalid; the UI reports parse/compile failures instead of pretending a stale PDF is current. AI results are syntax-checked but still require PDF compilation and factual review.

PDF previews and AI snapshots are stored under the repository Git common directory's vita-tree/pdf/ and vita-tree/runs/. AI input text is retained in vita-tree/run-inputs/, and run status and review state in vita-tree/job-history/. These can accumulate; keep the server stopped before manually cleaning old cache folders. Do not delete worktrees/ directly; use Git's worktree commands.

Development and verification

python -m pip install -e ".[test]"
python -m unittest discover -s tests -v
python -m playwright install chromium
python tests/browser_check.py --browser chromium
python tests/browser_jobs.py --browser chromium
python tests/browser_workflow.py --browser chromium

Browser tests cover editing, branch creation, provider settings, themes, conflicts, and real PDF compilation and text selection. They use synthetic CVs and a fake AI provider. Screenshots and PDFs are saved in ignored .test-artifacts/.

PDF.js is pinned in package-lock.json; browser assets and its Apache license are vendored into the Python package. Node is only needed to refresh these assets:

npm ci
npm run vendor

Use Node 22.13+ or 24+ when updating PDF.js. The application itself runs on Python 3.10+.

Releases

Use feat: and fix: commit messages. Release Please opens a PR updating the version and changelog; merging it creates the release tag. Code, browser/PDF, package installation, and privacy checks run before TestPyPI and PyPI uploads. Publishing uses PYPI_API_TOKEN in the pypi environment, TEST_PYPI_API_TOKEN in testpypi, and the repository's RELEASE_PLEASE_TOKEN.

Author

Created and maintained by Abdellah HALLOU, the project's main contributor. Contact: abdeallahhallou33@gmail.com.

Vita Tree is MIT licensed. Bundled PDF.js components retain their original licenses; see third-party notices.