Services
Six clear entry points
AI audit, private knowledge base, process automation, private workstation, media lab and CastelOS pilot deployment.
View services ->The operating system for the one-person department.
CastelOS is one serious workstation doing the work of a small department — writing, analytics, knowledge, media, automation: seventeen studios in all. Nothing leaves the machine unless you say so, and every run leaves a trail.
10 roles1 workstation1 operator
Solo studio — limited slots. Response within 48h. Deposit confirms your slot in the queue.
Need something built or produced now? We do that.
Fixed scopes, written quotes, and a person who answers. Pick the door — the scan is the cheapest way in.
Choose by situation
Start with the problem, not the tool name.
If the catalogue feels broad, use this map first. Pick the closest situation; the quote can refine the scope after a human reads your request.
DOCS
We have documents and cannot find reliable answers.
Private knowledge base with cited answers and a clear 'not found' boundary.
Build knowledge base →FLOW
We repeat the same admin process every week.
One controlled automation with logs, checkpoints and a human approval step.
Automate a process →SCAN
We do not know where AI is actually worth using.
Written audit: where to automate, where not to, and what to do first.
Order a scan →MEDIA
We need content, voice, video or a repeatable media pipeline.
Human-reviewed media work with rights, consent and synthetic-use boundaries.
Plan media work →LOCAL
Our data should stay on our own machine.
Private AI workstation: hardware, models, data classes and operator training.
Set up local AI →OS
We want the whole operating system, not one isolated task.
CastelOS pilot with governed studios, deployment notes and support boundaries.
Scope a pilot →TASK
We need a small fixed task now: site, audio, data, Telegram, hardware.
Pick one of the 23 catalogue offers and send the task in plain words.
Open catalogue →Core services
Founding-client rates — reduced while we build the public portfolio. Limited slots.
01 · SCAN
AI Quick Scan
Where AI pays off in your workflows, mapped + roadmap.
your process notes → audit report + roadmap → 1-2 weeks
A decision aid — you stay the decision-maker.
490€ Light / 890€ Standard
02 · RAG
Private Knowledge Base
Your documents, searchable and answerable — citations verified by code.
your document corpus → local RAG system → 2-4 weeks
Works on your machine; answers cite sources or honestly say "not found".
from 1 900€ (typical project 2 900-4 500€)
03 · AUTO
Process Automation
One repetitive process automated, with human checkpoints. Includes scheduled reporting.
one process + sample data → running automation → 2-4 weeks
A person signs off at every defined checkpoint.
from 990€ + care plan 290-490€/mo
04 · WKST
Private AI Workstation
Your own AI, running on your own hardware. Air-gapped modes, data classes, training included.
your requirements → installed workstation + training → 3-6 weeks
An engineering setup, not a certification; your compliance stays yours.
from 2 900€ + hardware (transparent quote)
05 · MEDIA
AI Media Lab
Content, audio and video production through governed pipelines.
your brief → reviewed deliverables → per pack
Human-reviewed; synthetic voices by default; likeness only with written consent.
content from 490€/mo · video from 490€ · talking-head from 1 490€
06 · PILOT
CastelOS Pilot Deployment
The system itself, installed and operated for you.
scoping call → pilot deployment → by agreement
Operator-installed today; self-serve edition — join the waitlist.
from 3 900€ · broader deployments quoted after scoping
AI assists — it does not replace professional judgment. You validate and approve all deliverables.
Platform capabilities
Available on request — built on the same platform.
Websites & landing pages
Multilingual, CMS publishing.
Document intelligence
PDF analysis, version diff, extraction.
Marketplace seller assistant
Listings, fee-aware margins.
Marketing strategy
ICP, funnels, launch plans.
Browser automation
Legitimate monitoring & workflows.
Book-to-audiobook production
Voice cloning
With written consent only.
AI tutoring systems
In development
Built and tested, not open yet. No date — they open when their gates are done.
07 · JOBS
Private job-search monitoring
Waitlist open
Scans public listings against criteria you define. Not a recruitment agency.
Learn more →08 · LOTS
Auction lot data
Private beta
Public lot data as structured fields with source notes. Not investment advice.
