We build the AI your customers and teams use, and everything it runs on.

Assistants that serve your customers and help your staff, with every action checked, confirmed and recorded. The platform they run on. The models and infrastructure underneath, built and run with our partners.

Tell us what you need How it is controlled

Confirmation before every action
A record of every step
Sign-in through your systems
SOC 2

Swipe to see each example.

Start with what you need

Start anywhere in the stack, from the assistant to the GPU.

Come to us for the AI people use, the middleware it depends on, the models behind it or the infrastructure underneath. We do one layer or all of them, with partners where it makes sense.

For your customers

Serve every customer, on every channel, at any hour.

Your products, prices and policies behind one assistant on your website, WhatsApp, Instagram, SMS, ChatGPT and Claude. It answers, books and takes payment. Your team steps in when it should.

AI for customers

For your employees

Give every team answers, and a hand with the work.

Assistants for HR, IT, operations, finance and sales, inside the tools your staff already use, such as Slack, Microsoft Teams and Google Chat. They answer from approved policies with the source shown, and prepare work for people to approve.

AI for employees

For regulated industries

Private AI, with every layer in your own environment.

For financial services, healthcare, insurance and the public sector. Models, platform and assistants run where your data has to stay, and every action is confirmed and logged.

Financial services Healthcare Other industries

For technology providers

Turn AI capacity into services your customers buy.

For telcos, data centres and cloud providers. We build the layers above your compute, put your own AI to work on them first, then help you sell the same solutions to your business customers.

For technology providers

For software companies

Put your product inside ChatGPT and Claude.

Your users already ask their assistant to get work done. We build the MCP servers, apps and connectors that let them use your product there, with sign-in through your accounts, permissions checked before every action and a record of each one.

For software companies

What it runs on

AI platform and middleware

Gateways, tools, integrations and retrieval.

The layer between models, systems and users: AI gateway, MCP and tool layer, integrations, retrieval, workflows, evaluation and observability.

AI platform and middleware

Models and data

Models made for your task.

Data curation, annotation through partners, fine-tuning, distillation, training and evaluation, with open-weight models you own.

Model services

AI infrastructure

GPU clusters to model serving.

Kubernetes for AI, inference, storage and networking, sovereign deployments and day-to-day operation, on your hardware or partner capacity.

AI infrastructure

When it goes wrong

Anyone can demo AI. We build AI you can trust with a payment.

The difference shows in the moments a quick build breaks. Our engine declares what each assistant may do, checks it before it goes live, guards every action and records the result. Here are three of the failures it handles.

See all ten failures Read the code

Our platform

Every conversation, with your customers and with your teams.

One concierge for every way your customers shop, whether they ask directly or send an AI to act for them. The same platform gives your staff assistants in the tools they already work in.

CustomersWebsite chatWhatsAppInstagramSMSChatGPTClaude

EmployeesSlackMicrosoft TeamsGoogle ChatChatGPT and Claude for work

See the platform

Services

Seven layers, from compute to the conversation. We work at every one.

We use the layers NVIDIA and cloud providers use to describe an AI factory, so your teams and ours describe the work the same way. Pick a layer to see what we have, what we build and what we deliver with partners.

Why Noodle Seed

We run what we build.

Our own platform serves businesses on six channels today, and we have taken 19 businesses through ChatGPT app review. What we build for you has already been tested where customers are.

One plan, every layer.

From the assistant your customer talks to down to the servers it runs on, with cloud and infrastructure partners for the layers below.

You keep what you pay for.

Open-source foundations, your data in your environment where it has to be, and the documentation and runbooks handed over.

Solutions

AI your customers and employees talk to, shaped for your industry.

Start ready-made where it fits. Build custom where it does not. Every solution runs on our platform, in your environment if you need it.

Three ways to start

Ready-made, extended or built for you.

Weeks

Ready-made

Our platform set up on your products, prices, policies and channels. The fastest way to serve customers with an assistant.

1 to 3 months

Extended

Ready-made, plus the journeys and connectors your business needs: your booking system, your store, your core records.

3 months and up

Built for you

A solution designed around your organisation, on our platform or on Noodle Core in your own environment.

Any industry

Built for your industry, whichever it is.

