Run Any AI Model We Handle Everything Else

    From model selection to production scale, the entire AI stack, hosted in Indonesia, managed by us

    ionext API · Preview
    Live loop
    01 · Talk to IONA
    I need infrastructure for shelf-stock detection. Real-time, peaks at shopping hours.
    Got it. What throughput are you expecting at peak?
    Bursty. Roughly 4x baseline for three hours a day.
    Sizing for peak, not average: 8 vCPU, 32GB, one 8GB GPU. Want me to provision it?
    02 · IONA provisions it
    IONA is preparing your model…
    ·Scanning model catalog…
    ·Benchmarking 12 candidates on id-text…
    ·Selected: Qwen3.8-27B
    ·Provisioning Jakarta GPU node…
    ·Endpoint built · https://api.ionext.id/v1
    Endpoint live · ready to call
    03 · You call the endpoint
    curl https://api.ionext.id/v1/chat/completions \
    -H "Authorization: Bearer sk-ionext-…" \
    -d '{ "model":"qwen3.8-27b", "messages":[…] }'
    200 OK · 184ms · jakarta-1
    > Ringkasan: Dokumen membahas strategi adopsi AI berdaulat di Indonesia, dengan fokus pada kepatuhan data, latensi rendah, dan skalabilitas regional…
    ✓ Shipped. Data never left Indonesia.
    04 · Watch it in ORBIT
    !
    vm-db-01
    vm-db-01Degraded
    192.10.0.9· Jakarta-1 · Debian 12
    System Check
    CPU88%3.5 of 4 vCPU
    Memory91%14.6 of 16GB
    Storage93%37 of 40 GB
    GPU-Not attached
    NetworkLogsMonitoring
    MonitoringLast 15 Minutes
    Storage
    Last updated: 8:09:28 PM
    100%50%0%
    06.15 AM06.25 AM06.35 AM06.45 AM06.55 AM
    CPU
    Last updated: 8:09:28 PM
    100%50%0%
    06.15 AM06.25 AM06.35 AM06.45 AM06.55 AM
    Memory
    Last updated: 8:09:28 PM
    100%50%0%
    06.15 AM06.25 AM06.35 AM06.45 AM06.55 AM
    Ask

    Home of Indonesia's Most Ambitious AI Builders

    PitjarusBTTFeedloopIndonesia 17NodefluxVyPrAIPitforceNubix
    Why IONEXT

    The AI Stack You Were Missing

    “GPUs are either unavailable, oversized for the job, or locked behind a year-long commitment.”

    GPU as a Service.

    Provision GPU compute when the workload needs it and release it when it does not. Sized to the use case, hosted in Indonesia, with a clear view of what you are spending it on.

    See available GPUs

    “Serving a model in production means owning the serving stack, the scaling, and every 3am page.”

    Managed Inference.

    One OpenAI-compatible endpoint and one API key. A drop-in replacement for your existing client code, with routing, batching, and scaling handled on our side of the line.

    “Open-weight models are the right call, until you have to host, update, and monitor them yourself.”

    Open Model Hosting.

    Run open-weight models on sovereign infrastructure without building the hosting layer. Deployed models live alongside the VMs and pipelines that use them, monitored in one place.

    IONA

    Meet IONA, Your Builder Companion Powered By Intelligence

    Tell IONA what you're building. It recommends the infrastructure and model that fit, provisions it in ORBIT, then keeps watching and tells you when something needs your attention.

    ✓
    Recommends infrastructure and models for your actual use case, not the trendiest one
    ✓
    Provisions what it recommends, directly in ORBIT
    ✓
    Watches production health, model drift, and pipeline failures as they happen
    ✓
    Escalates with the options in front of you when something needs your decision
    Explore IONA
    Preview
    IONAworking
    I need infrastructure for shelf-stock detection. Real-time, peaks at shopping hours.
    Sizing for peak: 8 vCPU, 32GB, one 8GB GPU. Want me to provision it?
    Go ahead.
    Running in ORBIT. Six resources, connections wired.
    I'll keep watching health, drift and pipelines.
    vm-gpu-01 memory is at 94% and climbing. This one needs your call.
    Type message here
    Ready
    Waiting for your request…
    ORBIT

    ORBIT Is Where Your AI Operations Actually Run

    Build infrastructure, assemble pipelines, deploy models, and monitor all of it from one canvas, instead of five scattered tools.