Learn more →Built on what we run ourselves.
TLX
An autonomous AI media channel, publishing around the clock with self-healing pipelines.
Anna
A voice companion that calls, listens and remembers — built for elder care.
French Tutor
A complete AI tutoring product: lessons, voice, pronunciation scoring.
Self-heal
A runtime that detects, diagnoses and repairs its own failures — live, in production.
No stock logos. No invented testimonials. These systems are real and running.
Some of it you can check yourself
gpu-memory-guard ↗
A small open-source CLI that checks GPU memory before a model loads — so the machine doesn't freeze. MIT-licensed, in use on our own workstation.
awesome-local-ai ↗
A curated list of tools for running AI on your own hardware — the map we navigate by, maintained in the open.
castelos-public ↗
CastelOS build-in-public artifacts: principles, architecture notes and updates, as the system grows.
- Teardown: a cold e-mail rewritten line by line →
- Teardown: one enquiry, read properly before anyone answers →
- Teardown: an answer with a page number, not a memory →
01Source material
The task, documents and boundaries are set before work begins.
02Approval point
A consequential step stays visible and is approved by a person.
03Verifiable result
The outcome is a file, note or working artifact you can keep.
This is how pilots work: a clear input, control during the work, and an independent result.
PORTFOLIO
Products and systems, not vague capabilities.
The site should show what the studio can actually build. This is the portfolio map: public where it can be public, bounded where it must stay private.
CastelOS
Pilot deployments
Local-first AI execution system for knowledge, automation, media, analytics and governed operations.
Evidence: Architecture notes, system pulse and live demo by request.
Evidence ->TLX — @trustlinex
Public channel, running
A personal news radar: it reads hundreds of sources in many languages, keeps only what matches a narrow interest profile, and publishes the analysis on a schedule.
Evidence: Open the channel and read the output — that is the proof.
Telegram ->Provence & Cote d'Azur news — @provence_cotedazur
Public channel, running
A regional news channel run by the system: sources scanned, items drafted and published on a schedule, in French and English.
Evidence: Open the channel and read the output — that is the proof.
Telegram ->Anna
Private product line
Voice companion workstream focused on calls, memory and care-oriented interaction design.
Evidence: Referenced as a real build with private details kept out of the public site.
Evidence ->French Tutor
Product prototype
AI tutoring system covering lessons, voice practice and pronunciation-oriented feedback.
Evidence: Included in the proof stack as a product artifact.
Evidence ->GPU Memory Guard
Open-source utility
Small CLI that checks GPU memory before model loading to reduce workstation freezes.
Evidence: Public GitHub repository.
GitHub ->Awesome Local AI
Open public map
Curated reference map for running AI on owned hardware and choosing local-first tools.
Evidence: Public GitHub repository.
GitHub ->THE SYSTEM
CastelOS
Most businesses run on a patchwork: ten tools, ten subscriptions, ten context switches, ten points of failure. CastelOS is the alternative — one governed system on one powerful workstation, run by one operator.
Content
Drafts, documentation, and copy move from idea to reviewable text without leaving the system.
Analytics
Your data becomes a readable report instead of a spreadsheet nobody opens.
Knowledge
Everything your team has written becomes something your team can actually find.
Media
Visuals and audio are produced under the same governance as everything else.
Automation
Routine steps run on their own, and pause for a person when it matters.
Reporting
The state of the operation is always one glance away, not one meeting away.
17 studios across the platform — one operating system.
How it's built
Four layers, one machine. Click a layer to see what it does.
One workstation you can point at. GPU and memory routing decide where each task runs — a small model for a small job, the big one only when it earns its power draw.
your building · your power socket · your data
Every run passes through the same gate: policy checks before, an audit trail after. Risky actions stop and wait for a person — that's a feature, not a bottleneck.
policy layer · audit trail · human checkpoints
Seventeen specialised studios — text, analytics, knowledge, media, automation and the rest — behind one front door. The router reads the task and picks the studio; you never file paperwork to the wrong department.
one request in · the right studio takes it
Not chat transcripts — files. Reports, documents, media, each one traceable to the run that made it. If a number is in the report, the trail shows where it came from.
real files · every run tracked
What it is not
- just a chatbot
- a wrapper over one model
- a collection of prompts
- a studio zoo
- a SaaS dashboard
IS
- a system
- installable
- local-first
- governed
- extensible through packs
- built for execution
Editions
Lite
single-domain starter profile
Pro
full multi-studio workstation
Enterprise
custom deployment and governance
Pricing on request.
