In Pro mode, WaxFrame sends prompts directly to each AI automatically — no copy/paste, no tab switching. Each AI needs its own API key. Follow the guides below to get set up.
In Pro mode, WaxFrame sends your prompts directly to each AI and fills in the responses automatically — no copy/paste, no tab switching. An API key is a private password that lets WaxFrame talk to each AI on your behalf.
Without API keys you can use WaxFrame Free — the same hive workflow, but you copy and paste prompts manually between WaxFrame and each AI's web interface. API keys are only needed for the fully automated Pro experience.
Your API keys are stored in your browser's localStorage and sent to each AI provider only when WaxFrame makes a request on your behalf. WaxFrame has no backend — there is no database holding your keys, your prompts, or your documents.
One exception — Claude. Anthropic's API does not currently support direct browser requests (no CORS), so Claude traffic routes through a small Cloudflare Worker operated by WaxFrame's developer at waxframe-claude-proxy.weirdave.workers.dev. The worker passes your request straight through to Anthropic and does not log or persist your API key or prompt content, but it does see them in-flight while forwarding. (The link goes to the live Worker endpoint, which expects API requests — opening it directly in a browser returns an API error rather than a page. That's normal.) All other providers (OpenAI, Gemini, Grok, Perplexity, DeepSeek) connect directly browser-to-provider with no intermediary.
Where exactly they live: your API keys are saved in your browser's localStorage (the same mechanism that remembers your dark-mode preference, your hive setup, and your project state). This is per-browser-profile, per-machine — keys you save on your work laptop don't appear on your phone, and clearing browser data for this site removes them. They are not encrypted at rest; any code or extension running in this browser profile that can read localStorage can read them. That's the same threat model as any browser-stored credential. For shared or untrusted devices, use a separate browser profile or clear data when finished.
WaxFrame — open source, no backend, no tracking. Learn more →
gemini-2.5-flash
Google's fast multimodal model
Gemini has a generous free API tier — no credit card required to start. Important caveat: once you enable billing on your Google AI Studio account (add a credit card), the same Gemini model can start routing through paid-tier paths and charging per token, even when you think you're still on free. Costs can climb quickly with the Builder role specifically — it reads project setup, reference material, the full working document, and every reviewer's suggestions every round. For casual use, keep billing off. If billing is on, watch your Google AI Studio usage page during longer runs.
deepseek-chat
DeepSeek's flagship model
DeepSeek is 10–20x cheaper per token than OpenAI or Anthropic. Add credit at platform.deepseek.com → Top Up — $5 gets you a lot of usage. Excellent Builder choice on a budget.
gpt-4.1
OpenAI's flagship model
Your ChatGPT subscription does NOT include API access — the API is completely separate and pay-as-you-go. Go to platform.openai.com → Billing and add credit ($5–$10 gets you plenty of testing).
claude-sonnet-4-6
Anthropic's capable everyday model
Requires credit balance. Go to console.anthropic.com → Billing and add credit (~$5 minimum). You are only charged for what you use.
grok-4
Requires credit balance — add at console.x.ai → Billing. API availability may vary — check the console for current status.
sonar-pro
Requires credit balance — add at console.perplexity.ai → Billing → Add Credits.
Perplexity offers a recurring API subscription as low as $5/month — but you must enable auto-pay during signup to lock in that price. Without auto-pay, the recurring rate jumps to $50/month. The $5 tier provides a solid monthly credit allocation that covers most WaxFrame usage without thinking about per-call costs.
Every AI in your hive has its own personality, strengths, and quirks. Knowing how each one behaves helps you pick the right Builder, get the most out of your reviewers, and decide when to toggle them off. This guide is based on real-world observations using WaxFrame in production — and on the hive itself, since each AI's profile below was refined by passing it through the hive for self-description and peer review.
A solid all-rounder for both Builder and Reviewer roles — fast, reliable, and consistent. A strong alternative Builder if DeepSeek is unavailable. As a reviewer, it effectively signals document convergence during final refinement rounds.
When ChatGPT returns NO CHANGES in later rounds, treat this as a sign of document convergence rather than a reason to toggle it off.
As a reviewer, Claude excels at catching subtle issues with tone, redundancy, and flow that others miss. As a Builder, it is reliable and format-compliant. Higher cost, but worth it for quality-sensitive projects.
Claude tends to give the most thoughtful reasoning for its suggestions — clarifying the logic behind each change.
The default Builder choice. Rarely deviates from instructions, making it a stable foundation. Fast, cost-effective, and reliable across document types.
DeepSeek + Gemini is the most cost-effective hive combination — one paid Builder at rock-bottom prices, one free reviewer.
Your primary free reviewer. Great for catching grammar, word choice, and structural issues. Use it as a reviewer and let a paid AI serve as the Builder.
Gemini + DeepSeek is the minimum viable paid hive — two AIs, one free, total cost often under a cent per round for short documents when Gemini stays on its free tier. If you've enabled billing on your Google AI Studio account, Gemini can start billing per token — keep it as a Reviewer rather than the Builder in that case, since Builders read everything every round and amplify cost.
A valuable reviewer in early rounds, pushing hard for stronger phrasing and action verbs. However, its tendency to revisit resolved points may prolong final convergence.
Toggle Grok off in later refinement rounds once key decisions are finalized to avoid repetitive suggestions.
A specialized reviewer for factual or research-heavy documents. Its search-aware training makes it uniquely effective at spotting unsupported claims, though its inconsistent output formatting may complicate document reconstruction.
Perplexity is particularly useful early in a project when the document is rough and needs fact-checking input from the hive.
You don't need all six AIs every round. Toggle them on or off at any time, except for your Builder. For a lean setup, consider DeepSeek as your Builder, Claude for nuanced feedback, Grok for early assertive input, and Gemini for free support — that's a well-balanced hive. A smaller hive completes rounds faster, while the full hive ensures the most thorough review possible.
Toggle, don't remove. Use the checkboxes on the work screen to toggle individual AIs on or off between rounds while preserving their API keys. This is the most efficient way to manage persistent reviewers.
Your Builder is the most important choice. The Builder processes the entire document along with all reviewer suggestions each round, using significantly more tokens than any reviewer. Choose an AI you trust with large amounts of text, one with sufficient rate limits and token budget for your project scope.
Watch for convergence signals. When multiple AIs start returning NO CHANGES in the same round, the hive signals that the document has converged and may be ready for finalization or manual review.
WaxFrame is fully open source under the AGPL-3.0 license. Every line of code is public and auditable.
Your API keys go directly from your browser to each AI provider, with one exception: Claude traffic routes through a small Cloudflare Worker (no CORS support from Anthropic) that does not log keys or prompts. See the API Setup page for the full disclosure. Read the source on GitHub →
About
Built with ❤️ by WeirDave and Claude.
Your WaxFrame Pro license is active.