How to use LLM Fusion
LLM Fusion is a governed AI gateway for professional firms: your team chats with 300+ frontier models, but names, contact details and other identifying data are masked before any model sees them, and every exchange lands in a tamper-evident audit log. This page is the complete start-to-finish guide. An agent-readable copy lives at /llms.txt.
1. Create your organization
Sign up at web.llmfusion.app/register with your work email, then confirm the verification email. Your first login creates your organisation automatically and grants $5 in trial credits — no card required. You are the organisation's admin. Nothing is billed without your explicit top-up.
2. Talk to your team
The chat lives at web.llmfusion.app. What matters in the model picker:
- Direct — pick any model from the catalog (GPT, Claude, Gemini, DeepSeek, Llama and many more). Requests pass through the anonymization layer first.
- Auto — four strategy presets (Balanced, Lowest Cost, Speed, Quality); Fusion routes each message to a suitable model and bills at pay-as-you-go rates.
- Council — three models answer independently, then review each other's anonymised answers (no model can favour a brand name), and a chairman model synthesises the final verdict. Use it when the answer carries weight: a client memo, a risk call, a valuation assumption.
Identifying entities — people, organisations, addresses, phone numbers, emails — in English and Arabic are de-identified before provider calls and restored on the way back. Detection is probabilistic, which is why you should still treat any AI output as a draft: it removes the identifiers from the wire, and the audit trail shows what happened.
3. Invite your team (Org Console)
Open Org Console from the sidebar (or web.llmfusion.app/console). Admins can:
- Invite members by email — they join your organisation with one click, no separate signup friction.
- Choose roles: admin (full console, invites, spending) or member (chat + granted models).
- Set per-member model access: by default members can use the full granted catalog; you can narrow access for specific people.
- Watch per-member usage so spend is attributable.
4. Credits, top-ups and plans
Usage is credit-metered. Add credits at web.llmfusion.app/billing — fixed packs or any custom amount from $5 (card via Stripe). Org plans start at $50/month for small teams; pay-as-you-go works without a subscription. Every credit movement is recorded in your organisation ledger.
5. The API (for developers)
Programmatic access uses per-organisation API keys, created in the Org Console. The gateway is OpenAI-compatible: point any OpenAI SDK at https://api.llmfusion.app/v1 with your Fusion key and existing code works unchanged — while anonymization, model policy and audit apply underneath. Pay-as-you-go pricing, no subscription needed for API-only use.
6. Governance and audit
Every request is recorded — who asked, which model answered, how much it cost — in a hash-chained log: entries are cryptographically linked so any tampering or deletion is detectable when replayed. That is the artifact you show a client, a partner, or a regulator when they ask what your firm's AI actually saw and said. Deny-first posture: models and tools are unavailable until granted.
Our own house rules: site and app analytics are cookieless (no cookies or local storage, IP masked, no click/screen capture) — details in Privacy §9.
7. What is deliberately not here yet
Honest limits, so you plan right: single sign-on (Okta/Entra) is not offered — onboarding is deliberately fast instead, and SSO is on the roadmap. Bulk audit export is not yet self-serve. Compliance certifications (ISO/SOC 2) are not claimed. The anonymizer masks identifiers; it does not make legal review unnecessary.