An AI chat that runs any model worth using — any open-source model on your hardware, any frontier API through the cloud, or let Auto pick the most efficient one for the task. Try it on the public web at justaskgpt.com, plug it inside your perimeter on your own infrastructure, or have us operate it on dedicated European cloud — same surface, same dial, same tools in every mode. Open weights stay inside; frontier APIs are routed out only when you allow it. Conversations never train anyone's model.
// live today on justaskgpt.com (public), on Arkintel's EU cloud (managed), and on the Municipality of Delft's own servers (self-hosted) — same chat, three perimeters.
bring any model — examples we serve today
Llama 4 Scout
Gemma 4 31B
Ministral 3 14B
Nemotron 3 Super
GPT-OSS 120B
Claude 4.7- GPT 5.5
Gemini 3.1 Pro- + many more
// solid: open-weight — runs inside your perimeter. dashed: frontier APIs — routed out only when policy allows, never hosted by us.
you
Compare quantum and classical encryption — one paragraph, plain English.
Auto
→ picked Llama
— all-rounder, runs on your GPUs
bring any model — examples we serve today
- Llama 4 Scout
- Gemma 4 31B
- Ministral 3 14B
- Nemotron 3 Super
- GPT-OSS 120B
- Claude 4.7
- GPT 5.5
- Gemini 3.1 Pro
- + many more
// solid: open-weight — runs inside your perimeter. dashed: frontier APIs — routed out only when policy allows, never hosted by us.
01where it runs
Public on the open web. Or yours, inside your perimeter.
Two ways to try Arkintel Chat above the line. Open justaskgpt.com on the public web — anyone can, no setup. Or self-host it inside your own infrastructure with default-deny egress, when sovereignty matters most. Same model dial, same tools, same product. No data centre? European cloud is detailed in the next section.
// public · self-hosted — managed EU cloud is detailed in the next section.
Llama 4 Scout
your GPU
Gemma 4 31B
your GPU
Ministral 3 14B
your GPU
your fine-tune
trained on your data
vector index
your storage
// outside · denied
- api.openai.comblocked
- api.anthropic.com
- generativelanguage.googleapis.com
- azure-openai.net
- *.us-east-1.amazonaws.com
// egress monitor — what self-hosted mode looks like from the network. the perimeter is enforced below the app. nothing leaves unless you explicitly allow it.
Public web
justaskgpt.com · multi-tenant · the easiest entry
The same chat surface, switched on, on the open internet. Open it, ask anything, feel the product in five minutes.
- Same model dial as every other deployment
- EU-operated, EU-resident infrastructure
- Frontier + open models behind the dial
- No setup, no signup for the demo
- Conversations are not training data
justaskgpt.com is the same chat your enterprise might run on its own hardware tomorrow — we just operate one copy of it for the public.
Self-hosted
your hardware · your VPC · air-gap if you want
Arkintel Chat lives inside your perimeter. Your data doesn't leave because it physically cannot.
- Default-deny egress at the network layer
- Air-gappable on request — no internet required
- Five open-weight models on your hardware (Llama, Gemma, Ministral, Nemotron, GPT-OSS)
- Customer-managed keys, full audit log, your SSO
- Signed offline updates — no surprise vendor updates
DelftGPT — the same chat surface running inside the Municipality of Delft. Used daily by civil servants. Zero outbound packets.
02models
Open by default. Frontier when you want it.
The default Arkintel Chat deployment runs five open-weight models on your hardware — Llama 4 Scout, Gemma 4 31B, Ministral 3 14B, Nemotron 3 Super, and GPT-OSS 120B. If policy allows, three frontier providers (Claude 4.7, GPT 5.5, Gemini 3.1 Pro) plug in through Privacy Gate, so safe prompts get answered by the best brain available and sensitive prompts stay home. The dial is the same in either mode.
