Praus Review: $0 Sovereign AI Harness, $10 and $20 Plans Coming Soon

The pitch is simple: stop paying flagship prices for questions a small model could answer. Ortelan Labs launched Praus on November 3, 2026, a "sovereign AI harness" that sits between you and open-source AI models, routing each request to the model best suited to the task and checking the result before you see it. The free tier is live now at $0 a month, with Plus ($10) and Pro ($20) plans announced but marked "coming soon" — you cannot buy them yet. The headline cost figure, $0.00069 per answer, is the company's own estimate of its infrastructure cost, not your bill, and the service currently works only in the United States. Here is what Praus actually is, what its pricing fine print says, and where the honest catches are.

What Praus is: a harness, not a model

Praus is not a new AI model. As its own site puts it, it is "a harness, the layer that sits around AI models and decides how they are used." Because it is not tied to any single model, the company says newer open-source models can be added or swapped in as they improve, with no change for the people using it.

Behind Praus is Ortelan Labs, an applied AI lab in Salt Lake City that builds AI products, builds and operates AI systems for organizations, and invests up to $5 million per company in early-stage AI startups. The lab is led by chief executive Kyle Nel with partners Christopher Harper and Bradden Blair. Praus is one part of its work; the lab also describes a second product, an autonomous genealogy researcher called Lost Tribes, and client engagements including an audit of more than 80 systems at a national home services provider that it says identified over $12 million in potential savings.

The product's core loop is three steps: Auto routes each request to the models and tools that suit it, the response is checked for inconsistencies, and you are shown which model answered. "Most people don't need the biggest model for every question," Nel said in the launch announcement. "They need the right one, checked, at a cost that makes sense."

The feature set at launch includes web search and deep research reports, projects that keep chats, files and notes together, group chats with friends or coworkers, coding assistance with a reason given for each change, and voice conversations with read-aloud answers. Agents that carry out multistep tasks using tools are in preview for selected accounts; the company says they work through tools that run in the customer's own application or terminal, so every action happens in the customer's environment and leaves a checkable result.

The $0.00069-per-answer claim, unpacked

The number in the launch headline is $0.00069: Ortelan Labs' estimate of its own on-premises infrastructure cost for a typical answer. Praus runs open-source models on its own on-premises infrastructure, with additional capacity from cloud providers, and its pricing page frames a typical workload as 8,717 input tokens and 935 output tokens per answer, based on 165 real answers.

Its cost page then stacks that figure against public list prices for 17 frontier models. A few rows from the comparison: GPT-6 Luna at $0.00134 (1.9x the Praus figure), Mistral Small 4 at $0.00187 (2.7x), DeepSeek V4.1 Flash at $0.00374 (5.4x), Claude Sonnet 5 at $0.02678 (39x), and GPT-6 Astra and Claude Fable 5.1 both at $0.13392 — 194x the Praus estimate. Three cautions belong here, and the page itself supplies two of them: the figures are inference estimates, not subscription or API prices; they do not show equal answer quality; and the Praus number is a self-reported estimate while the rest use public list prices. In other words, the 194x headline compares the company's own infra arithmetic against rivals' retail prices.

On benchmarks, Ortelan Labs published Praus's results on nine public tests to Hugging Face. In the company's own testing, Praus ranked first among the listed systems on two — GraphWalks (100) and SimpleQA (81.6, ahead of Gemini 3.1 Pro at 74.8, GPT-6 Astra at 73.1 and Claude Opus 5 at 62.2) — and between third and sixteenth on the other seven, including MATH-500 (96.6, 3.0 points behind the top score shown) and GSM8K (93.3, 2.1 points behind the top score shown). The launch release carries its own honesty clause: benchmark scores and cost figures are self-reported estimates from the lab's own testing, have not been independently verified, and are not guarantees of performance, accuracy or price. Treat them as a company benchmark, not an independent verdict.

Pricing: Free is live; Plus and Pro are not

The pricing page, read live at publication, lists three personal plans. Free costs $0 a month with no credit card required: web search, image and file upload, notes and projects, and a monthly message allowance. That tier is live now.

Plus ($10 a month) and Pro ($20 a month) both carry the same line: "Paid plans are not available yet. Coming soon." Plus promises five times longer chats, a deep research request allowance, an agent task allowance, extended memory across tools and access to new features. Pro adds higher monthly allowances, more storage, and API access for integrations, and is for individual use.

The fine print worth reading: chat, deep research, agent and API requests share processing and web allowances, so complex tasks can hit those limits before the message count does. Agent tasks get a separate allowance, and their internal steps do not use monthly messages. For teams and businesses, the site lists Team and Enterprise tiers and API access priced on volume — all as "talk to us" custom arrangements with no published numbers.

In rough regional terms at market rates on November 3, 2026 (1 USD = 0.745 GBP / 1.401 CAD, verified at publication): Plus $10 is about £7.45 or C$14.00 a month, and Pro $20 about £14.91 or C$28.00 — when they become available. But Praus is currently available only in the United States, so for UK and Canadian readers these are reference figures for now.

