coldread.dev

1. What Coldread is

Coldread is proprietary software developed and operated by PeachStateAI (Newnan, GA). The hosted service at coldread.dev, the CLI binary, and the MCP server binary are all distributed under a subscription. The CLI and MCP authenticate to the hosted service with a token minted at /mcp and are subject to these same terms when used.

2. Subscription & billing

Coldread offers four tiers: a free first run (no card), Early Access at $19 / month, Pro at $29 / month, and Max at $120 / month. Enterprise is contact-only. Each tier carries a monthly inclusion (50–75, 200, or unlimited subject to fair use) and a hard per-day cap. Runs that exceed your tier's caps are blocked until the period rolls over; switching tiers takes effect immediately via the Stripe billing portal.

Teams (up to 5 people) are included with Pro and higher at no additional charge. Team membership requires each member to hold their own active, team-eligible subscription (Pro or higher), and every run a member makes — in personal or team scope — counts against that member's own plan caps. Larger teams with central billing and custom limits are available as Enterprise Teams; contact hello@peachstateai.com.

Billing is handled by Stripe. PeachStateAI never sees or stores your card. Cancel any time from the billing portal (Profile menu → Manage subscription); your subscription remains active through the end of the current billing period.

2a. Money-back guarantee

Money-back guarantee. If you decide Coldread isn't right for you, contact us within 30 days of your first charge by way of /contact. We'll hop on a quick call (or trade emails if you prefer), hear what didn't land, and — on request — refund that first charge in full and cancel the subscription. The guarantee covers your first paid month; it doesn't stack across multiple billing periods, and it doesn't require you to prove anything beyond telling us why it wasn't pulling its weight.

3. Acceptable use

What you agree not to do, and a few notes on what we don't police.

You agree not to use Coldread to:

• Generate, score, or refine content that is illegal in your jurisdiction or in the United States — including but not limited to CSAM, content that solicits violence against an identifiable person or group, or content designed to enable a specific crime.
• Generate content intended to defraud consumers, deceive a regulator, or impersonate a real individual without that individual's consent.
• Use Coldread output to misrepresent that real people, real demographic groups, or real market segments hold any particular opinion. The reactions are written-persona output; attributing them to real humans without disclosure crosses from preflight signal into fabricated research.
• Copy, modify, redistribute, sublicense, or train any model on the persona definitions, simulation prompts, or simulation outputs. The Software is proprietary; subscription grants you a limited use right, not a redistribution right. Licensing inquiries: hello@peachstateai.com.
• Attempt to extract, reverse-engineer, or scrape Coldread infrastructure (the web UI, the serverless functions, the rate-limit boundaries) beyond normal product use.
• Use the URL-fetch endpoint (/api/fetch) as a generic SSRF or proxy — the endpoint exists to extract a marketing asset, not to relay traffic.
• Submit content that violates Anthropic's Usage Policies for the API.

Things we explicitly do not police:

• Whether your asset is good or bad. Coldread is here to surface that signal, not to pass judgment on your taste.
• Whether you act on the recommendations. Most users won't — that's fine.
• Whether the asset is real, hypothetical, satirical, or fictional. The system runs either way; the responsibility for what you ship is yours.

3a. Limitations and known constraints

What the system can't do, in plain English.

• Coldread reactions are generated by language models (xAI Grok or Anthropic Claude, depending on what the operator has configured) speaking as written personas. They are not real human responses and not a replacement for talking to your customers.
• The Americans panel is U.S.-anchored. International audiences are out of scope today; non-U.S. localization is a known gap.
• Asset size: very long assets (multi-page documents) are truncated to fit the model context. The web UI's text input is sized for marketing-copy lengths.
• Run length: a single run is capped at five minutes by Vercel function timeouts. Hybrid (full Americans + audience) runs may push that limit on slow Anthropic days.
• Cost: included in your subscription. PeachStateAI absorbs the model-provider charges within your tier's monthly + daily caps. See /privacy.

4. Output and accuracy

Coldread reactions are generated by language models speaking as written personas. They are not real human responses and are not market research. Use them as directional signal only. PeachStateAI makes no representation that any score, theme, or recommendation produced by the system reflects the views of any real person, demographic group, or market segment. Treat the output as an input to your judgment, not a substitute for it. See /methodology for the longer version.

5. No warranty

The hosted service, the CLI, and the MCP server are provided "as is," without warranty of any kind, express or implied. PeachStateAI does not warrant that the service will be uninterrupted, error-free, or available at any particular time.

6. Limitation of liability

To the maximum extent permitted by applicable law, PeachStateAI is not liable for indirect, incidental, special, consequential, or punitive damages arising out of your use of Coldread. Your sole remedy if you are dissatisfied is to stop using the service and cancel via the billing portal.

7. Changes

We may update these terms; material changes will be posted on this page with a dated revision note. Continued use after a change constitutes acceptance of the updated terms.

8. Contact

Questions? /contact.