Stop reading résumés.Start meeting candidates.

Yes — Calibr is BYO-key for the LLM tier. Each organisation adds their Anthropic key in Settings → API Keys and we store it encrypted in Supabase Vault. Operators can opt in to a platform-fallback key as a safety net, but the default is your key, your bill, your scoping.

Every tenant table has Postgres Row-Level Security pinned to get_active_org_id(). The frontend reads via supabase-js (RLS-enforced) and writes via the BFF which double-checks the active org on every request. There is no "view all orgs" mode — even platform admins are scoped to one active org at a time.

PDF and DOCX, single-file or batch (up to 20 at a time). Image-only PDFs are OCR'd via Tesseract before extraction. The parsing pipeline tries PyMuPDF first, falls back to pdfplumber, and ends with a Claude cleanup pass when raw extraction is too sparse to score.

The assistant is a Claude Sonnet 4.6 chat with 17 tools wired into your live data — query candidates, compare two CVs, draft outreach, propose interview slots, save the email to your Drafts, schedule the meeting on your calendar, take notes, and tune the scoring rubric. You stay in the loop: nothing is sent or scheduled without your click.

Calibr ships as a remote Model Context Protocol server. Mint a Personal Access Token in Settings, drop the URL into Claude Desktop, Cursor, or any MCP client, and ask "what's open in Calibr?" — the LLM pulls live pipeline data and can even leave notes on candidates with a write-scoped token.

Every time you reject, hire, or flag a candidate as over/underscored, Calibr captures the signal alongside the candidate's component scores. After 5+ feedback events on a job, the assistant proposes weight adjustments with rationale and confidence. If you've enabled auto-apply and confidence is high, the new weights take effect immediately — and a Postgres trigger queues a rescore on every active session for that job. No manual rescore step.

Three pools, each opt-in per recruiter: Files (read-only access to SharePoint / OneDrive / Google Drive for batch CV import), Calendar (read-write so the assistant can propose slots and schedule), and Mail (compose-only — Calibr saves drafts to your Drafts folder; we deliberately never send on your behalf). You can disconnect any pool at any time in Settings → Connections.

pages.marketing.faq.q8Answer

Yes — Calibr is multi-tenant from day one. Owners invite members with five role tiers (viewer, hiring_manager, recruiter, admin, owner) and can override individual module permissions per user. The audit log captures who did what; the Usage tab breaks down LLM cost per feature so finance has visibility.

We're in early access — pricing is being finalised based on what early users tell us is fair. Today the cost you bear is your own Anthropic API spend (typically a few cents per CV scored, cached aggressively). Reach out via the contact link below and we'll set up your workspace.

Yes — Voice Screening is live for UAE phone numbers today (Vapi + Twilio under the hood). Calibr drafts the script from your job spec, dials the candidate, runs a 2-3 minute structured screener, captures expected salary + notice + language fit, and produces a recommendation of proceed / borderline / reject. Per-org daily cost ceilings, Do-Not-Call list enforcement, and configurable recording retention all ship with v1. Indian and global rollouts are planned for Q2.

English-only at launch. Hindi, Tagalog, and Gulf Arabic are next on the roadmap — each language needs (a) a native-recorded screener script per locale, (b) the Vapi assistant configured with the right Speech-to-Text + Text-to-Speech model, and (c) compliance copy reviewed by a regional recruiter. The UI itself is already trilingual (English / हिन्दी / العربية).

Yes — the assistant ships a generate_sourcing_queries tool. Hand it a job and it produces three ranked Boolean strings tuned to that role (skills, seniority, location hints), three LinkedIn search URLs you can click straight from chat, and 5-10 target-company suggestions based on the preferred-industries you've set. No external crawling, no LinkedIn scraping — just a focused prompt to Sonnet 4.6 that pulls the same job context the scoring agents see.

Yes — the entire dashboard, marketing site, voice-screening detail panels, every settings tab, all status badges and empty states flip between English / हिन्दी / العربية via the language picker in the top bar. Arabic auto-renders right-to-left from the very first server-rendered paint. The cookie is sticky so the recruiter only picks their language once.