Manual processes in.
Working software out.
OltenAI builds custom AI agents and LLM pipelines for small and mid-sized companies — knowledge bases, document extraction, support and workflow automation — and runs them after launch. GDPR-compliant, EU-hosted or on-premise, working prototype in 2–4 weeks.
We build —an agent that answers order-status mail — with a human gate before it sends
- 00:00.0TRIGGERMail from customer: “Where is order 48213? Needed by Friday.”
- 00:00.4PLANlook up order → check shipment → draft reply → ask approval if the promised date is at risk
- 00:00.9TOOLerp.get_order(48213) → status: shipped, 2 items, carrier DHL
- 00:01.6TOOLcarrier.track("JD0148…") → in transit, delivery expected Thu
- 00:02.3CHECKpromised date met · no discount or refund involved
- 00:02.5DRAFTreply drafted (DE, formal), 86 words, cites tracking link
- 00:02.6GATEapproval requested in Slack #support — pending → approved after 41 s
- 00:44.1ACTIONmail sent · ticket #5531 set to “answered”
This is what an agent run looks like — see custom AI agents →
Eight ways SMBs put AI agents and language models to work
Each service is a page of its own — with the problem it solves, how it works and what it needs. Every project starts with a free audit.
Custom AI agents
Agents that do multi-step work inside your systems — built on LangGraph, hosted on EU infrastructure we run, and extended every month from your team's feedback.
IN: a job with steps, tools and judgmentOUT: an agent that runs it — supervisedSERVICE — RAG
Knowledge base (RAG)
Chat with your own documents. Employees get sourced answers from manuals, wikis and contracts in seconds instead of searching folders for hours.
IN: PDFs, wikis, manuals, ticketsOUT: sourced answers in secondsSERVICE — EXTRACTION
Document & data extraction
Invoices, orders and forms read themselves. AI turns scans and PDFs into validated, structured records inside your ERP or accounting system.
IN: invoices, orders, forms, scansOUT: validated records in your ERPSERVICE — SUPPORT
AI customer support
Tier-1 questions answered instantly, in 30+ languages, grounded in your own documentation. Your team keeps the cases that need judgment.
IN: tickets, chats, mailsOUT: instant answers, clean escalationsSERVICE — TRIAGE
Email & inbox automation
Shared inboxes that sort themselves. AI categorizes, routes, extracts order data and drafts replies — a human approves, nothing disappears.
IN: info@, sales@, support@ mailOUT: routed mail, drafts, extracted dataSERVICE — WORKFLOWS
Workflow automation
Recurring multi-step office processes become monitored n8n workflows with LLM steps: quotes, reports, data sync, meeting notes into the CRM.
IN: recurring manual processesOUT: monitored automated workflowsSERVICE — AUDIT
AI agent & chatbot audit
Already running a chatbot or agent — bought, built or inherited? A fixed-scope audit of answer quality, security, GDPR / EU AI Act compliance and running costs, with a prioritized fix list.
IN: your live chatbot or agentOUT: scored report + prioritized fix listSERVICE — STRATEGY
AI consulting & EU AI Act
From "we should do something with AI" to a prioritized roadmap: use cases with ROI estimates, model selection, GDPR and EU AI Act obligations mapped.
IN: "we should do something with AI"OUT: prioritized roadmap + compliance map
Three ways to work with us
A project when something should be built, an audit when something already runs, a monthly engagement when it should keep improving.
PROJECT
Something should be built.
- Free 30-minute audit to pick the use case with the fastest payback
- Fixed-price prototype on your real data within 2–4 weeks
- Integration into ERP, CRM, mail and ticket systems
- Handover to your team, or straight into operations
Fits — a process you want automated, an agent you want built
AUDIT
Something already runs — and nobody has tested it.
- Chatbot or agent from any vendor, agency or in-house team
- Answer quality, security, EU AI Act / GDPR, running costs
- Scored report, prioritized fix list, reusable test set
- Fixes optional — by you, your vendor or us
Fits — a bot that went live after a demo
OPERATE & EXTEND
It should keep improving after launch.
- Hosting on EU infrastructure we operate — or your servers
- Monitoring, updates, review of failed runs, monthly report
- Your team's feedback becomes the next capability each month
- Code and data are yours at every point
Fits — agents and pipelines that should grow with use
How we build
Data protection is architecture
Model choice, hosting and data flows are designed around GDPR from day one — EU cloud or your own hardware, contracts excluding training use.
No island solutions
AI that lives outside your ERP, CRM and mail is a toy. We build into the systems your team already uses.
Human gates by default
Reads are free; writes, sends and money wait for a one-click approval until the traces prove an action type is safe. Autonomy is widened per action, never all at once.
Payback decides the use case
We only build automations whose saved hours are measurable. If AI is the wrong tool for your problem, we say so.
Working software before big contracts
Within 2–4 weeks you test a working prototype on your real data — then you decide.
One engineer, the whole stack
The person who builds your agent also hosts, monitors and extends it. No handover between an agency and an ops team, no ticket queue between you and the code.
From first call to running system
Discovery call
30 minutes on your processes and bottlenecks. We identify the use case with the fastest payback — free and without obligation.
Proof of concept
A working prototype on your real data within 2–4 weeks. You validate the value before committing further.
Integration
The solution is wired into your existing systems via API, your team gets trained, guardrails go live.
Operate & extend
Hosting we run, monitoring, model updates and a monthly feedback loop that turns what your team learns into the next capability — or a clean handover to your team.
Tools we build with
Langfuse
PostgreSQL
Docker
Hetzner
OllamaFrequently asked questions
Is our company data safe with LLM automation?
Yes, if the architecture is built for it: EU-hosted or on-premise models, contracts that exclude training use, and data flows documented for your GDPR records. That is our default setup, not an add-on.
How fast do we see results?
The first working prototype runs on your real data within 2–4 weeks. You evaluate actual output — not slides — before any larger commitment.
What does a project cost?
The 30-minute initial audit is free. Prototypes are fixed-price after the discovery call, sized to the use case. Ongoing operations are optional — your team can take over.
Can everything run on our own servers?
Yes. Open-weight models (e.g. Mistral) and self-hosted components like n8n, pgvector and Qdrant allow fully on-premise setups when data must not leave the house.
We already have a chatbot. Can you check it?
Yes — that is the AI agent & chatbot audit: a fixed-scope review of answer quality, security (prompt injection, data leakage), GDPR / EU AI Act compliance and running costs, whoever built it. You get a scored report and a fix list; implementing the fixes is optional.
Who runs the system after launch?
Either your team, with documentation and handover — or we do: hosting on EU infrastructure we operate, monitoring, updates and a monthly feedback loop that turns what your team learns into the next capability. Cancel any month; code and data stay yours.
Get a free 30-minute AI audit
We look at your processes, name the use case with the fastest payback and tell you honestly if AI is not the right tool. No slides, no obligation.
Or write directly: hello@olten.ai