The industries above have pages of their own. The same assistants, platform and services work for any organisation with customers to serve or staff to help. Where the rules require it, every layer runs in your environment.

InsuranceGovernment and public sectorEducationLegalProfessional servicesReal estateLogistics and transportManufacturingEnergy and utilitiesHealth and wellnessHospitality

Regulated? Every layer can run in your environment.

Financial services, healthcare, insurance, legal and the public sector usually need data and models to stay inside. We build and run the whole stack there, with each action confirmed and logged.

Applications: Customer and employee assistants that confirm before they act.
Models: Open-weight models in your environment, evaluated on your own content.
Infrastructure: Your cloud or data centre, or dedicated capacity from our partners.

Solutions / Financial services

Financial services

Private AI for your customers and your staff, in your own environment.

For banks, microfinance banks, insurers, payment companies and fintechs. Customer and employee assistants, the models behind them and the platform they run on, built around how a regulated business has to work.

Talk to us See services

Where banks start

The questions customers ask every day, and the work staff repeat every day.

For customers

In your app, on WhatsApp and in the contact centre

Account, card and fee questions, answered from approved content
Onboarding and document collection, with progress the customer can see
Transfers, bill payments and remittances, confirmed by the customer before they happen
Complaints recorded, routed and followed up

For employees

For branches, operations, risk and head office

Product and policy answers for branch and contact-centre staff, with the source shown
Compliance and credit research prepared for review
Operations requests and reconciliations drafted for approval
HR and IT requests handled end to end

Every layer

What we bring, and what we build with you.

Layer	Noodle Seed solutions	Services to implement or build
Applications	Customer assistant on every channel. Confirmation before action and a log of every step. Payments in the conversation.	Employee assistants for compliance, credit, operations, HR and IT. Connectors to core banking, cards and case systems. Custom journeys.
Models	Choice of model provider, with routing between models.	Open-weight models deployed in your environment. Fine-tuning and distillation on your own data. Evaluation and red-teaming in English and local languages. Model documentation for risk review.
Infrastructure	Noodle Core, our open-source runtime, self-hosted.	AI platform on your cloud or data centre, or on capacity from our cloud partners. GPU servers and systems with our hardware partners. Monitoring and run services.

Built for how a bank has to work

People decide. Assistants prepare, look up and act within limits.

The customer confirms every action that moves money or changes an account
Each action logged: who asked, for whom, what was approved and what happened
Sign-in through your own identity systems, never typed into a chat
Answers drawn only from approved content, with the source cited
Data and models in your environment where your rules require it
Evaluation reports and model documentation for your risk teams
Payments run as durable workflows on Temporal and are reconciled with the provider, so none is lost or made twice

How it usually goes

Weeks 1 to 4

Review

Use cases ranked by value and risk, data and systems mapped, and a plan with a price.

Months 2 to 6

First assistant live

One customer journey or one team, measured against the goals agreed at the start.

Years 1 to 3

Programme

More channels, teams and layers on the same foundation, with our team and partners.

Solutions / Telecoms and technology providers

Telecoms and technology providers

Turn your AI capacity into services your customers buy.

For telcos, data centres and cloud providers building AI factories. GPUs give you capacity. The layers above them, and the customers who use them, make it a business.

Talk to us See infrastructure services

Build, use, sell

From virtualized compute to paying customers.

Build

The layers above your compute

Container platform, GPU scheduling, model serving, per-token metering and billing feeds, on your own virtual machines.

Use

Your own AI, first

Employee and subscriber assistants as the first workloads. Proven on your own business before you sell them.

Sell

Solutions for your customers

Customer assistants, employee assistants and private models for your enterprise and public-sector customers, under your brand or ours.

Sell tokens and solutions, not only GPU-hours.

GPU rental competes on price. The layers above it compete on what customers can do with it.

Model access: A catalogue of open-weight models, metered per token.
Private hosting: Dedicated models for enterprises and government that keep data in the country.
Model services: Fine-tuning, distillation and local-language models for your customers, delivered by our team.
Assistants: Ready solutions for your business customers’ customers and staff.
Services: Implementation delivered behind your account teams.

Partnership models

White label

Your brand, our platform

Solutions sold under your name, bundled with your cloud and connectivity. You bill the customer; we deliver and run it.

