perfox

What Powers Perfox ​

When Asha at Acme Diagnostics activates her first agent, she doesn't think about servers, embedding models, or real-time media layers — she writes a personality, attaches a Knowledge Base, and her agent is live on WhatsApp, email, and voice. This page explains what Perfox operates for you underneath, so you know exactly what you're building on.

If you're new here, start with Bring Your Own Provider to understand which connections you bring and which ones Perfox manages for you.

AI models are included and managed ​

The AI models your agents think and speak with are included and managed by Perfox. There are no model keys to bring. Leave an agent on Platform default, or pick a specific model on its AI Model sub-node: a lighter model for speed or a stronger one for complex conversations, plus the models used for voice calls (for example Gemini Live, OpenAI Realtime, or a Sarvam / ElevenLabs speech pipeline). Model usage is paid from your workspace credits.

Perfox also does its own AI work behind the scenes, such as indexing your Knowledge Base for meaning-based search, summarising long conversations, transcribing audio and reading images. That's included too, with nothing to configure.

Storage, semantic search, and web voice are managed for you ​

  • Storage. Files you upload — Knowledge Base documents, call recordings, attachments — are kept safely in managed object storage. There's nothing for you to set up or maintain.
  • Semantic search. Your Knowledge Base gives agents retrieval by meaning, not just keyword match. When you add a file under Connect → Knowledge Base, Perfox splits it into passages and indexes them so an agent can find the most relevant passages for any question. The original file stays untouched — the Knowledge Base is the searchable index over it. Wire the folders you want an agent to draw from on its Knowledge Base settings.
  • Web voice. Voice in the browser works out of the box on any active agent that has voice enabled — Perfox runs and secures the real-time media layer for you, with nothing to configure. Phone calls run over your own Plivo connection, using the numbers you've provisioned.
  • Memory. Agents remember each customer across conversations and channels (facts they've shared, preferences, earlier conversations), with nothing to set up. You control it per agent in the AI Agent's Memory section.

Walked example: a new Knowledge Base file goes live ​

Setup. Asha uploads a PDF of Acme Diagnostics lab reference ranges to Connect → Knowledge Base. She wires that folder to her triage agent on the agent's Knowledge Base settings.

Action. Priya, a patient, messages the agent: "My haemoglobin result was 10.5 g/dL — is that low?" The agent searches the indexed passages for "haemoglobin reference range" and pulls the relevant passage from the PDF.

Result. Priya receives a direct answer comparing her result to the reference range, without Asha having to write a single rule or upload a lookup table.

What just happened. Perfox split the PDF into passages and built a meaning-based index the moment Asha uploaded the file. The agent queried that index at conversation time — no keyword matching, no manual mapping. Asha's only task was uploading the file and wiring the folder.

Agents run reliably across every channel ​

Your agents reach customers over chat, the embeddable website widget, email, SMS, WhatsApp, and voice — all from the same design. The platform is built for high availability and scales elastically under load, so bursts of inbound messages or a wave of concurrent calls are handled without dropping conversations. In-flight chats and live calls are drained gracefully during routine updates so nothing is cut off mid-turn, and voice runs in a redundant configuration so an in-progress call survives an underlying hiccup. Updates ship safely and without downtime, and can roll back instantly if a release ever misbehaves. All of this is operated for you — your agents simply stay responsive.

Your data resides in India (Mumbai) ​

Your workspace runs on managed cloud infrastructure in the India (Mumbai) region for data residency. Your conversations, customers, files, and configuration are private and yours alone. Perfox continuously monitors the platform behind your workspace and responds automatically when something needs attention, so there's nothing to run or patch on your side.

You can now picture what's running behind your agents ​

Perfox manages the AI models, storage, semantic search index, web voice layer, and reliability plumbing — you manage the agent design and the providers you choose to bring. Here's where to go next:

  • Bring Your Own Provider — which connections you supply (phone numbers, email, storage, databases) and which ones Perfox operates for you.
  • Agent Memory — what agents remember about customers and how to control it.
  • Knowledge Base & RAG — learn how Perfox indexes your files and how to wire folders to an agent.
  • Glossary — quick definitions for terms used across the docs.