Compare settings, not just brands.
Two or three configurations in parallel. Test the same model at different reasoning or verbosity levels.
A MAC APP. YOUR OWN API KEYS.
One question.
Several AIs.
Your choice.
Compare answers from several models, protect sensitive information on your Mac, then challenge the results with an audit.
Explore the screensPrivate beta in preparation · macOS 26 or later
Compare AI models. Keep your data private.
PRIVATEARENA
Four steps. You stay in control.
Illustrated interface previews with a completely fictional example. This page sends no AI requests.
PrivateArena♧ Protected on this MacYour request
Your request · TXT, Markdown, DOCX, PDF
Prepare a launch plan for recycled notebooks from the brand Boreal Studio. Our lead, Camille Laurent, wants to compare an Instagram campaign with a newsletter. Budget: €3,000.
▧ launch-brief.txt
Fictional exampleContinue
Models and settings
Compare the same model with different settings, or several models.
gpt-6.1-sol
Reasoning HighVerbosity MediumWeb search Yes
Claude Opus 5.5
Reasoning HighVerbosity —Web search Yes
Grok 4.7
Reasoning HighVerbosity —Web search Yes
BackFictional exampleCompare
Answer comparison
gpt-6.1-solReasoning : High · Web : Yes
A focused campaign
Present the notebook in use, show how it is made and prepare a short series of posts. Measure useful visits, then adjust the budget.
- A clear goal
- A tracked budget
- Results to measure
Estimated cost : ≈ $0.019
Fictional exampleClaude Opus 5.5Reasoning : High · Web : Yes
A launch sequence
Start with a newsletter to a qualified audience, then support the campaign on Instagram. Compare conversion rates over the same period.
- A clear goal
- A tracked budget
- Results to measure
Estimated cost : ≈ $0.025
Fictional exampleGrok 4.7Reasoning : High · Web : Yes
Test before expanding
Split a test budget between both channels. Keep the same message and offer to identify what really works.
- A clear goal
- A tracked budget
- Results to measure
Estimated cost : ≈ $0.018
Fictional exampleBackFictional exampleAudit
Audit reports
8.8/ 10 · Fictional example
gpt-6.1-solGood request coverage
Strengths: the plan remains consistent with the budget.
Check: expected conversions and acquisition costs.
The score is an AI assessment, not a certification.
Estimated cost : ≈ $0.031
7.5/ 10 · Fictional example
Claude Opus 5.5Assumptions to check
Strengths: the plan remains consistent with the budget.
Check: expected conversions and acquisition costs.
The score is an AI assessment, not a certification.
Estimated cost : ≈ $0.035
8.2/ 10 · Fictional example
Grok 4.7A concrete, measurable test
Strengths: the plan remains consistent with the budget.
Check: expected conversions and acquisition costs.
The score is an AI assessment, not a certification.
Estimated cost : ≈ $0.028
BackFictional exampleExport PDF
The session remains available until you close the app. Export a PDF to keep it.
Protect what needs protecting.
Sensitive information is analysed on your Mac. Review is optional: correct a missed item or keep the brand names your question needs.
Keep a useful record.
Return to answers and audits without new requests. Export your question, settings, answers, scores and costs as a PDF.
WHO IS IT FOR?
For people who want to choose,
not simply accept the first answer.
Consultants & independents
Compare analyses, prepare a recommendation and make its limitations clear.
Marketing & communication
Challenge strategies, editorial angles and launch plans.
Teams & professionals
Test models and settings on the questions that matter to their work.
PrivateArena is initially for users who already have API accounts. No programming knowledge is needed.
PRIVACY
Privacy starts
before you send.
Selected information is replaced locally. Your chosen providers receive the prepared text; names are restored on your Mac in the answers. No PrivateArena server receives your questions.
Pseudonymisation reduces exposure without guaranteeing complete anonymisation. You can review detected items before continuing.
Read the privacy policy →Your MacLocal protectionChosen providerRestored answerLocal PDF
Before you begin
Does everything work offline?
Preparation and protection are local. Comparison and audits need an Internet connection to AI providers. PDF export happens on your Mac.
Why use my own API keys?
Choose providers and use your accounts. Add keys in Settings; they stay in macOS Keychain. AI request costs are separate from the app purchase.
Can I compare a model against itself?
Yes. Test the same model with different settings, including within a single provider. Up to three configurations run in parallel.
Does an audit guarantee accuracy?
No. It looks for inconsistencies, omissions and errors using web search, then assigns a score out of 10. It is still an AI assessment: check its sources and limitations.
Is my question saved?
PrivateArena creates no permanent conversation history. Results remain available during the session. A PDF is the archive you choose to keep.
COMING SOON
A place for PrivateArena on your Mac.
A private beta comes before the Mac App Store release. Contact us to join the testing panel.
Join the testing panelMac App Store · coming soonPlanned launch offer: a 7-day app trial, then a one-time purchase of €29.99. AI requests are billed separately by your providers.