Implementation partner

You win the deal, we deliver

We scope, build and run AI projects behind your account teams, on your infrastructure.

Revenue share

Grow together

A share of revenue on solutions your customers activate, with hosting and inference staying on your infrastructure.

The layer model follows NVIDIA’s AI factory reference. NVIDIA does not endorse Noodle Seed.

Solutions / Healthcare

Healthcare

Private AI for patients and staff, under your own controls.

For hospitals, clinic groups, laboratories, pharmacies and health insurers. Assistants for appointments, questions and paperwork, and for the staff who handle them, with models and platform running where patient data has to stay.

Talk to us See services

Where healthcare starts

The calls and paperwork that keep people away from patients.

For patients

On your website, app and WhatsApp

Appointments booked, moved and reminded, each confirmed by the patient
Opening hours, preparation instructions and directions, from approved content
Billing and insurance questions, with hand-over to your team
Prescription refill requests routed to the pharmacy team

For staff

For clinics, pharmacy and administration

Policies and protocols looked up, with the source shown
Referral and prior-authorisation paperwork drafted for review
Scheduling, shift and leave requests
HR and IT requests handled end to end

Every layer

What we bring, and what we build with you.

Layer	Noodle Seed solutions	Services to implement or build
Applications	Patient assistant on every channel. Confirmation before any booking or change, and a log of every step.	Staff assistants for clinics, pharmacy and administration. Connectors to scheduling, patient record and billing systems.
Models	Choice of model provider, with routing between models.	Open-weight models in your environment, so patient data stays inside. Fine-tuning and document models on your own content. Evaluation in your languages.
Infrastructure	Noodle Core, our open-source runtime, self-hosted.	AI platform on your cloud or data centre, or on capacity from our cloud partners. GPU servers with our hardware partners.

Built for how healthcare has to work

Clinicians decide. Assistants handle the admin around them.

Assistants do not diagnose or give medical advice. Clinical questions go to your clinicians
The patient confirms every booking, change and request
Each action logged: who asked, for whom and what happened
Sign-in through your own identity systems, never typed into a chat
Answers only from content your clinical and administrative teams approve
Patient data and models kept in your environment, under your own compliance controls

How it usually goes

Weeks 1 to 4

Review

Use cases ranked by value and risk, systems and data mapped, and a plan with a price.

Months 2 to 6

First assistant live

One patient journey or one team, measured against the goals agreed at the start.

Years 1 to 3

Programme

More sites, teams and layers on the same foundation, with our team and partners.

Platform

Serve your customers, and the AI acting for them.

One concierge for every way your customers shop. Same prices, same stock and the same answers, whether your customer asks directly or sends an AI to decide and buy for them. Your teams get the same platform in Slack, Microsoft Teams and Google Chat.

Talk to us See the architecture

Noodle Seed platform

Build it once. It works everywhere your customers and teams are.

CustomersWebsite chatWhatsAppInstagramSMSChatGPTClaudeVoiceEmail

EmployeesSlackMicrosoft TeamsGoogle ChatChatGPT and Claude for workYour intranet

Your information, connected

Products, prices, stock, bookings and policies from your own systems, so every answer is current.

Actions, confirmed

Orders, bookings and changes happen only after the customer says yes.

Payment in the conversation

Card, bank and wallet payments through Stripe and Safepay.

Your team, when it matters

Hand-over with the full history, so the customer never repeats themselves.

A record of every action

Who asked, what was confirmed and what happened, for every conversation.

Pay when it works

Credits are used when an action succeeds, not for every message.

How it works

Connect your information

Your website, catalogue, booking system or store. We bring in what the assistant needs to know.

Choose what it can do

Answer, book, sell, take payment, hand over. You set the limits and what needs confirmation.

Turn on your channels

Your site, WhatsApp, Instagram, SMS, ChatGPT and Claude. One assistant, the same rules on each.

Open source

Noodle Core

The open-source runtime behind our platform. Run assistants and the tools they use on your own infrastructure, under the Apache-2.0 licence.

Self-hosted, in your cloud or data centre
Connects to your systems through MCP, the open standard assistants use
Hardening, qualification and engineering help from our team

Every action checked, confirmed and recorded.

