Joblit pulls Australian roles into one workspace and builds the exact tailoring prompt for each JD. Paste it into the chatbot you already use, paste the result back, export clean LaTeX PDFs — Joblit's servers hold no model keys.
Ship accessible, high-performance UI across the platform.
Partner with design-systems and product teams end to end.
The workflow
Pull fresh roles into one list with a single run. Duplicates and dead links are filtered before you ever see them.
Scan each ad with the fine print already pulled out, and keep only the roles worth your time.
Copy the prepared prompt into any chatbot, paste the JSON back, review, publish. Only the summary and the skills order change — never your bullets.
Every draft comes back for review, then compiles to a clean LaTeX PDF — named, filed, ready to send.
Architecture
Joblit builds the exact prompt and checks what comes back — the writing happens in whatever chatbot you already pay for. The servers hold no model key and cannot call a model. That is enforced by the architecture, not promised by a policy.
Builds the exact prompt from your resume profile and rules — and holds no model keys.
Any assistant you already pay for — Claude, ChatGPT, Gemini.
Joblit never sees your model account.
Summary lint and index resolution — fabrication is rejected, not detected.
Finalize is the only render. Clean LaTeX, in English and Chinese.
AI, kept honest
Joblit reads every job ad and pulls out the hard asks — 8+ years of Java, security clearance, work rights — as tags you can click to jump to the exact sentence in the ad. No AI guesswork here: the same ad always reads the same way.
Tailoring rewrites your summary for the role — then three checks run before it can land. Fail any one, and the paste is rejected with the reason named.
For skills, the model can only answer with positions in your own skill bank — it never writes a skill name. A position that doesn't exist is rejected outright.
The model returns positions, never names — it cannot invent a skill.
Every claim above is how the code is built — and the code is public.
Yes. Sign in with GitHub or Google and everything works. Generation runs on the AI subscription you already have, so there is no per-token bill from us and no paid tier hiding the good parts.
Your jobs, resume profile, and drafts live in your Joblit workspace. Generation happens in your own chatbot, on your own account — the servers hold no model key and cannot call a model. That is enforced by the architecture, not promised by a policy.
There is nothing to install and nothing to run. Joblit prepares the exact prompt; you paste it into whatever chatbot you already use and paste the JSON back — that loop is the product, not a workaround. If you use Claude Code, an optional Skill Pack automates the same loop with your own rules.