Deployed as operator-installed pilots today. Self-serve edition — join the waitlist.
LIVE DEMO — SIMULATED
Ask the OS
Try a sample request below. This demo shows how CastelOS classifies and routes work — not a live connection to a production system.
Response will appear here once you send a request.
SYSTEM PULSE
The system behind this site, in numbers.
This studio doesn't run on slideware — the platform that does the work audits itself, and these are its own figures. Dated snapshots, not live dashboards: we'd rather show you a real number from last week than a fake one from right now.
- 17
- studios in the registry
- studio registry
- 128
- text task types
- text engine task types
- 1751/1755
- suite tests passing
- TLX suite run, 2026-07-03
- 52
- routing chains
- routing chains
- 691
- MITRE ATT&CK techniques mapped
- MITRE ATT&CK techniques mapped
- 1600+
- security registry entries
- live security registry entries
PUBLIC PROOF, PRIVATE MOAT
How it thinks
We publish principles, not internals: local-first by design, the smallest capable model for each task, human approval where it matters. If it explains a pattern — we share it. If it enables cloning — we don't.
FAQ
Common questions, honest answers
Do you work with my data locally?
By default, yes — the standard deployment runs on your own workstation, and your documents and data stay on your machine. Some capabilities (e.g. certain media or web-facing tools) may call external services when there's no practical local equivalent; when that happens, we tell you which service and why, before it's part of your setup.
What does a project cost?
See the Core services pricing above — real published starting prices, from 490€ for an AI Quick Scan to custom quotes for a full workstation or pilot deployment. These are founding-client rates while we build the public portfolio. We give a transparent quote before any work starts; nothing is billed on estimate alone.
Do I need my own hardware?
For a Private AI Workstation, yes — we help you choose or supply the hardware and install on it. For smaller engagements (a scan, a knowledge base, one automation), we can scope around hardware you already have.
Which languages do you work in?
This site and our delivery work in English and French; the platform itself has been used with Russian, Ukrainian, and German content as well. Ask about your specific language needs when you get in touch.
Is the demo a real AI?
No — the "Ask the OS" widget above is a scripted routing preview: fixed sample prompts and pre-written responses that illustrate how requests get classified and sent to the right studio. It doesn't call a live model and it doesn't collect data. A real, live demo on an actual CastelOS instance is available on request.
Who reviews what the AI produces?
You do. Every deliverable — a draft, a report, an automation run — passes through a defined approval checkpoint before it goes anywhere final. We don't ship a system that acts on your business unsupervised.
Contact
One person reads this inbox — the same person who builds what you order.
Order path
What happens after you send a request.
A site form is not an order and not a payment demand. You do not need a technical brief: write in normal words what is happening, and we turn it into a clear written scope before work begins.
Send in simple words
- What is happening now and what should be better.
- Files, links, examples or rough notes you already have.
- Deadline, language, privacy limits and budget boundary if they matter.
No hidden subscriptions: extra work, changed briefs and longer support are quoted separately before they begin.
Send request- 01
Human reply within 48 hours
You receive a fixed quote, a clarifying question, or an honest refusal if the job is not a good fit.
- 02
Scope and checkpoints in writing
The quote names the deliverables, price, deposit if needed, review points and what support is included after delivery.
- 03
Work starts after approval
Risky outputs are reviewed by a person before delivery. You receive files, notes and the next practical step.
Book a Project
Talk to us
Phone · WhatsApp · Telegram
Book a 20-min intro call →
Opens on cal.com — pick a slot, video link comes by email. No account needed.
A one-person AI studio in Nice. Whoever answers your message is the one who does the work — no account managers in between. Files are processed on our own machine, not in someone else's cloud.