Your language
Write in your own language. On Auto, model selection adjusts with each message, and you can see which model was selected and why.
The mid-sized Mistral: quick, good in many languages, and the model Alpha starts on.
Alpha, by AI Potluck
Your language. Your data. Your rules.
Use your own language, see which model responds and where it runs, control what gets stored, and decide how Alpha responds to you.
Which open projects work on speech, and what can you actually say about them?
I can name projects our catalog lists, and I will not invent details the sources do not support. The summary under this reply shows what Alpha checked for this answer.
What you get
Using AI shouldn’t mean accepting settings you can’t change or trusting a process you can’t see. Alpha brings together models that work in your language, control over your data, and settings for responses that work for you. With an X-ray view of every response, you can also see which model generated it and where it ran.
Your language
Alpha’s model picker detects your language and selects an open model based on its performance in that language. Where available models fall short, you can see the gap and find ways to contribute. If a response isn’t useful, switch models in a click.
Write in your own language. On Auto, model selection adjusts with each message, and you can see which model was selected and why.
Mistral Medium 3.5 by Mistral AI CatalanMedium taskThe mid-sized Mistral: quick, good in many languages, and the model Alpha starts on.
Your data
See what gets stored, where it’s kept and who stores it. Your chats are not used to train AI models. With an account, you can view and delete stored files, erase your data without deleting your account, and download an export of your conversations and files.
Your chats are stored by Current AI in Alpha’s database and are not used to train AI models. Guest chats are automatically deleted after 30 days, or you can erase them sooner.
See the energy used to generate your responses and the time you’ve spent. Alpha is designed to be useful, not to keep you hooked.
19% of one phone charge About 17 min of a 10-watt LED bulb
Your rules.
Choose a theme, set the tone and length of responses, and leave model selection on Auto or choose an open model yourself. Your settings apply from your next message, in new and existing chats.
Choose a theme or create your own in Settings. Your colors, corners and type apply throughout Alpha.
Here’s a simple menu that travels well, plus a checklist so nothing gets forgotten…
Alpha · light
Short responses or detailed explanations? Bullet points or prose? Emojis or none? Choose what works for you. Your settings apply from your next message, in new and existing chats.
A heat pump moves heat instead of making it. A fluid absorbs warmth from the outside air, a compressor squeezes it so it gets hotter, and that heat is released inside your home. In summer the cycle can run in reverse to cool.
Use Auto to match each message with an open model, or set a favorite as your default.
See inside every answer
Alpha puts the details behind each response in one place. Open its X-ray view to see who built the model, which organization ran it, and which sources and checks were used—step by step, in plain language.
Select X-ray beneath any response for a quick overview, or explore Step by step for a closer look.
X-ray · “Our city wants to use AI to triage citizen emails. What does the EU AI Act require of us?”
Mistral AI (France). A European AI lab known for strong multilingual models.
Mistral AI in France. From this answer’s routing record.8 steps · about 13 s from sending to done. Only what actually happened is listed.
Your message was sent to Alpha.
Your message was checked for harmful content. Nothing was flagged.
This runs first, so a harmful request never reaches the AI.
Searched for “EU AI Act deployer obligations public authority” on Oct 29, 2026, and read 2 pages from the open web and Wikipedia.
The numbers in the answer, like [1], point to these pages so you can check them.
Mistral Medium 3.5, made and served by Mistral AI, wrote the reply word by word.
It was told to stick to the sources above.
Each sentence was checked before you saw it. All clear.
Looked for invented links, sources or names. Nothing needed fixing.
AI can sound sure about things that aren’t real — this catches the common cases.
Stored by Current AI in Alpha’s own database. You can delete it any time in Data & privacy.
You asked 2 other open models to answer too — they showed high agreement. Where ALIA 40B ran was not recorded. Where Bielik 11B ran was not recorded.
Before you start
Before you trust a new AI with your questions, you deserve a few answers.
Alpha is an AI assistant built by Current AI with contributions from across the AI Potluck ecosystem. Ask questions, explore ideas and work with your files using open models. Each response includes an X-ray view showing which model generated it, which organization ran it, and which sources and checks were used.
No. You can start chatting as a guest; guest conversations are deleted automatically after 30 days. Create an account when you want to keep your history and your settings across devices.
Yes, in many languages. On Auto, Alpha’s model picker detects the language you’re writing in on your device and selects an open model for each message based on its performance in that language. You can see which model generated each response and switch if an answer isn’t useful. Where available models perform poorly, Alpha highlights the gap and offers ways to contribute language data and evaluations.
Your conversations and settings are stored by Current AI in Alpha’s database. Your chats are not used to train AI models. See our Privacy Notice for details on what gets stored, where it’s kept and for how long.
Yes, any time. Settings → Data & privacy has one button that erases your conversations, settings and uploads. You can delete a single chat from its menu, and export your history before you go.
Each model is run by a named organisation — Apertus by Infomaniak in Switzerland, for example. Open X-ray on any answer to see who made the model and who ran it for that reply.
You check it, and Alpha makes that easy. When an answer uses sources, they’re numbered and linked. X-ray’s Step by step shows whether Alpha looked something up or answered from memory, and you can ask other open models for a second opinion. For anything medical, legal or financial, treat Alpha as a starting point, not the final word.
Every message is checked for harmful content before a model sees it, and Alpha checks answers for invented links, sources and names. Alpha is also honest about being a machine: no pretend feelings, no flattery, no personality designed to keep you chatting.
An estimate of the electricity your answers used this session, next to the time you’ve spent. We show it because AI has a real cost, and because Alpha isn’t built to keep you here longer than you need.
You shouldn’t have to take a company’s word for how its AI works. Open models give researchers, public institutions and people like you ways to inspect, test and improve the technology. How much you can check—including what a model was trained on—depends on what’s been shared. Explore the Gap Map to see what’s available across the open AI ecosystem and what’s still missing.
Open calls
Some of what Alpha needs is code. Much of it is data in more languages, people willing to test it in the real world, and partners to run open models. If one of these is yours, come build it with us.