Early access — The platform is under active development. The Free tier is available now.
Cloud or Self-hosted AI Automation Platform
Your AI team that never sleeps
AI agents that check your email, process invoices and documents, answer phone calls, browse websites, query databases, and post to social media — automatically, on schedule, with zero coding.
No business analyst, no project manager, no developer. Describe the job in plain words, attach a document or a diagram — “read the invoices in this folder and put them in Excel”, “answer the phone and book appointments” — and our Process Audit module tells you what can be automated and builds the workers for you. Or build them yourself, with AIK’s help, one click at a time.
Ask, and it builds.
Tell AIK in the chat what you need — “an agent that reads my support mailbox every morning and puts the open questions in a spreadsheet” — and it creates the agent or the whole workflow for you, tools and schedule included, ready for you to look over before you switch it on.
Runs where you want it.
On our cloud, or on your own server so your documents never leave the building. Works with any AI model — the big cloud ones or free local ones — and you switch with one click.
Works from Claude.
Connect AIK to Claude Desktop or Claude Code and run your agents and workflows just by asking. Anything Claude builds there arrives switched off, for you to check first.
Nothing happens behind your back.
New workflows start switched off until you turn them on, any step can wait for a person to approve, and every account has a spending limit. Every run leaves a signed receipt: what was done, by which agent, what it cost. Everything AIK produces is marked as AI-made, as EU law now requires.
A hidden order cannot make it act.
A document, e-mail or web page can hide an instruction for the AI — it is called prompt injection. With Ask before acting, anything that sends, posts, pays or deletes waits on your Dashboard, shown in full, and AIK does exactly what you approved. An allow-list limits each agent to the addresses and sites you name — everything else is refused before the tool runs.
Every agent in its own fence.
An agent's tools run locked inside their owner's folder by the operating system, not by a promise: other people's files, the server's keys and the rest of the machine simply are not there. Even an admin sees other users' agents read-only, and every look goes into a log that cannot be quietly changed.
Decisions, not guesses.
A workflow can ask a plain question at any step — “did this step really finish the job?” — and Jev, a model that writes no text, answers yes or no with how sure it is. The branch and the confidence go into the run log, and if it cannot answer, the step stops with the reason instead of guessing. Optional, and it runs in the cloud.
Every agent has a passport, every run a certificate.
An Agent Passport is a signed ID card for one agent — who it is, what it may do, its track record and the hours it has saved — and an Action Certificate is a signed receipt for one run: what was done, when, at what cost. Both open as a PDF and anyone you send them to can check them on a public page, no login needed. A “Verified AI” badge puts the passport on your own website.
One assistant, everywhere.
Website chat, phone, SMS, WhatsApp and Viber — one brain, in English and Bulgarian, with appointment booking built in.
40+ ready-made agents, one click to deploy.
Invoice extractor, universal Mail, Excel and SQL agents, a web surfer, an API agent, a video maker and editor, five voice assistants (receptionist, support, sales, booking), a chatbot, agents for Slack, Teams, Google Workspace, SAP, Stripe, PayPal, GitHub and every major social network.
One sentence, a finished file.
Ask for a video and get a narrated MP4, ready to post. Ask for a website and get the finished pages, with a plain guide to putting them online. Hand over your old CV and get a new one as PDF and Word — better wording, no invented facts.
Posts to LinkedIn for you.
The LinkedIn agent publishes text, pictures and videos on your profile — make a video with the Video Maker and post it in the same run, or put it on a schedule. If LinkedIn refuses, the run says so and why.
Documents in, data out.
A ready-made Invoice document type reads PDFs, scans and photos, checks the numbers, and writes the result to Excel, your database, your ERP or an email — define your own document type once and reuse it forever. Search finds INV-1043 exactly, never its look-alike INV-1034.
Checks the company, not just the numbers.
Every invoice’s ЕИК is checked for a valid number and looked up in the Trade Register and EU VIES: a company that does not exist, a name that does not match, or VAT from a firm that is not registered for it is stopped for a person. The VAT and totals are recalculated too.
Try it before you trust it.
Test models lets up to three AI models read the same sample documents and shows which one got most fields right, how fast, and at what cost. Then a trial compares every result with your own spreadsheet, and the job goes live only after enough correct runs in a row.
Personal numbers stay hidden from the cloud.
Switch it on and ЕГН, IBAN, e-mail addresses and phone numbers are replaced with placeholders before anything is sent to a cloud AI model, and put back in the answer when it returns. The model works without ever seeing them. It covers text, not pictures, and it does not recognise names — the setting says so. With a local model nothing leaves at all.
Never stuck on one model.