Identity, permissions, confirmation, records and data handling on every action, and durable workflows on Temporal for every payment.

How it is controlled Payments: consent and reliability

Run it your way

Our cloud

The Noodle Seed platform, managed by us. The fastest start.

Your cloud or a partner’s

Noodle Core in your cloud account, or on capacity we arrange with our cloud partners, run by our team or yours.

Your data centre

Noodle Core and open-weight models on your own hardware, for data that cannot leave.

Services

We build and run AI at every layer, with cloud and infrastructure partners underneath.

Advice, implementation, day-to-day operation and supply, from the assistant your customer talks to down to the compute it runs on. Start with one layer. Add the rest on the same foundation, under one plan.

Model services

Models made for your task, from data to deployment.

When a general model is too slow, too costly or not good enough at what you need, we prepare the data, adapt or train an open-weight model, make it efficient and prove it works. You own the result.

Data

Ready to learn from

Collection and curation of your examples
Annotation and labelling through specialist partners, with your experts on the edge cases
Synthetic data where real examples are scarce or sensitive

Adapt

Better at your task

Fine-tuning on your language, products and documents
Preference tuning and reinforcement fine-tuning against what good looks like
Continued pre-training for a domain or a local language

Make efficient

Smaller, faster, cheaper

Distillation of a large model’s skill into a small one
Quantization and pruning to fit your hardware
Routing so each request uses the cheapest model that does it well

Specialise

Beyond chat

Speech recognition and voice in local languages
Document understanding for forms, statements and records
Embedding and ranking models that make search find the right thing

Prove

Measured before it ships

Evaluation on your own test sets, in every language you serve
Red-teaming for safety, misuse and data leakage
Model documentation for risk and compliance review

Operate

Kept good in use

Training pipelines you can rerun
Serving in your environment or ours
Monitoring for drift, with retraining when results slip

Annotation and labelling are delivered through specialist partners. Training capacity comes from your environment or our cloud partners.

Reviews

Built by someone else? We will test it.

Two fixed-fee reviews that test an assistant, or a plan for one, against the failures in our failure lab. Every finding is reproduced, and each review ends with a plan to fix what it finds.

2 weeks, fixed fee

Production readiness review

For a business with an AI assistant built in-house or by another firm. We test payments, outcomes, hidden instructions, identity, tokens and records.

You get: a scored report, each failure reproduced, and a fix plan with a quote.

3 to 4 weeks, fixed fee

Governance review

For a bank, insurer, hospital or public body planning AI that acts. We map every intended action to identity, permission, confirmation, records and data rules.

You get: a control design your risk team can sign off, and the build plan.

How we deliver

Six ways to work with us. Most engagements combine two or three.

Engagements

From a two-week review to a multi-year programme.

Partners

The layers below, delivered with the people who run them.

We bring in cloud and infrastructure partners for compute, capacity, storage and networking, and manage them inside your plan. One plan, one contract and one team to call.

Cloud providers

Compute, storage and managed services

Managed Kubernetes, storage and model services, in your account or one we set up. Noodle Seed is a Google Cloud Partner.

GPU capacity and data centres

Capacity by the hour or reserved

GPU cloud, colocation and data centre operators, for capacity without building it yourself.

Hardware

Servers, storage and networking

Supplied and installed through manufacturers and distributors, for private deployments and AI factories.

Models and software

Open-source and commercial providers

Models, databases and tools chosen for your needs and licensed correctly, with our team responsible for how they fit together.

Option 1

Your own cloud or data centre

We build and run on what you already have.

Option 2

Capacity we arrange

Cloud or GPU capacity from our partners, set up and billed within your plan.

Option 3

A mix of both

Sensitive work on your hardware, peaks and experiments on partner capacity.

In every engagement

A named team

A delivery lead and engineers, and for programmes, specialist partners working to one plan.

Measures agreed up front

What success means is written down before work starts, and reported against throughout.

Nothing locked in

Open-source components, documentation and runbooks, so your team can take over at any point.

Company

Noodle Seed is an AI company from Silicon Valley.

We build the AI that businesses’ customers and teams use, and the platform it runs on. Our team comes from Google and Microsoft and studied at UC Berkeley, MIT and the University of Oxford. Our angel investors come from Google, Arm and Meta.