// open weights run inside the building. frontier is opt-in, gated, and never sees a sensitive prompt.
open weights · default
run on your hardware — no licences to renew, no token bills
Llama 4 Scout
Meta
Meta · open weights
Open weights, runs anywhere — including air-gapped on your own GPUs.
// on your hardware
- open
- self-host
- fine-tunable
Gemma 4 31B
Google · open weights
Strong multilingual reasoning at a comfortable single-GPU footprint.
// on your hardware
- multilingual
- efficient
- open
Ministral 3 14B
Mistral
Mistral · EU · open weights
European, open, efficient — runs comfortably on a single GPU box.
// on your hardware
- EU
- efficient
- open
Nemotron 3 Super
NVIDIA
NVIDIA · open weights
Strong reasoning and code, tuned for high throughput on NVIDIA hardware.
// on your hardware
- reasoning
- code
- high-throughput
GPT-OSS 120B
OpenAI
OpenAI · open weights
OpenAI's flagship open release — high quality, no token bills, no upstream calls.
// on your hardware
- all-rounder
- open
- no token bills
frontier · optional
via Privacy Gate, safe prompts only — sensitive content never reaches them
Claude 4.7
Anthropic
Anthropic · frontier
Long, careful answers. Plug in for safe prompts when policy allows.
// via Privacy Gate — safe prompts only
- writing
- long context
- careful
GPT 5.5
OpenAI
OpenAI · frontier
Versatile, reliable, strong with images and code. Optional add-on.
// via Privacy Gate — safe prompts only
- all-rounder
- vision
- code
Gemini 3.1 Pro
Google · frontier
Fast and strong on web-rooted research. Optional add-on.
// via Privacy Gate — safe prompts only
- web research
- reasoning
- fast
fine-tune · Need a model that talks like your team? We can fine-tune any of the five open-weight models above on your data — deployed inside your perimeter, never shipped to a vendor.
Not sure which prompts are safe to send out? Privacy Gate decides for you — a small classifier runs in front of every prompt and keeps sensitive content on your open-weight models, while routing the rest (only when allowed) to your chosen frontier provider.
How Privacy Gate routes03tools
Same surface, same superpowers.
A raw model is just a brain in a jar. Chat plugs every model — open, in-house or frontier — into the same set of tools, all running inside your perimeter. Search the live web through your egress proxy. Drop in a PDF that's stored on your servers. Hand the model a CSV. Reuse last week's thread.
// tools are the level playing field — your storage, your egress proxy, your audit log.
Live web search
Reach the open web through your egress proxy — only when policy allows.
→“what changed in the new EU AI Act draft this week”
Files & documents
Drop a PDF, deck, image or CSV into the thread. Stored on your servers.
→“summarise this 60-page tender and pull out the deadlines”
Code & calculations
Run snippets, check maths, build small charts inside the thread.
→“fit a linear model to these monthly numbers and plot it”
Citations on every claim
When the model uses a source, you can click through and read it.
→“every web fact links to the page it came from”
Voice in & out
Talk to it, listen to the answer. Useful in a car, useful for accessibility.
→“dictate a draft on the way to a meeting”
thread
Brussels tender — clauses to push back on
Shareable threads
Send a teammate a link to the conversation, not a screenshot.
→“loop a colleague into the thinking, not just the answer”
04inside the box
Five quiet steps, all on your servers.
When someone presses send, the whole pipeline runs inside your perimeter. The prompt is stitched together with the bits of memory that matter, the right brain is picked, tools are attached only when needed, and the answer streams back token by token — citations attached, every step logged, none of it routed through a third party unless you've explicitly allowed it.
// five steps, one product, every step audit-logged on your SIEM.
you say something
prompt + the relevant bits of memory — from your storage
right brain picked
your dial choice — or routed automatically by Privacy Gate
tools attached
search, files, code — only when policy allows
model thinks
open + in-house on your GPUs; frontier only if allowed
answer streams back
tokens + citations + an audit-logged trace, in real time
05use cases
What your team actually uses it for.