The catches, honestly stated

US-only. The site says Praus is currently available in the United States, and the FAQ limits it to people in the United States who are 18 or older. The "sovereign" story has a US flavor: the release calls it "a U.S.-based sovereign AI harness."

"Sovereign" does not mean on-device. Praus's own FAQ spells it out: sovereign "does not mean that every request stays on your device or within one country." Runs happen on Ortelan Labs' on-premises infrastructure plus cloud overflow, and customers can set rules for how their data is used. On training, the FAQ says the lab does not use non-public conversation content for model training "without a separate adult opt-in," and temporary chats are excluded.

The paid features are announced, not available. Deep research allowances, agent task allowances and API access live on the Plus and Pro plans — which cannot be purchased yet. Agents remain in preview for selected accounts, the iOS app is "planned for later this month" (the release's forward-looking-statements section says such plans may change), and API access opens account by account.

Everything big is company-reported. Benchmarks, the $0.00069 cost figure, and the customer-savings figures: all come from Ortelan Labs' own testing or engagements, with no independent verification yet. That is not a disqualifier for a one-day-old product, but it is the frame everything should be read in.

Where Praus fits the sovereign-AI wave

Praus lands in the middle of a busy sovereign-AI season. Europe's Polarise cloud launched a sovereign AI cloud for enterprises in October, and Switzerland's Giotto models ship with published API pricing. Praus is the consumer-side cousin: a personal assistant pitched on data control and low cost rather than national infrastructure. The comparison below keeps the three apart:

On the personal-assistant side, Praus also overlaps with tools like Perplexity Computer (a Mac-local AI computer with its own credit catch) — different architecture, similar pitch about control over your AI.

Pro tips and common mistakes

Do not read the cost figure as a bill. $0.00069 is the company's estimated infrastructure cost per answer, not a subscription or API price. Your bill on the free tier is $0; on paid tiers it will be $10 or $20 flat — when those tiers exist.

Watch the shared allowances. Messages, processing and web allowances are shared across chat, deep research, agents and API requests, and complex tasks can hit the processing ceiling before the message count. If you burn through deep research fast, this is likely why.

Do not wait for the paid tier to try it. The free tier is the only live product. Sign up expectations should be: web search, notes, projects, and routed open-source answers — not the full agent/deep-research suite, which sits on coming-soon plans.

Take the benchmark table as marketing-adjacent data. The nine-test table is published on Hugging Face, which is good transparency, but the scores are company-run on company-chosen comparisons. Compare them against independent evaluations before making a switch decision.

FAQ

What is Praus?

Praus is a sovereign AI assistant launched November 3, 2026, by Ortelan Labs. It routes each question to the open-source AI model best suited to the task, checks the response for inconsistencies, and shows which model answered. It handles research, writing, code and voice conversations, with agents in preview.

Who made Praus?

Ortelan Labs, an applied AI lab in Salt Lake City that develops AI products, builds and operates AI systems for organizations, and invests up to $5 million per company in early-stage startups. It is led by chief executive Kyle Nel with partners Christopher Harper and Bradden Blair.

How much does Praus cost?

The Free tier is $0 a month with no credit card required, and it is live now. Plus ($10/mo) and Pro ($20/mo) are announced but marked "coming soon" — they cannot be purchased yet. Team, Enterprise and volume-priced API tiers are custom "talk to us" arrangements.

Is Praus available outside the United States?

Not yet. The site says Praus is currently available in the United States, limited to people 18 or older. UK and Canadian readers can watch the pricing reference figures (about £7.45/C$14.00 for Plus, £14.91/C$28.00 for Pro) but cannot sign up yet.

Does "sovereign" mean my data stays on my device?

No. Praus's own FAQ says sovereign "does not mean that every request stays on your device or within one country." Workloads run on Ortelan Labs' on-premises infrastructure plus cloud overflow, and customers set rules for how their data is used. The lab says it does not train on non-public conversation content without a separate adult opt-in.

Are Praus's benchmarks and cost claims independently verified?

No. All benchmark scores and the $0.00069 cost figure are self-reported estimates from Ortelan Labs' own testing. The launch release states this itself, and the cost page notes the comparisons are inference estimates that "don't show equal answer quality."

What about the iOS app and the API?

An iOS app is planned for later this month, and API access opens account by account; both are forward-looking statements in the launch release, which warns such plans may change.

Sources

• Praus official site — praus.ai (homepage copy, features, benchmark and cost pages; read live November 4, 2026)
• Praus pricing page — praus.ai/pricing (Free $0 live; Plus $10 and Pro $20 "not available yet. Coming soon"; read live November 4, 2026)
• Ortelan Labs Launches Praus — GlobeNewswire release, November 3, 2026 (syndicated copy; read in full)