Persona

A desktop app for building a structured profile of your work, tracking jobs against it, and tailoring applications to each one.

Persona is a Spiritus application: it supplies configuration and domain assets, and the spiritus package supplies the runtime — window, agent execution, storage, providers, and the JS↔Python bridge. It ships its own UI via AppConfig.ui_dir rather than using the shared chat UI.

Run

uv sync
uv run python run.py

For a source checkout, the bootstrap installs the pinned OpenCode engine and Playwright's Chromium on first launch. The packaged Windows and macOS releases already include both, so they do not require a separate OpenCode installation. No API key is needed to start — the app defaults to a free OpenCode Zen model.

Connecting a provider

Click Settings in the sidebar to paste an API key (Anthropic, OpenAI, …) or pick a model. Credentials are stored in the app-local .opencode-home/, not in ~/.opencode, so this app is isolated from anything else using OpenCode on the same machine. Provider setup is available from the Settings panel.

Layout

Path	Contents
main.py	The whole Core↔app contract: one AppConfig, one run() call.
opencode.json	Agent definitions (profile, profile-pdf, jd-match, job-extract, resume-composer, profile-writer, headline-writer) and the active model.
app_bridge.py	PersonaBridge — extends Core's Bridge with Scanner methods.
scanner/	LinkedIn scan: URL/facet construction, card parsing, feed store.
ui/	The front-end (vanilla JS, Shoelace, Lucide). No build step.
schemas/	JSON schemas for the profile and job records.
workspace/	User data — documents, profile, jobs, browser profile. Gitignored.

Profile imports accept PDF, text, Markdown, and JSON. PDF imports retain the original document plus its extracted selectable text; scanned and password-protected PDFs are reported as unsupported until OCR is added.

How it consumes Spiritus

pyproject.toml uses the sibling Spiritus checkout as an editable dependency, with its bundle extra, alongside Playwright. main.py imports spiritus like any other installed package — there is no sys.path manipulation and no vendored copy of the Spiritus runtime.

The release build delegates PyInstaller bundle assembly and manifest verification to Spiritus. Persona supplies its UI, configuration, OpenCode engine payload, and Playwright browser payload through spiritus.bundle.toml. The resulting bundle ships Spiritus, the pinned OpenCode engine, Chromium, and Persona's resources in one Windows installer or macOS DMG. Installed app data and credentials remain in the platform's normal per-user application-data folder. Public releases use the clean v<version> tag and Persona <version> title; CI build numbers are not exposed in the release name.

Persona also uses Spiritus's update configuration from Settings to check the public GitHub release, download and verify the matching installer, and hand it off to the platform installer. Persona decides when to launch the installer; updates are not applied silently.

Spiritus carries no knowledge of jobs, resumes, or profiles. Everything domain- specific lives here, which keeps the runtime reusable.

Scanner

The Scanner tab pulls jobs from your logged-in LinkedIn session using the shared workspace/browser-profile Chromium profile. Scanned results land in workspace/jobs/scanner-feed.json, separate from tracked jobs in jobs.json, until you promote them.

Currently manual only — a "Scan now" button that reads the recommended feed plus any configured keyword/location searches, deduped on card-level fields. Running on a schedule while the app is closed, desktop notifications, and full per-job detail extraction are all deliberately unbuilt; see the roadmap notes in CLAUDE.md before starting any of them.