Getting started with JobEmber

JobEmber is a pipeline. Five stages, in order, each one consuming what the previous one produced. Here’s what the path looks like end-to-end:

Stage 1 — Search. The job-searcher agent runs against the job sites enabled for your account, returns up to N postings filtered by your target role + location preferences. You pick the ones worth pursuing; the rest are archived (you can un-archive any time).

Stage 2 — Research. For the postings you picked, the employer-profiler agent runs in parallel — one company-researcher sub-agent per company. It produces an “employer profile” with an overview, role needs, and a fit summary based on your CV.

Stage 3 — Edit. Generation reads your CV templates + writing style + the employer profile + the job description, and drafts a tailored CV plus cover letter for each application (a batch click writes every selected application in parallel; a single-app regenerate writes just one). You can edit anything that doesn’t sound right.

Stage 4 — Export. Generate a PDF (or DOCX in ATS mode). The Stage 4 surface also runs the AI-detector and offers an anti-AI rewrite if the document looks too obviously machine-generated.

Stage 5 — Apply. Track that you sent the application. Q&A answers (the form-filler agent) and follow-up notes live here. The Chrome extension can scrape the apply form and prefill it from the documents you generated.

What the agents see

Every agent reads from your profile — target role, work preferences, languages — and your assets — CV templates, cover-letter templates, writing style, voice samples. Settings → Profile + Settings → CV Templates is where you keep these current. The wizard you walked through on signup is exactly the same data, just guided.

What the agents never see

Other users’ data. Every database read is scoped to your userId — no agent of yours can pull another user’s CV, profile, or application history into its prompt. This is structural, not just policy: the MCP tools the agents use don’t accept a userId argument; they always read the calling user’s row.

When things go wrong

The active-runs indicator in the top-right header shows any agent run in flight, on any tab. Click “Stop” to cancel one mid-flight (the agent terminates and the partial work is discarded). Stage-level errors surface on the stage card; they include the agent run ID so support can find the logs.

If you hit something stuck or confusing, the floating Help chat (bottom-right) answers most procedural questions. If it’s actually broken, send feedback from Settings → Send feedback.