Built for fast answers at a lower cost, with the context to make them useful to you.
Praus can make mistakes. Check important information. Answers aren't legal, medical or financial advice.
Research, projects, notes and group chats in one sovereign assistant. Keep your work together, with clear choices about how your information is used.
Deep researchResearching
What should I know before renting in a new city?
Reading sources · 0 of 20
Report ready · 20 sources
Get a quick answer with sources, or a full report when the question is bigger.
Kitchen remodel
4 notes · 2 chats
Keep each project’s chats, files and notes together, ready when you come back.
Lisbon tripASJ
Can we keep it under $1,500 each?
And somewhere near the water.
Here are three plans under $1,500 each, all within a short walk of the river.
Bring friends, family or coworkers into one conversation, with Praus there to help.
format-name.jsFixing
1function formatName(name) {2return name.split(" ")[0];3const parts = name.trim().split(/\s+/);4return parts[0] || "";5}
Handles extra spaces and empty names.
Fix a bug, learn a language or plan a build. Every change comes with a reason.
Agent4 tool calls
Reconcile the October invoices and report what is still outstanding.
Verified$55 outstanding · $5 credit
Illustrative task · results are checked before completion.
Give Praus a file or data task. It works in a private cloud workspace and checks saved results before returning them.
The intelligence each task needs, without unnecessary compute. Praus brings open-source models together to help keep answers fast and costs down.
New models can be added as they arrive, without changing your workspace or rebuilding your integrations.
Models work alone or together to suit your question, balancing answer quality, speed and cost.
Answers are checked for inconsistencies before you see them. Important details still need your judgment.
You choose the context Praus can use. Permissions govern access to your chats and files, with private deployment options for businesses.
Compare answer quality across public tests, or explore the estimated running cost of a typical answer.
Factual questions answered with web search
Highest of the 4 models shown
Multi-step reasoning over long documents
Highest of the 4 models shown
Competition math problems
3.0 points behind the top score shown
Multi-step math word problems
2.1 points behind the top score shown
More Praus results
Accuracy in percent, higher is better. How we measured · Full data
Praus, estimated
$0.00069
per answer
Next lowest
1.9×
GPT-6 Luna costs 1.9× as much
Flagship models
Up to 194×
the running cost of Praus, for the most expensive model shown
Log scale. Each step on the axis is ten times the cost of the one before. The line shows how much more each model costs than Praus.
A typical workload of 8,717 input and 935 output tokens per answer, based on 165 real answers. Praus’s figure is a self-reported estimate of running cost; other figures use public model list prices. These are inference estimates, not subscription or API prices, and they don’t show equal answer quality. Methodology and full data
Upgrade when you want more, or bring Praus to your team.
Plan details are temporarily unavailable. You can still start free.
For everyday tasks.
Includes
Unlock full intelligence.
Paid plans are temporarily unavailable.
Includes
For demanding individual work.
Paid plans are temporarily unavailable.
Includes
Chat, Deep Research, Agent and API requests share processing and web allowances. Complex tasks may reach those limits before the message count. Agent tasks have a separate allowance; their internal steps do not use monthly or purchased messages. Pro is for individual use. Family plans and household sharing are coming soon. View your current allowances and available one-time credits in Settings → Usage.
For you, your household and your team.
Research, writing, planning, decisions and code. Bring files, ask follow-ups, and keep ongoing work organized in projects and notes.
Praus doesn’t send every request to one large model. Auto routes each request to the models and tools that suit it, checks the response, and shows which model answered.
Pro includes everything in Plus with higher monthly allowances for messages, Deep Research and Agent tasks, plus API access. It is for individual use. Family plans and household sharing are coming soon.
It means an assistant built around control and choice. You retain your rights in your content. Data handling and processing locations are explained in our Privacy Policy; sovereign does not mean that every request stays on your device or within one country.
We do not use non-public conversation content for model training without a separate adult opt-in. Temporary Chats are excluded. Commercial content is not used for training under our API terms. See our Privacy Policy.
Yes. Research answers show their sources, so check anything important against them, especially before a big decision.
Praus combines specialized models, routing and verification in one system. Our public benchmark results show where it leads and where it trails, measured on nine public tests.
Many assistants use the same large model for every request. Praus picks the right model for each task, which avoids unnecessary compute and keeps quality high. That lets teams roll AI out more widely without costs rising at the same rate.
You retain your rights in inputs, and we assign any rights we have in outputs to you, subject to applicable law. Outputs may not be unique or eligible for copyright. Commercial content is not used for training under our API terms.
Contact us to discuss deployment requirements. Availability, processing locations and specific safeguards must be agreed in writing for your organization.
Start with a conversation. API availability and custom deployments depend on your requirements and the offering available at launch. We’ll recommend the right setup and timeline. Talk to us.
Praus is built by Ortelan, a laboratory for applied intelligence.
People in the United States who are 18 or older.