07 / WHY / VENDOR BRIEFSHEET A-007
VENDOR SELECTION BRIEF

Why YuanASI
for custom AI agent development

We're an enterprise AI agent studio based in Shenzhen, China, building on open-multi-agent, our own open-source orchestration framework. This page puts the vendor-selection questions in one place: who we are, who we fit, how we deliver, what the investment looks like, what you can verify, and how we differ structurally from the alternatives.

Last updated

01The profile in one place

Shenzhen YuanASI Technology Co., Ltd. (brand name YuanASI) was founded in 2023 and is based in the Software Industry Base, Nanshan District, Shenzhen. We are an enterprise AI agent studio, and our foundation is open-multi-agent, our own open-source multi-agent orchestration framework.

ItemDetail
Legal nameShenzhen YuanASI Technology Co., Ltd.
BrandYuanASI (元定义科技)
Founded2023
OfficeRoom A701-20, Block 1, Software Industry Base, No. 81/83/85 Gaoxin South 10th Road, Nanshan District, Shenzhen, Guangdong, China
PositioningEnterprise AI agent studio, delivering systems that run in production and that your own team can maintain
Foundationopen-multi-agent (OMA): our own open-source multi-agent orchestration framework, TypeScript-native, MIT-licensed
ServicesCustom AI agent development · Multi-agent system integration · Enterprise AI advisory
ContactEmail co@yuanasi.com · WhatsApp +86 13760249135

02Who we fit — and who we don't

We fit companies that already have a defined, high-frequency workflow and need AI wired into existing systems. The table below sets our fit conditions beside the boundary conditions — read it before booking a call.

A good fitNot a fit
A defined, high-frequency, rule-based workflowYou just want a general-purpose chatbot
Integration with existing CRM / ERP / internal APIsNo business owner to define the workflow and acceptance criteria
A Node.js / TypeScript backendYou want to buy outsourced hours
Private deployment or on-premise modelsYou need a team on site 24×7
A technical lead on your side

03How we deliver

Delivery runs in five stages: diagnosis → demo / POC → phase-one build → production deployment → handover. The demo / POC stage is optional; you can go straight into the build. Every stage has deliverables you can accept against, timelines are scoped per scenario, and payment is staged with the milestones written into the proposal.

StageDeliverablesTimelinePayment
01 DiagnosisWorkflow, data, and existing systems mapped; feasibility and priority callOne 30-minute call to startFree
02 Demo / POC (optional)Runnable prototype, agent architecture and model selection, acceptance criteriaMeasured in weeks; scoped per scenarioStaged payments, milestones set in the proposal
03 Phase-one buildCustom implementation on open-multi-agent, prompt engineering, evaluation harnessScoped per scenarioStaged payments, milestones set in the proposal
04 Production deploymentProduction deployment and system integration; private deployment and on-premise models supportedScoped per scenarioStaged payments, milestones set in the proposal
05 HandoverCustom project source, tests, SOPs, and team trainingScheduled by stageStaged payments, milestones set in the proposal

04What the investment looks like

Investment is published as bands. A single-scenario MVP sits in the tens of thousands of RMB; multi-agent system integration sits in the hundreds of thousands of RMB; advisory is a fixed price per stage. The exact figure depends on scenario complexity and integration surface, and is set in the proposal after diagnosis.

ItemBasis
Custom AI agent developmentSingle-scenario MVP: tens of thousands of RMB
Multi-agent system integrationHundreds of thousands of RMB
Enterprise AI advisoryFixed price per stage; set in the proposal after diagnosis
Pricing modelFixed project price, quoted against the agreed scope
PaymentStaged payments
SubscriptionProject-based settlement, no subscription lock-in

05Evidence you can check

open-multi-agent is used in production by several third-party projects. Every row below carries its nature label — third-party production adoption, official cookbook recipe, or an interactive demo on fictional data — and links to something you can open and check.

EvidenceNaturePath
Multi-agent tool chain in a WordPress security platformOpen-source production adoption · third-party project/cases/temodar-agent-wordpress-security
Fully offline, local multi-node agent stackOpen-source production adoption · third-party project/cases/offline-sovereign-stack
Contract review: task DAG with step-level retriesOpen-source example · official recipe/cases/contract-review-dag
Support escalation: expert agents picked on demandOpen-source example · official recipe/cases/adaptive-customer-support
Incident postmortem: parallel evidence → root cause → reportOpen-source example · official recipe/cases/incident-postmortem-dag
Commission reconciliation: repair what's repairable, escalate the restOpen-source example · official recipe/cases/commission-reconciliation-recovery
BOM quote review deskInteractive demo · fictional sample data/cases/bom-quote-review-demo
Interactive workflow demo catalogueInteractive demo · fictional sample data/demos
open-multi-agent production adoptersOpen-source production adoption · several third-party projectshttps://github.com/open-multi-agent/open-multi-agent#built-with-oma

06How we compare to the alternatives

In-house teams, low-code platforms, large integrators and YuanASI trade off differently on time to start, depth of customization, private deployment, who maintains the result, and the scale each suits. The table below sets out those five structural differences — match them against your own constraints.

OptionTime to startCustomizationOn-premiseMaintained byScale it suits
In-house teamDepends on hiringNo ceilingYour callYour own teamLong-term, many scenarios
Low-code platformFastest to startBounded by the platformDepends on the platformPlatform vendor + your teamStandardized, light scenarios
Large integratorProcurement / tender cycleScoped by contractSupportedThe integrator, long-termLarge cross-department programs
YuanASIDiagnosis and POC firstCode-level delivery on an open-source foundation, custom project source handed overSupportedYour own team after handoverOne scenario first, then expand

07Four questions we get during vendor selection

These four come up most while a vendor is being chosen. Every answer lines up with what's on the services and about pages; for anything more specific, book a call.

How are you different from a large vendor or systems integrator?

The difference is the shape of the delivery and who maintains it. We deliver at code level on open-multi-agent, our own open-source framework: the custom project source, tests, deployment scripts, and runbooks hand over together, the system runs in your own environment, and your team takes it forward after handover.

What does the smallest sensible starting project look like?

Pick one high-frequency, well-defined scenario and run a POC: we map the workflow and the existing systems, then give you a feasibility call, a runnable prototype, and the acceptance criteria. You then decide whether to enter the phase-one build.

Who maintains the system after handover?

Your team, by default. Handover includes the custom project source, tests, deployment scripts, runbooks, and team training, and the orchestration logic is open source and auditable line by line. We also offer ongoing iteration and operations support, and you choose whether to take it.

Can it be deployed on-premise?

Private deployment is supported directly: open-multi-agent is a Node.js library embedded in your own backend, and with local models via Ollama or vLLM it needs no API key and runs fully offline, so your data stays inside your network. The models, the orchestration, and your business systems all sit in your own environment.

If that lines up, take 30 minutes to walk through your scenario.