Explore

Overview

AgentX Architecture

⁠

🏗️ 1. Agent Builder: Setting Up the Brain

This is where you construct what the agent knows, how it speaks, and what it can do:

Knowledge: Import product FAQs, policy docs, and guides.

Channels: Decide where the agent lives — email, Instagram, telephony, etc.

Actions in Tools: Let the agent perform tasks via CRMs, calendars, databases.

Core Voice Capabilities: Define TTS/STT models and other telephony settings.

LLM Orchestrator: Connect and configure the language model(s).

Prompt Management: Set the goals, tone, and response structure for the agent.

Guardrails & Human Handoff: Define fallback routes, error handlers, and escalation logic.

All these plug into an internal Agent OS that orchestrates everything.

⁠

🔁 2. Workflow Engine / Multi-Agent Orchestrator

Once your agent is ready, you connect it to a workflow engine. This is the process brain:

Tells the agent what to do when, across multiple steps and possible conditions.

Can involve multiple agents or tools working together in sequence or logic-driven paths.

⁠

🧪 3. Evaluation: Testing If It Works

You simulate and validate the agent:

Define Scenarios (ex: "What if the customer wants to return a damaged product?")

Capture Metrics (ex: resolution rate, time to respond)

Get a Result (pass/fail or score based on real or synthetic conversations)

⁠

📊 4. Reporting: Monitoring Over Time

Once agents are live, track performance:

Conversation Logs: What was said, done, and when.

Business Funnel: How many leads converted, how many tickets resolved, etc.

Agent Performance: Compare different agents or use cases.

Qualitative Insights: Understand pain points, edge cases, or surprises.

⁠

Temp Bird’s eye

⁠

🔍 What is Nurix.ai?

Nurix.ai is building AgentX, a platform to create intelligent, multi-modal, voice- and text-based conversational AI agents that can operate across different functions of a business — customer support, sales prospecting, supplier coordination, and HR outreach. Think of it as the “operating system + workflow designer + orchestration layer” for AI agents embedded into enterprise processes.

The ambition is to replace repetitive human communication workflows by giving companies powerful, flexible agents that behave with intelligence, memory, and integration — across channels and use cases.

⁠

💡 Why Now?

B2B AI adoption is accelerating: Over 56% of AI use cases in B2B are tied to customer support — a primary wedge Nurix starts with.

Channels are fragmented: Customers communicate via email, WhatsApp, Instagram, voice calls, etc. Existing tools aren’t natively cross-channel or fluid.

LLMs are powerful but messy: Most businesses don’t want to wrangle prompts and OpenAI APIs — they want reliability, guardrails, and process adherence.

Voice is back: With AI-native voice stacks improving, conversational agents are now ready for nuanced tasks on phone or WhatsApp calls — beyond old-school IVR trees.

⁠

🛠️ What Does Nurix.ai Offer?

1. Agent Builder (Studio)

A modular platform to construct conversational agents that can:

Pull from a knowledge base (e.g. FAQs, product manuals, drive files)

Talk across channels (email, WhatsApp, Instagram, phone/SIP)

Take actions via tools (CRM updates, calendar bookings, API calls)

Use voice capabilities (TTS, STT, voice activity tuning)

Be instructed via LLM prompts and configured with fallbacks and handoffs

These agents live inside an Agent OS — a runtime hub that orchestrates messaging, voice, behavior, and execution logic.

2. Workflow Engine / Orchestrator

Once built, agents can be plugged into workflow logic: a visual canvas for building flows with:

If/else conditions

Multiple agents

API calls or tool integrations

Guardrails

Human escalation triggers

Global fallback behaviors

This lets companies go beyond isolated chatbots and design cross-functional, multi-step agentic processes.

3. Evaluation

Before going live, agents can be tested using:

Scenarios (scripted flows to simulate real-world queries)

Metrics tracking (success, fallback rate, escalation rate)

Results: data to validate agent readiness

4. Reporting

After deployment, companies can monitor:

Conversation logs (context, transcripts, channel-level detail)

Funnel analytics (drop-offs, conversions)

Agent performance (per use case, channel, goal)

Qualitative insights (manual review, failure types)

⁠

🧠 What Makes Nurix Different?

🟩 1. Multi-Channel + Voice-First Native

Unlike traditional chatbots that live in web widgets, Nurix supports telephony, social messengers, email, and switching between them. Upload a doc on WhatsApp mid-call? That’s part of the design.

🟩 2. Designed for Internal Customisation

Most products hand over bot-building to the customer. Nurix offers done-for-you agent creation + ongoing maintenance, because agent behavior is deeply linked to internal SOPs, tone, and context. Over time, this could become a premium service model (50-50 SaaS vs Services revenue split, not 80-20).

🟩 3. Visual Workflows + Prompt Abstractions

Instead of relying on complex prompt engineering, Nurix offers:

A flowchart-style workflow studio for CSMs/ops teams

Underlying agent block creation by more technical folks

Global prompts, fallback nodes, and escalation triggers — packaged as reusable primitives

This allows both low-code and pro-code creation, targeting different personas within an enterprise.

🟩 4. Granular Human Handoff

Agents aren’t islands. Nurix supports customizable escalation triggers — defined by conditions in the workflow or agent behavior — to API into a real human handoff channel (e.g. Zendesk, Twilio, Slack DM).

⁠

🧱 Product Design Philosophies Behind Nurix

AI as a Process Primitive Agent is not a chatbot. It's an executional actor that belongs inside a business process — like a UI-less employee.

Separation of Concerns Allow technical and non-technical stakeholders to collaborate without bottlenecking each other — workflows vs agents.

Reliability > Brilliance Focus on guardrails, context, and clear fallback behaviors. Enterprise adoption needs predictability more than surprise.

Service-Enhanced SaaS For now, agents will be built/maintained by Nurix to ensure depth and accuracy — this feeds into learning loops to improve the product itself.

⁠

🔮 Where It’s Going

GT Market Focus: Mid-market+ ($50M+ ARR) retail and insurance players in the US.

Use Cases:

Retail: customer support, product queries, returns, order issues

Insurance: FNOL (first notice of loss), claims initiation, lead qualification

India as a secondary market (founder connections, cost arbitrage, reputation flywheel)

Future roadmap includes:

Real-time agent monitoring

Agent cloning/templates by vertical

Embedding agent outputs into internal dashboards (BI / ops)

⁠

AgentX Bird’s Eye

recap includes:

tl;dr

features (agents, studio, monitoring)

roadmap (future features, stakeholders)

value prop

usecase (persona, journey, impact)

differentiation (landscape + usp)

GTM

audience - archetypes (industry, geography, ticket size, channels, behaviour, needs + goals) (primary + secondary + tertiary)

1. understand: product, market, audience

product - features, flow, usecase, value proposition

market - landscape (overview, features, audience, positioning)

audience - archetypes (industry, geography, ticket size, channels, behaviour, needs + goals) (primary + secondary + tertiary)

concept refinement

audience refinement

Want to print your doc?
This is not the way.

Try clicking the ··· in the right corner or using a keyboard shortcut (

CtrlP

) instead.