Getting started¶

This page takes you from an empty directory to a tailored resume on disk.

1. Install the prerequisites¶

Sira needs Python 3.13 or newer and uv — a fast Python package manager that also runs commands inside the project environment. Every command below goes through uv; there is no pip install step and no virtual environment to activate by hand.

curl -LsSf https://astral.sh/uv/install.sh | sh

Just want to use Sira?

Install the released package instead of cloning:

uv tool install sira    # or: pipx install sira
sira setup

then continue at step 4, writing sira where the guide writes uv run sira.

2. Clone and sync¶

git clone https://github.com/Tiqni/sira
cd sira
uv sync

uv sync reads pyproject.toml and uv.lock, creates a .venv/ directory, and installs the exact pinned versions. It is safe to re-run at any time.

3. Install the browser Playwright needs¶

The job scraper drives a real headless Chromium browser, because many job boards render their posting with JavaScript and return an almost-empty page to a plain HTTP request. Playwright ships as a Python package, but the browser binary is a separate download:

uv run sira setup

sira setup runs playwright install chromium with the same interpreter Sira uses, so the browser build always matches the installed Playwright version.

Skipping this step

Without it, the first tailor run fails while fetching the posting with an error about a missing executable. See Troubleshooting.

4. Set your API key¶

Sira defaults to OpenAI's gpt-5-mini. Export the matching key:

export OPENAI_API_KEY=sk-…

To use Anthropic, Gemini, Groq, Mistral, or a local model through Ollama instead, see Models and providers. Each provider reads its own environment variable.

5. Run it¶

uv run sira tailor <JOB_URL> <RESUME_PATH>

Both arguments are positional — Sira never prompts you interactively for them.

JOB_URL must start with http:// or https://.
RESUME_PATH points at a .md, .docx, or .pdf file. DOCX and PDF are converted to Markdown before parsing.

A real example:

uv run sira tailor \
  https://www.linkedin.com/jobs/view/12345678 \
  ~/Documents/resume.pdf

6. Read the output¶

Files land in output/<company>-<job-title>/:

output/
└── acme_corp-senior_engineer/
    ├── acme_corp-jane_doe.md          ← tailored resume (Markdown)
    ├── acme_corp-jane_doe.pdf         ← same resume as PDF
    ├── acme_corp-jane_doe.docx        ← same resume as DOCX
    └── acme_corp-jane_doe_report.md   ← self-review report

The report is the part worth reading first: it lists what changed, which job keywords your resume covers, which skills you are genuinely missing, and an overall verdict of Strong Match, Partial Match, or Weak Match. See Output and reports for a full walkthrough.

The run also prints the report to your terminal, along with two identifiers: a job ID — a UUID you need if you later want to re-run the tailoring with feedback — and a run ID, which sira resume uses to continue a run that was killed or failed (see the CLI reference).

7. Iterate on the result¶

If the report suggests something, feed it back without re-scraping the posting:

uv run sira re-tailor <JOB_ID> "Put more emphasis on cloud infrastructure work"

re-tailor reuses the stored job posting and your stored original resume, so it costs one fewer scrape and always starts from your original resume — never from a previously tailored one.

Making runs faster and cheaper¶

A default run makes a lot of model calls. Two flags cut that down:

# Speed preset: cheaper model for mechanical stages, lower gate threshold
uv run sira tailor <JOB_URL> <RESUME_PATH> --fast

# Or pick a cheaper model outright
uv run sira tailor <JOB_URL> <RESUME_PATH> --model openai:gpt-4o-mini

Every knob is documented in the CLI reference.

Watching what the agents do¶

By default you get a live dashboard that updates as each stage finishes. To watch the agents' reasoning stream by token instead:

uv run sira tailor <JOB_URL> <RESUME_PATH> --verbose

Why --verbose looks cleaner than the dashboard

The workflow still writes some progress lines with plain print(), which can interleave with the dashboard's live panel in an interactive terminal and garble the display. Nothing is actually wrong with the run — only the drawing. In a non-interactive environment (a pipe, or CI) the dashboard degrades to plain line-by-line logging and the problem disappears.

What happens on the second run¶

Sira remembers. Your original resume is stored in a local SQLite database on the first run, and its parsed form is cached by content hash — so if the file has not changed, the second run skips the parsing model call entirely. Details in Resume memory.