    ✓Create VMs and assemble pipelines on one canvas
    ✓Real-time production health in a single view
    ✓Models sit beside the VMs running them
    ✓Public or private status from the topology
    Explore ORBIT
    Workspace/retail-vision/Production
    OrbitVirtual MachineDiskSSH KeySettings
    internet
    Public ingress
    gateway-01
    Public Gateway
    lb-prod-01
    Managed Load Balancer
    vm-api-01
    4 vCPU · 16GB RAM
    vm-gpu-01
    8 vCPU · 32GB RAM · 8GB GPU
    vm-gpu-02
    8 vCPU · 32GB RAM · 8GB GPU
    !
    vm-db-01
    4 vCPU · 16GB RAM
    vol-data-01
    500 GB · Block Storage
    !
    production
    8 resources · vm-db-01 warning
    staging
    3 resources
    vol-shared-01
    500 GB · shared volume
    ssh-keys
    shared credentials
    egress
    project gateway
    project · retail-vision
    shared-vpc
    workspace network
    !
    retail-vision
    11 resources · vm-db-01 warning
    doc-ai
    5 resources
    fraud-detect
    3 resources
    workspace · Jakarta-1
    production
    vm-gpu-01Running
    192.10.0.11· Jakarta-1 · Ubuntu 22.04
    System Check
    CPU71%5.68 of 8 vCPU
    Memory40%12.8 of 32GB
    Storage80%35 of 40 GB
    GPU80%6.4 of 8 GB
    NetworkLogsMonitoring
    MonitoringLast 15 Minutes
    GPU Utilization
    Last updated: 8:09:28 PM
    100%50%0%
    06.15 AM06.25 AM06.35 AM06.45 AM06.55 AM
    CPU
    Last updated: 8:09:28 PM
    100%50%0%
    06.15 AM06.25 AM06.35 AM06.45 AM06.55 AM
    Storage
    Last updated: 8:09:28 PM
    100%50%0%
    06.15 AM06.25 AM06.35 AM06.45 AM06.55 AM
    Model Catalog

    Every Model IONA Recommends, Ready To Call

    Open weights, served from our own hardware in Indonesia. Pick one, or let IONA decide for you.

    DeepSeek

    DeepSeek-V4-Flash-0731

    FEATURED
    DeepSeek · Running on IONEXT hardware in Jakarta

    A 304B mixture-of-experts model that activates only a small slice of those parameters per token, paired with DSpark speculative decoding. Reasoning effort is selectable per request across low, high, and max. The weights are MIT licensed, so the whole thing runs inside our infrastructure, nothing calls out.

    Context
    384K
    High and max reasoning effort
    Parameters
    304B
    Mixture of experts
    Licence
    MIT
    Open weights
    Built for
    Agentic work
    Tool calling, long-horizon code
    Benchmarked on our hardware/Jakarta
    0.25s
    Time to first token
    Hosted APIs took 1.73s and 3.10s.
    1.66×
    Faster over 498 tasks
    Same weights as the hosted DeepSeek API.
    −4.4 pts
    Mean accuracy gap
    Against far larger frontier models.
    Time to first token
    Measured from Jakarta. Shorter is better.
    IONEXT on-premDeepSeek V4 FlashGPT-5.6-LunaSonnet 50.25s1.73s3.10s
    Accuracy on Indonesian data
    Identical prompts and sampling on every endpoint.
    On IONEXTGPT-5.6-LunaSonnet 5
    05010074.276.477.8CORD-v2receipts · field F179.383.183.8IndoCareerprofessional exams79.584.580.5IndoMMLUschool exams

    Internal benchmark, run by IONEXT, not independently audited. 498 identical tasks over three public, human-annotated datasets. Hosted-API latency is measured from Jakarta and includes the network path. Benchmark code available on request.

    Also hosted

    Same infrastructure, same API. Choose a model, or let IONA route each request.

    Qwen

    Qwen3.8-27B

    Qwen (Alibaba Cloud)

    Compact open-weight dense model built on the Qwen3.5 architecture with native vision-language understanding. Thinks by default with preserved thinking across turns, and reasoning depth is tunable per request via reasoning_effort. 262K context natively, extensible to 1M with YaRN.

    Context
    262K
    Parameters
    27B
    Best for
    • ·Agentic coding and repository-level reasoning
    • ·Multi-step reasoning with preserved thinking across turns
    • ·Document and visual understanding (charts, OCR)
    ReasoningCodeMultimodal
    How it works

    From Idea To Running AI Infrastructure In Three Steps

    IONA decides what to run. ORBIT runs it. You hear about problems first.

    IONAreading
    Your use case
    Shelf-stock detection for a supermarket chain. Real-time inference, peak load at shopping hours.
    computer visionreal-timebursty load
    Computer vision with bursty traffic. Sizing for peak, not average, so inference stays under your latency target.
    IONA recommends
    Infrastructure
    VM profileg1.gpu.8x
    Compute8 vCPU · 32GB
    GPU8GB
    Model
    SelectedQwen3.8-27B
    Whymultimodal, 262K ctx
    Indonesia archipelago map
    Data Sovereignty

    The data never leaves Indonesia

    Every inference, every token, processed and stored in-country. Built for Indonesia's PDP Law (UU No. 27/2022). Compliance from day one.

    ISO/IEC 27001 certified by TÜV NORD Indonesia
    ISO/IEC 27001Certified by TÜV NORD IndonesiaSee certificate details →

    Your AI solution. Our infrastructure. One platform.

    Full stack, one platform, zero data leaving Indonesia. IONA and ORBIT handle the infrastructure decisions and the workspace to run them, so your best engineers stay on what wins the next deal.