What we believe

Start with the need.

The right answer may be a product, a project or a partnership. We decide that with you after we understand the problem.

People decide.

Assistants prepare, look up and act within limits. Anything that matters is confirmed by a person and logged.

You should own what you pay for.

Open foundations, your data where it belongs, and work your own team can carry on.

Trust

SOC 2

Independently audited controls, with confirmation before action and a log of every step built into the product.

Partners

Google Cloud Partner

Payments through Stripe and Safepay. Cloud, data centre and infrastructure partners for the layers below.

Open source

Noodle Core

Our runtime is published under the Apache-2.0 licence for anyone to run and inspect.

Talk to us

Tell us what you are trying to do.

We will suggest where to start, at whichever layer and at whatever size.

Platform / Architecture

Architecture

The control system around every assistant.

Quick builds put the rules in a prompt and hope. We put them in the engine: what an assistant may do is declared, checked before it goes live, guarded on every action and recorded. The model holds the conversation; the engine decides what happens.

The control system

Declared, checked, guarded, recorded.

Failure lab

What happens when it goes wrong.

Pick a failure. Each is a moment where a quick build does the wrong thing, and each names the mechanism that handles it.

Anatomy of one action

One booking, followed through every control.

A table for four with a $20 deposit, from the first message to the record, with what the engine produces at each step.

Interaction language

Write a conversation once. Serve it properly everywhere.

Every capability is described in our own channel-independent interaction language: who takes part, the steps they go through, what each step shows, asks and changes, and every possible outcome. The runtime then serves it on each channel in the form that channel supports.

// Simplified for illustration
interaction('reserve-item', {
  roles: ['customer'],
  steps: [
    inform('availability', { facts: source('stock') }),
    choose('size', { from: source('variants') }),
    confirm('reserve', { effect: 'orders.reserve' }),
    outcome({ success, pending, conflict,
              rejected, unknown }),
  ],
})

How it adapts

Each step takes the form each channel does best.

Native The channel’s own wayAdapted Same meaning, simpler formSecure page Continues on a protected page, then returns

Channel	Show facts	Offer a choice	Confirm an action	Card or identity details
Website chat	Native	Native	Native	Secure page
WhatsApp	Native	Native	Native	Secure page
Instagram	Native	Adapted	Adapted	Secure page
SMS	Adapted	Adapted	Adapted	Secure page
ChatGPT and Claude	Native	Native	Native	Secure page
Slack, Microsoft Teams, Google Chat	Native	Native	Native	Secure page
Voice	Adapted	Adapted	Adapted	Secure page

Honest outcomes

What the assistant says is written before anything runs.

Every action declares its possible results, and each has approved words. The assistant never announces a success the system has not confirmed.

Success

“Your table for four is booked for 8:00 pm on Friday.”

Pending

“Your payment is being processed. I will confirm as soon as the bank does.”

Conflict

“That time was just taken. 8:30 pm is free. Shall I book it instead?”

Rejected

“The card was declined. Nothing was charged and nothing was booked.”

Unknown

“We could not confirm the booking yet. I am checking and will message you shortly.”

Under the hood

Durable waits and follow-ups

A step can wait for a payment, an approval or an event for up to 90 days, then carry on. A change that ended unknown is checked, not repeated. Built on Temporal.

Secure pages for sensitive steps

Card numbers, documents and identity checks move to a protected page and come back to the conversation, so they never sit in a chat.

Confirmed changes

Each step can change at most one thing, through a registered operation with checked inputs, and only after a yes.

One build, every assistant platform

Capabilities are written in TypeScript and compiled into one runtime artifact that serves both current MCP protocol versions, including apps in ChatGPT and Claude.

A broker for credentials

Each person’s access is exchanged for narrow, delegated credentials for each system. Tokens an assistant presents never reach your backend.

An open runtime

Noodle Core, our runtime for assistants and the tools they use, is open source under Apache-2.0.

How it is controlled

Platform / Open source

Open source

Read the code that enforces the controls.

The engine behind our platform is open source under Apache-2.0. Your security team can check how every action is guarded instead of taking our word for it.