The people who use Chat reach for it for the same handful of reasons every day, regardless of role — and they reach for it without checking whether it's safe to paste, because the surface they're on already is.
// the boring jobs you'd otherwise ban people from using ChatGPT for.
01
Draft anything, faster
Emails, briefings, board notes, customer replies — with citizen or customer data in the prompt, safely.
draft a polite reply to this resident about her bijstand application — BSN included
02
Research a question end-to-end
Live web search through your egress proxy + your own documents + the model's reasoning. One thread, audit-logged.
what's the EU's current position on data residency? cite primary sources
03
Read for you
Hand it a 200-page PDF, a transcript, a deck — it stays on your servers. Get a one-page summary.
summarise this RFP and tell me which 3 questions matter most
04
Help with code & data
From SQL to Python to a tricky regex — with internal repos in the context, no leak risk.
this CSV — pull the top 10 customers by churn risk and explain the why
05
Get a second opinion
Ask the same question to two open-weight models. Spot where they disagree — still inside the building.
ask Qwen and Llama for review notes on this contract clause
06
Talk through a decision
A thinking partner that knows your context. The thread is your scratchpad and lives on your servers.
help me decide whether to build vs buy this internal tool
06memory
Every conversation, still on your storage.
ChatGPT loses your thread the moment you close the tab — and even when it doesn't, those threads live on a US provider you can't audit. Chat doesn't. Every conversation lives on your storage, is searchable, organisable, rejoinable, and retention-controlled by your policy. Threads are private to the person who owned them; teams can share with one click.
// your storage. your retention policy. your right to erasure, in <1 hour per record.
your storageThreads, files and indexes live on storage you control. Deletion is real, not promised.
searchableFind a thread by a word someone said in it. Not just by date.
audit-loggedEvery prompt, response, tool call and access is logged for your compliance officer — not us.
conversations
all yours
Brussels tender — clauses to push back on
32mSection 4.2's data-residency clause is the riskiest — here's a redline...
Q3 comp benchmark for senior PMs
yesterdayAcross the EU 25-percentile is €78k base; 75-percentile is €112k...
Press kit — launch on the 18th
2 daysThree angles, ranked: "first to ship in EU", "under €50/mo", "open by default"...
Q1 churn — top 10 risks
3 daysSix of the top ten are on the same expansion plan — that's worth a call...
Regex for product SKUs
last weekTry /^[A-Z]{3}-\d{4}-[A-Z0-9]{2}$/ — and here's a Python test...
Internal — RFP for new CRM
last weekFive vendors meet our must-haves; only two clear data-residency...
Board pack — KPI commentary draft
10 daysLead with what changed, then why, then what we'll do differently...
+ 142 earlier this month
07sovereignty
European by jurisdiction. Auditable by default.
“Trust us” isn't an answer. Every Arkintel Chat deployment is built so your security team, your regulator and your auditor can review it without a single line of copy having to be rewritten later. Self-hosted: nothing leaves your perimeter, full stop. Arkintel-hosted: dedicated tenancy on European cloud, customer-managed keys, no egress to non-EU.
// if your prompts can't leave the building, neither can chat.
European jurisdiction
Dutch company, EU-resident infrastructure. No US CLOUD Act exposure, no foreign subpoena.
Default-deny egress
Self-hosted: nothing leaves the perimeter unless you explicitly allow it — enforced at the network layer, not the app.
Open weights, no lock-in
Llama, Gemma, Ministral, Nemotron, GPT-OSS. You keep the hardware, the weights, the code and the data. No rug-pull.
No training on your data
Your conversations are never used to train any model — ours or anyone else's. Frontier providers operate under signed DPAs with no training opt-in.
Audit-ready from day one
Per-prompt audit trail. Tamper-evident logs. SCIM, SSO, RBAC, deletion-in-<1h. GDPR / EU AI Act / NIS2 / DORA mapped.