Pick a backup model: if the main one’s service is down or out of credit, the step runs on the backup and the run log says so, with the cost of the model that really answered. Repeat runs on Claude reuse the unchanged part of the request, which in our test cost about a sixth of the first run.
Connects to what you already use.
Email, files, databases, browsers, Slack, Teams, Google Workspace, SAP, payment providers, social networks and any REST API — and a small Local Worker lets the cloud version read and write files on your own PC, so nothing needs to be uploaded.
Proof, not promises.
Stamp any document or PDF: a signature that proves it was not changed, marks it as AI-made, and locks the bank account and amount on an invoice — so a swapped IBAN is caught before it is paid.
Secure by design.
Two-factor sign-in, encrypted keys, rate-limited access, a tamper-evident activity log, signed updates with one-click rollback, and a published security brief mapped to NIS2 and ISO 27001 controls — with a plain list of what we do not claim.
See what it earns you.
The Analytics page shows every run, its cost, its success rate and the hours it saved — the ROI of your AI team in one screen.
Built to be the simplest AI agent platform there is.
A single automation professional built it, installs it himself and answers the phone — no ticket queue. The goal from day one: turn a complex technical agent set-up into a click, and replace fragile RPA robots with something that thinks.
New
Agent Passport
Give any AI agent a signed ID and track record — who it is, what it's allowed to do, how many times it ran and how well.
Digitally signed, so anyone can verify it's genuine — offline, no login. Proof you can hand a client, not "trust me".
From one-off tasks to complex multi-step pipelines — AIK handles it all.
AI Agents
Create specialized agents with their own tools, memory, and instructions. Run them on demand or on a schedule.
Automated Workflows
Chain agents together in multi-step pipelines. Visual scheduling, dependency tracking, auto-retry on failure, and pause for human approval when a step needs sign-off.
Drive it from Claude
Connect AIK to Claude Desktop or Claude Code and run your workflows and agents by asking. Anything Claude builds there arrives switched off, for you to check first.
MCP Integrations
Connect to email, files, databases, browsers, APIs and more via the Model Context Protocol server ecosystem.
Multi-LLM Support
Use OpenAI, Anthropic, Google Gemini, Groq, Ollama Cloud (large models, free plan), or local Ollama — switch anytime in Settings.
Live Monitoring
Track every run, token usage, cost, and success rate. Get notified via email or Slack on failures.
Team Workspaces
Invite team members, share agents and workflows, manage roles — collaborate without stepping on each other.
Document Processing
Turn invoices, contracts and forms into structured data — read PDFs and scans, extract fields, validate, and export to Excel, your ERP or a database. Define a document type once, reuse it forever.
Voice & Omnichannel
One AI assistant reachable everywhere — website chat, phone calls, SMS, WhatsApp and Viber — sharing the same brain. Bilingual, with appointment booking built in.
Enterprise Reliability
Health checks, failure isolation, per-server timeouts and graceful degradation. One-click safe rollback (code + database), signed updates with optional hands-free auto-update, and a tamper-evident audit log — chain-tampering prevention for GDPR & EU AI Act record-keeping.
How it works
Up and running in minutes
No complex setup. No coding. Just describe what you need.
1
Create an Agent
Pick a type — Email, Browser, Database, File, API — and describe what it should do in plain language.
2
Build a Workflow
Chain agents into multi-step pipelines. Add conditions, approvals, and schedules — all from the visual editor.
3
Let it Run
Hit Run or set a schedule. Monitor progress in real-time. Get notified on success or failure. That's it.
Real-world examples
What teams automate with AIK
These workflows run daily for real businesses — on schedule, with zero manual work.
Finance
Invoice Processing Pipeline
Check email for invoices → OCR extract data → validate amounts → upload to ERP → notify accountant. Runs every morning at 8 AM.
Marketing
Social Media Monitoring
Scrape competitor posts → analyze trends → draft content ideas → post to LinkedIn → send weekly report via email.
Operations
Daily Data Sync
Query SQL database → compare with API data → flag discrepancies → save report to shared folder → alert team on Slack.
Pricing
Simple, transparent pricing
Start free. No credit card, no hidden fees.
Hybrid deployment
Cloud, self-hosted, or both
Use our cloud platform instantly, install on your own server for full control, or mix both — cloud AI models with local data.
Cloud Ready
Sign up and start in seconds. No installation needed. We handle everything.
Self-hosted Option
Installed on your own server, by me — full privacy, data never leaves your infrastructure. Ships compiled so your code stays private, with signed updates. GDPR by design.
Hybrid Mix
Use cloud AI models (GPT, Claude) with local Ollama models — switch anytime with one click.
Local Worker
Even as a cloud user, run AI on your own computer with Ollama. Your data never leaves your machine. Free, private, no API costs.
