Docs/Welcome
Neofir documentation
Neofir is a builder for conversational agents that run on the messaging surfaces your customers already use — Instagram DM, Facebook Messenger, WhatsApp Business, and an embeddable web widget. Configure one personality, plug in a knowledge base, choose the tools the agent can call, and ship.
Start here
Quickstart
Spin up an agent on a single channel in 15 minutes.
Concept
Personality (PETG)
Role, Environment, Tone, Goal, Guardrails — the five fields that shape every reply.
Concept
Tools
Skills the agent can invoke: book a slot, send a quote, escalate to a human.
Concept
Knowledge
Upload PDFs, URLs, or markdown. Retrieved per query, injected into the prompt.
What Neofir is not
Neofir is not a generic chatbot platform. It is opinionated about three things:
- Vertical-agnostic. The runtime has zero literals naming a vertical. Every tenant uses the same primitives. Behaviour differs because the JSON differs, not because of branching code paths.
- Personality over scripts. Agents are configured as PETG (Personality / Environment / Tone / Goal / Guardrails), not as dialog flows. The model handles the dialog; you handle the policy.
- Channels are first-class. The same agent answers across web, Instagram, Messenger, and WhatsApp — without channel-specific prompts. The wire format differs; the agent does not.
Architecture at a glance
The platform splits into four deployable surfaces:
| Surface | Stack | Hosting | Public URL |
|---|---|---|---|
| Backend API | Fastify (Node 20), TypeScript, Zod | Railway, EU | api.neofir.com |
| Dashboard | Next.js App Router, TanStack Query | Vercel, EU | app.neofir.com |
| Widget bundle | Preact + Vite (≤47 KB gz) | Romarg CDN | cdn.neofir.com |
| Landing & docs | Static HTML/CSS | Romarg FTP | neofir.com |
Storage and authentication run on Supabase in EU region. LLM inference is provider-agnostic with Gemini as the current default. Email and messaging are routed through best-in-class providers (Resend for email, Twilio for SMS / WhatsApp). Payments use Stripe.
Next steps
- If you want to ship a first agent today, read the Quickstart.
- If you want to understand the model before configuring, start with Personality.
- If you are integrating the widget into a host site, jump to the Web widget reference.
- For long-form background on how an agent operates across channels, see the Playbook.