Runtime

Noodle Core

The TypeScript SDK, compiler, runtime, protocol layer, credential broker, CLI and self-host service. Run assistants and the tools they use on your own infrastructure.

Noodle Core on GitHub

Language

Strand

The channel-independent interaction language and its checks. A capability that skips a read-back, leaves money untyped or an outcome without words cannot be published.

Tests

Readback

An open test suite for any AI that acts: payments that time out, outcomes it cannot prove, instructions hidden in a page, tokens passed through. Run it against your own assistant.

Reference

Channel catalogue

What each channel can really do, from WhatsApp to Microsoft Teams: native components, limits and fallbacks, each citing the provider’s own page and the date it was read.

The boundary

Controls open. Operations ours.

Open

Everything that enforces a control

The language, the checks, the runtime, identity, permissions, confirmation, records and the credential broker.

Ours

What runs the business

The managed platform and portal, billing, distribution and organisation-wide governance tools, built on the open core.

Platform / How it is controlled

How it is controlled

Every action checked, confirmed and recorded.

Most AI automation is built to talk. Ours takes real actions for real businesses, so every action passes through the same five controls. Plain words first, with the detail for your security team underneath.

Identity

Customers and staff sign in through your own systems, never by typing a password into a chat.

For security teams

OAuth sign-in with your identity provider. Each session is bound to one business and one user. Tokens an assistant presents are never forwarded to your systems; the platform uses its own scoped credentials, held separately for each business.

Permissions

Before anything runs, we check that this person may do it, for this business, right now.

For security teams

Business-scoped authorization and policy are evaluated before every tool call, and again at confirmation. Third-party connectors and interactive views run isolated from the platform and from each other.

Confirmation

An order, booking, payment or change to a record happens only after the customer says yes to the exact details.

For security teams

Actions declare whether they change state. State-changing actions require explicit confirmation of the values shown, and those values come from your records, not from model output.

Records

A record of every action: who asked, for whom, what was checked, what was confirmed and what happened.

For security teams

Structured events for each request, check, confirmation and result, linked end to end and available for export and review.

Data

Passwords, card numbers and sensitive details stay out of logs and out of the assistant’s view. Each business’s data is kept apart.

For security teams

Structural redaction in logs. Credentials never appear in client payloads, logs or artifacts. Data and credentials are isolated per business, and Noodle Core runs in your own cloud or data centre when data must stay inside.

Assurance

Checked by others, not just by us

SOC 2. Noodle Core, our runtime, is open source under Apache-2.0, so your team can read how it works.

Payments

The consent and reliability layer for AI payments.

Every payment an assistant makes is tied to what the customer asked for, what they agreed to and what actually happened. None can be lost or made twice.

Intent

What they asked for

The request kept in the customer’s own words, with what the assistant understood next to it.

Consent

What they agreed to

The exact payee, amount and terms shown and confirmed, recorded with the time and channel.

Execution

What actually happened

Through licensed processors such as Stripe and Safepay, counted only on the provider’s signed confirmation.

Evidence

One record

Intent, consent, execution and the provider’s confirmation joined in one record you can show anyone.

Built on Temporal

Every payment is a durable workflow.

Card details are entered only on the provider’s hosted checkout, never in the conversation.

Survives failure: A crash or network loss mid-payment resumes from the last completed step.
Runs once: Each step is idempotent and provider events are deduplicated, so no one is charged twice.
Defined outcomes: Timeouts and retries end in success, reversal or hand-over to a person.
Reconciled: Every payment checked against the provider’s records, with the full history kept.

Who builds what

Our own code where it matters most. Proven open source and partners underneath.

Layer	Who builds it	What that means for you
Applications and agents	Built by Noodle Seed	Customer and employee assistants, running in production today.
Agent and application platform	Built by Noodle Seed	Our own code and our own controls: permissions, confirmation, records, connectors and MCP.
Data and knowledge	Ours, on open source	Our connectors and retrieval, on established open-source databases.
Models, inference and AI platform services	Open source, run by our team	Open-weight models, fine-tuning, distillation, training, serving, gateways and metering, built and operated by us. Data annotation through partners.
Containers and infrastructure	Partners, run by our team	Kubernetes on your compute or on capacity from cloud and data centre partners, under one plan.

Compared with typical AI automation