Security
Secure by design
Built for teams that can't compromise on data protection — especially on your own servers.
In plain words
Your data can stay in your building. AIK runs on your own server — nothing has to leave it.
Every action is written down, permanently. Each entry is sealed to the one before it, so any later change — even by us — shows.
Anything the AI writes says so. Documents, emails, phone calls and website chat all declare they are AI.
A copy is taken every night, and a bad update can be undone with one command.
An order hidden in an email or PDF cannot make an agent act on its own: sending, posting, paying and deleting wait for your approval, and an allow-list refuses any address or site not on it.
Encrypted secrets
Every API key and credential is stored encrypted in the database (Fernet: AES with a tamper check) — never in plain text.
Tamper-evident audit log
Every action is SHA-256 hash-chained — any tampering is provable. EU AI Act Art. 12 record-keeping.
NEW · Agent Passport
Give any AI agent a signed ID + track record — who it is, what it's allowed to do, its runs and success rate. Digitally signed, so anyone can verify it's genuine offline, no login. Proof you can hand a client, not "trust me".
Every agent in its own folder
An agent's tools are locked inside their owner's folder by the operating system — the rest of the server simply is not there for them.
No third-party tool servers
Every tool server is our own code — no external MCP servers, so there's no supply-chain attack surface.
Access & transport security
Optional 2FA (Google Authenticator), expiring sessions with instant logout-all, enforced HTTPS/HSTS, rate-limiting and revocable API keys.
Work with me
On-premise setup, by the person who built it
Want AIK running on your own servers? I install and configure it for you — your models, your data, fully private,
GDPR by design — and stay on hand for support and tailoring it to your processes. No cloud vendor in the loop, no DevOps team needed.
I am Alexander Krastev, an RPA Developer & AI Automation Engineer in Sofia, Bulgaria.
Before automating the work, I did it myself: 11 years in banking (payments, SEPA/SWIFT, AML at ProCredit Bank)
and 5 years in accounting, doing every manual task by hand — invoices, reconciliations, payments, reports.
Today I automate those same tasks with UiPath and IBM RPA, and now with AIK.
“After years automating business processes with RPA, I watched AI capabilities explode and asked a different question:
what if automation could actually think? That’s what I built. AIK is a hybrid AI automation platform that
combines the reliability of traditional automation with the decision-making intelligence of AI agents — it doesn’t just
follow rules; it understands context, handles exceptions, and completes tasks end-to-end without human involvement.”
Most of us have run into the same thing: you want to do something simple — put together a CV, turn a PDF into Word or back — and you land in an ocean of sites, adverts and “free” services that walk you to the last step and then ask for money to download what you made yourself, or offer a stripped-down version with half the fields missing, or one free try and that’s it.
When you are curious and want to learn or try something new, the same thing waits for you: endless courses, tools and platforms that confuse more than they help. In the age of AI it got worse. Every day there is a new tool, a new “revolution” — and the pace is such that you feel helpless to keep up.
Read the rest →
I have been through different fields and companies, and I have always wanted to learn, to get better, and to fix or improve what I thought was not working as it should. I have not always met understanding — change is often met with reluctance. But when you see a process that is complicated, slow and full of mistakes, the only thing you want is to make it easier, faster and more reliable.
Right now I automate processes with RPA robots, and I see the weaknesses not only in the robots but in the whole road from the first meeting to the finished product. On one side: unclear and incomplete requirements, unstructured information, conditions that change halfway through, unrealistic deadlines, quality pushed to the back, pressure from managers and endless meetings without a result. On the other side, the robots themselves. An RPA robot is fragile: it sees the screen, not the meaning, and when the system it works with changes one button or one field, the robot stops. Every change means maintenance. It cannot read a scanned invoice unless it looks exactly as expected, it cannot decide in an unexpected situation, and it does not know how to say “I’m not sure”. To build one you need a licence, an orchestration server and a team to watch it — and still it does only what it was programmed to do, nothing more.
When the AI boom began, everyone wanted to understand what it was, touch it, play with it. Being curious, I tried to build agents — not chatbots you only talk to, but ones that act and get work done the way people do; an improved version of the RPA robot, the next level of process automation. And again I ran into pointless courses, platforms with heavy, complicated setup, and the feeling that I had to know the whole world of technology to make it possible.
That is how the idea was born: one complete platform for agents and automation, where everything is easy and fast and you do not have to be an engineer. I started it from an AI course — a small product that gave “hands” to agents, which until then were used only as a brain for thinking and for searching documents. I have been working on it for more than a year now, with Claude Code as a coding assistant, because the work is too much for one person: backend, frontend and everything else.
The platform has grown big, with countless features. But the idea is the same as at the start: take complexity away instead of adding new, and solve real problems faster and easier.
