01Legal teamExample
Summarise the termination clauses in this lease.
qwen3:4bLocal
Soriku measures what every model can do and sends each question to the one that fits best. Fully local if you want, in the EU if you prefer, built in the Netherlands.
Download freemacOS and Windows now · Linux soon · no account needed
01Legal team
Works with
Run locally
- Ollama
- LM Studio
- llama.cpp
- vLLM
- Your own model
Open models
- Llama
- Qwen
- Mistral
- DeepSeek
- Gemma
- Phi
Your own keys
- Claude
- OpenAI
- Mistral API
- Groq
- Gemini
Tools via MCP
- Gmail
- Google Drive
- Notion
- Slack
- GitHub
- Exact Online
How it works
Connect your models once, and Soriku takes it from there.
Ollama7 modelsLocalOnline
LM Studio2 modelsLocalOnline
Soriku EUHosted in the EUEUOnline
Claudeclaude-sonnetOwn keyOnline
Connect your models
Local through Ollama or LM Studio, remote through your own keys. Soriku finds what is on your machine by itself.
CategoryChosen modelScore
Code generationqwen2.5-coder:14b0.97
Code reviewqwen2.5-coder:7b0.85
Security reviewqwen2.5-coder:7b0.83
Long contextgpt-4o · own key0.88
Creative writinggemma3:4b0.94
Soriku measures them
A short test per model and per category, on your own hardware. The result is a score you can read, with how sure it is.
Write a short story about a lighthouse keeper who finds a letter in a bottle. Around 300 words, calm tone.
Chosengemma3:4bCreative writing
Highest measured score for creative writing on this machine.
- gemma3:4b0.94
- qwen3:4b0.86
- llama3.1:8b0.79
Each question goes to the best fit
The question is recognised, the map is read, and the highest measured score wins. You see which model answered, and why.
A walk through the app
Each question goes to the model that measured best for it.
Soriku recognises the kind of question, reads your capability map and compares the candidates before it answers. Under every reply you can see which model ran and why.
Summarise this rental contract and list the notice periods.
Your data goes where you say, and nowhere else.
Local by default, the EU when you allow it, never a guess.
Read about EU complianceWhat it does
Everything Soriku does starts with what it measured on your machine.
Measured on your hardware
Every model gets a scorecard across 13 categories, with how certain each score is. Your own numbers, not marketing benchmarks.
CategoryChosen modelScore
Code generationqwen2.5-coder:14b0.97
Code reviewqwen2.5-coder:7b0.85
Security reviewqwen2.5-coder:7b0.83
Each question to the best fit
Soriku reads the map and picks the model that scored highest for that task, within your rules for cost, speed and privacy.
Chosengemma3:4bCreative writing
Highest measured score for creative writing on this machine.
- gemma3:4b0.94
- qwen3:4b0.86
- llama3.1:8b0.79
You approve what matters
Anything that sends, writes or deletes waits for your yes. Mail, invoices, files, CRM.
ApprovalWaiting for you
Send this mail to anna@studio-rood.nl?Subject: The keeper's letter · 1 attachment · via GmailApproveNot now
Your whole network in one view
Every model, device and connection on one board. What runs where, what is online, what may go out.
Bring any subscription
Use the Claude, OpenAI or Mistral account you already pay for, through your own API key. No markup.
Train and share agents
Agents with their own memory and tone, each running on the model that measured best for its work.
How it picks
Soriku measures what each model can do. Then it picks for you.
When you add a model, Soriku tests it on your own computer: writing, summarising, code, translation and nine more categories. That gives you a capability map, a scorecard you can read. Every question goes to the model that measured best for that kind of task, and you can always see why.
13Categories
9Models measured
0.97Highest score
Example data. Your scores depend on your hardware and models.
Open the capability mapCategoryChosen modelScore
Code generationqwen2.5-coder:14b0.97
Code reviewqwen2.5-coder:7b0.85
Security reviewqwen2.5-coder:7b0.83
Long contextgpt-4o · own key0.88
Creative writinggemma3:4b0.94
Instruction followingqwen3.5:4b0.91
Instruction formattingphi4-mini:3.8b0.86
Multilingualqwen3.5:4b0.89
Reasoningdeepseek-r1:7b0.82
Speedgemma3:4b0.95
Summarisingqwen3:4b0.97
Tool callinggpt-4o · own key0.87
Translationphi4-mini:3.8b0.93
Who it's for
Made for people who want AI to work for them, not the other way round.
Photo · EqualStock / UnsplashYour own AI on your own laptop.
Free, no account, and it keeps working offline. Write, translate, plan and learn without your questions ending up in someone else's dataset.
Photo · Austin Distel / UnsplashAI for quotes, mail and admin, with customer data that stays in.
Connect the tools you already pay for. Soriku picks the right model per task and you approve anything that sends, writes or deletes.
Photo · Vitaly Gariev / UnsplashHealthcare, legal, government. Local or EU, never a guess.
Run every model on your own hardware, or in an EU data centre. Every request lands in a log you can show an auditor.
Trust
Who builds Soriku, and what we do with your data.
- Built in Utrecht
- A small team in the Netherlands. You can visit us, and you can email the people who write the code.
- EU servers
- Soriku EU runs on servers inside the European Union. Your questions stay there unless you connect a model that lives elsewhere.
- Open source core
- The router and the benchmark are open source. You can read how a decision is made and run it yourself.
- No training on your data
- We do not train models on what you ask, and neither do the models we host. That is in the contract, not only on this page.
- Every request logged
- Which model answered, where it ran and which rule applied. You can read the log and export it.
I started Soriku because I wanted to use AI without handing over everything I work on. Every choice we make starts there, and if one ever doesn't, I want to hear it.
Marten TimanFounder, SorikuWhy Soriku
More than a chat window.
SorikuA single AI chat subscription
Picks the best model per question–
Works fully offline–
Use any model, also your own subscriptions–
Data stays on your machine or in the EUdepends on plan
Train and share your own agentspartly
See your whole network in one view–
Price€0 local, €12 Prousually €20+ per month, one provider
Pricing
Free on your own computer. Fair when you want more.
The app and local mode are free, for good. You pay for hosting, team features and people you can call.
Starter
€5per month
For freelancers. A monthly allowance on Soriku EU and your setup synced across two devices.
Choose StarterPro
€9 for early adopters€12per month
Soriku EU for heavier models, access from anywhere, priority support.
Choose ProThe Atypisch family
Soriku is one of a family of tools built by Atypisch.
They share one login, one bill and the same rule: your data stays where you put it. Soriku already routes questions inside several of them.
The name Soriku joins Japanese words and means orchestration in an open space.
- SimezuAuth and payments
One login and one bill across every Atypisch product.
- SumeziNewsletters
Write, send and measure newsletters from one place.
- RozuroAccounting
Bookkeeping for freelancers and small businesses.
- WemazuDeployments
Ship sites and apps to European servers.
- Modular CMSOpen source CMS
A content management system you host wherever you want.
- AkyruLightweight language modelOur own model
A small model we train ourselves, built to run on an ordinary laptop. It is measured on the same capability map as every other model, so Soriku only picks it where it scores.
Start today, free on your own computer.
Download Soriku, connect a model and ask your first question.