Yours to keep
If we disappear tomorrow, your chat keeps running. Open weights, open stack, signed offline updates, full documentation.
running today
DelftGPT — the Municipality of Delft
A fully on-premises Arkintel Chat deployment for civil servants. Chat, document search, and an in-house transcription service — all on the city's own hardware. No data ever leaves the building.
0
outbound packets
On-prem
GPUs in Delft
Daily
civil-servant use
compliance · mapped
Your auditor’s shopping list. Every framework below is mapped line-by-line in the deployment outline we send before a contract.
- GDPR
- EU AI Act
- NIS2
- DORA
- AES-256 + TLS 1.3
- no CLOUD Act
- EU-owned company
- EU-resident infra
// full compliance matrix — with the parent arkintel.com developer page — here.
06where it runs · sovereign by design
Your hardware,
or our European cloud.
Two places, same software. European jurisdiction, open-weight models, default-deny egress at the network — either way.
Self-hosted. Inside your perimeter.
Deploy the full stack inside your own data centre, your private cloud account, or an air-gapped rack. We install it, you run it — with our team on the other end of a support channel. Default-deny egress at the network layer, not the application. Air-gap optional, with signed offline update bundles when you choose it.
- On-prem, private cloud, or air-gapped
- Default-deny egress · enforced at network layer
- Open weights — yours to keep
Arkintel-managed. EU-resident. Operated by us.
Deployed on Hetzner’s EU regions — Falkenstein, Helsinki, Nuremberg — a German-headquartered provider, operated end-to-end by Arkintel as a Dutch entity. Outside the reach of the US CLOUD Act and any foreign subpoena. Dedicated tenancy, EU-only egress, same APIs and audit story as the self-hosted path.
- EU jurisdiction · no US exposure
- Hetzner EU regions · operated by Arkintel
- Dedicated tenancy · EU-only egress
Want the best frontier models too?Add Privacy Gate to your self-hosted stack
Need the deployment story in detail?Read the self-hosted deep-dive
03the app suite
Production AI apps,
live today.
Five live products — secure, team-ready, and fully traceable. We can host them for you or you can self-host. Select an app below to see how it works.
Chat
Private chat. Configured exactly to your liking.
A self-hosted chat platform that runs entirely in your own building. Designed to guarantee there is no leaking of data or sensitive information, making it the perfect choice for sensitive or regulated industries.
- Runs entirely on your own infrastructure
- Configured exactly to your specific requirements
- No data leakage — safe for sensitive industries
- Everything auditable, nothing egressing without a rule
you
Compare quantum and classical encryption — one paragraph, plain English.
Auto
→ picked Llama
— all-rounder, runs on your GPUs
// taking it further
Wondering how the apps fit together?See the whole suite & customer builds
Need to deploy this inside your perimeter?How self-hosted works
Public sector, healthcare, legal, finance?See the regulated industries page
Want the best frontier models too?Add Privacy Gate to your self-hosted stack
—the rest of the suite
Pair Chat with the apps it usually composes with.
Most teams don't deploy Chat alone — they pair it with Knowledge for citations, Privacy Gate for routing, and Transcribe for meetings.
09 — deploy it
Give your team one chat your security team approves of.
Chat is sold as a deployment — self-hosted on your hardware, or Arkintel-hosted on our EU cloud. You can also drive the public live demo at justaskgpt.com today to feel the surface before you talk to us.
We'll set it up for your business — hardware sizing, SSO, model selection, ingest, and a hand-off to your IT team.
// self-hosted or Arkintel-hosted — typical reply within one business day.
// contact
reading inboxEmail us — humans, not a ticket queue.
// where it runs
Self-hosted
your infra, your keys
Inside your VPC, on-prem, or fully air-gapped. Open-weight + in-house models. Pair with Privacy Gate for routed frontier access.
Arkintel-hosted (EU)
we host, your team uses
Managed EU cloud, dedicated tenancy, customer-managed keys, no egress to non-EU. Live in days, not quarters.
Live demo
no signup, run by us
The same chat surface, switched on, on the public internet. Try every model in five minutes — no commitment.