On-Premise AI Automation — your data never leaves your servers
AIK runs AI agents entirely on your own infrastructure. Automate invoices, emails, documents and web tasks without sending a single record to a third-party cloud.
Why on-premise AI matters
Most "AI automation" tools ship your documents, emails and customer data to someone else's servers to be processed. For finance, healthcare, legal and public-sector teams — and anyone under GDPR — that's often a non-starter. AIK is built to run self-hosted: install it on a machine you control, and every step happens inside your own network.
- Your documents and database stay on your machine — a local SQLite file (or your own PostgreSQL), never a vendor cloud.
- AI can run on local LLMs (via Ollama) so prompts and data never leave the building — no external API calls required.
- One installer sets up everything (app, database, service) — no DevOps team needed.
- Your code is shipped compiled, and updates arrive as signed packages — no access to any external code repository required.
Self-hosted AI agents — private AI for your business
Private AI means the documents, e-mails and customer data your agents work with stay on hardware you control. AIK's self-hosted AI agents run next to your existing systems, use a local model or a cloud model you choose per agent, and keep a record of everything they do.
What you can automate — privately
- Invoice & document processing — read PDFs and scans, extract fields, validate, export to Excel/CSV or your ERP. See invoice automation →
- Email triage — read the inbox, classify, download attachments, draft replies.
- Database & reporting — query your systems in plain language and build reports.
- Web & RPA tasks — browse sites, fill forms, move data between systems.
GDPR & data residency, handled by design
Because processing happens where you install it, data residency is automatic: EU data stays in the EU, on hardware you own. There's no data-processing agreement to negotiate with a cloud AI vendor because there is no cloud AI vendor in the loop. Combine that with local LLMs and you get automation that's genuinely private end to end.
Hybrid: cloud when you want it, local when you don't
On-premise doesn't mean giving up modern models. AIK is hybrid by design — pick any model in Settings, OpenAI, Anthropic Claude, Google Gemini, Groq, or fully local Ollama, per use case. Run sensitive workloads locally and reach for a cloud model only where it's appropriate. Deploy the app directly on a server or via Docker.
Built to run reliably, unattended
Self-hosting only works if it keeps working. AIK is engineered for it:
- Failure isolation — each tool server runs as its own process, so one failing tool never takes down the platform.
- Health checks & dependency validation — start each built-in tool server and confirm it responds, so you know instantly if an update broke something.
- Per-server timeouts & graceful degradation — a slow or misconfigured component reports "unavailable" instead of hanging or crashing.
- Safe rollback — every update snapshots the app and the database first, so one click restores the previous version if anything goes wrong.
- Signed updates over your own channel — delivered via GitHub or your own server, verified before they're applied; optional auto-update.
How it runs on your own server
- One installer for Windows, Linux or macOS brings what it needs (Python, OCR, PDF tools, a browser for web agents); running it again upgrades and keeps your settings and data.
- Agents can be fenced — switch on the agent fence and each agent's tools run in a sandbox that sees only its own user's files; actions such as sending or deleting wait for a person's OK.
- A record you can prove — every action goes into an append-only activity log whose entries are chained, so a change after the fact shows.
- Backups and updates you control — nightly encrypted backups, signed updates, and a snapshot before each update so it can be rolled back.
EU-hosted AI automation, or entirely your own
Run AIK on a server in the EU — your own, or a provider's in an EU data centre — and pick a local AI model, and the documents your agents handle never leave the EU or your control. Agents can also ask for a person's approval before anything leaves: AI agents with human approval, not unattended black boxes.
Questions people ask
What hardware do we need?
An ordinary office server runs the platform. Local AI models need more memory and ideally a graphics card; with a cloud model the server needs neither.
Does it need the internet?
With local models, no — only for updates if you want them. Cloud models and online services (e-mail, Microsoft 365, Google) naturally need a connection.
Which AI models can we use?
Local models through Ollama, or cloud models from the major providers — chosen per agent, so sensitive work can stay local while the rest uses the cloud.
How are updates done?
Signed release packages or a pull from the repository, opt-in auto-update, and a snapshot of code and database before every update so it can be undone.
How do we keep control of what agents do?
Each agent has an allow-list of recipients and addresses and an 'ask before acting' switch, on by default: sending, uploading or deleting waits on the dashboard for a person's OK.
Does it help with GDPR and NIS2?
Running on your own servers keeps the data under your control, and the activity log and backups support an audit. The obligations themselves stay with your organisation — AIK is a tool that helps you meet them, not a certificate.
Are there per-call fees?
Not with local models — you pay for your own hardware, not per document or per run. Cloud models are billed by their provider, and AIK shows the cost of each run.
Can we try it before installing?
Yes — create a free account on the hosted version, build an agent or two, and install on your server when it fits.