Quickstart

Sumisura runs on your machine. Nothing is hosted, there is no account, and it calls the Claude API with your own key.

What you need

Requirement	Why
Docker + Docker Compose	Runs the web app (Go backend + React frontend)
An Anthropic API key	Selection, Rewrite, cover letters and research calls (Claude API, pay per use)
Claude Code	Runs the `tailor-cv` skill, which drives the tailoring pipeline
Typst on your `PATH`	Renders the PDF
`pdftotext` (poppler-utils)	Optional — the ATS-parsability check

1. Clone it

git clone https://github.com/gio-del/sumisura.git
cd sumisura

Pin yourself to a release rather than tracking main if you would rather upgrade deliberately:

git checkout v0.1.0

2. Add your API key

cp .env.example .env
# then set ANTHROPIC_API_KEY=sk-ant-...

Without a key the app still starts and you can browse and edit Master Data — only the calls to Claude fail.

3. Start the app

docker compose up

Backend: http://127.0.0.1:8080
Frontend: http://127.0.0.1:5173

Both are bound to localhost with no authentication, on purpose. See LAN-reachable mode if you want to reach it from your phone.

4. Replace the stub data with your own

data/ ships with stub content for a fictional “Jane Doe” so the app has something to show on first run. Replace it with your own career history:

data/profile.yaml — contact details, education, publications, awards, languages
data/experience/*.md — one file per job or client engagement
data/projects/*.md — one file per project
data/cover-letter-snippets/*.md — optional reusable cover-letter paragraphs

You can edit these in the web app, or in your editor — they are plain Markdown with YAML frontmatter, and both paths write the same files. See Writing your Master Data for the shape of each file.

5. Install the tailoring skill

The pipeline is a Claude Code skill served from this repo’s own plugin marketplace:

claude plugin marketplace add .
claude plugin install sumisura@sumisura-local

6. Tailor your first CV

In Claude Code, from the repo:

/sumisura:tailor-cv

Give it a job description (pasted text or a URL), point it at an Application you already track, or give it nothing at all for a general-purpose CV. It walks Selection → Rewrite → Text Review (you approve) → Render → Visual Review (you approve), and writes everything into output/<label>-<timestamp>/.

That directory holds cv.pdf, cover-letter.pdf, the assembled data.json, and selection.json — the record of which Entries were chosen and how each bullet was rewritten.

How it works — the vocabulary, and where the guarantees are
Tracking applications — Job Listings, Applications, the extension
Configuration — models, costs, environment variables