Trend synthesis
Public signals clustered into niches and scored for velocity, longevity and saturation — the free radar, and the input to everything else.
Open Trend Radar · a free tool for the community
Big sellers have trend analysts. Now every small seller does too — free.
We're building a free trend analyst for every indie print-on-demand seller. Every morning, Claude agents scan public trend signals across five regions, group them into niches and score each one for demand, saturation and trademark risk — then tell each seller which niches matter to their shop.
Our goal: 25,000 indie sellers on the free radar by the end of 2027.
- Free forever for individuals
- Open source at launch
- Powered by Claude agents
What we're building · Design Agent
From a trend to market-ready apparel, run by an agent
The radar tells you what is rising. The Design Agent — our paid product, planned for 2027 — will do everything after that: concepts, designs, mockups, marketplace listings and a synced catalog through our fulfillment partners Printful, Gelato and Prodigi. Our vision: an indie seller spots a trend in the morning and has approved, on-trend products live that afternoon — with an agent running the shop's busywork around the clock.
Product preview
Sample dataDesign Agent · concept
Preview with illustrative sample data — not live trends. Select a niche to see the kind of output the Design Agent will produce.
Design directives & artwork
Styling, palette and typography directives, turned into print-ready artwork and product mockups.
Listings for each marketplace
Titles, tags and descriptions written per marketplace and validated against a strict JSON schema — no broken listings.
IP & safety audit
Every design and listing screened for trademarked terms, protected characters, real-person likeness and harmful content. Risky items go to human review.
Catalog sync
Approved products pushed to your catalog through our fulfillment partners Printful, Gelato and Prodigi.
Continuous re-checks
As trends fade, the agent re-checks live listings and suggests refreshes or retirements — so your catalog keeps up without you.
Built on Anthropic Claude
The agent behind it
The Design Agent is a multi-step Claude pipeline. Every step is an API call; nothing is published without a person approving it.
-
Trigger
A trend is picked
The seller picks a niche from the radar, or a niche they follow crosses their velocity threshold.
-
Step 1 · Sonnet 5.5
Plan concepts
Reads the niche, the seller's shop profile and brand rules (prompt-cached) and writes 3–5 design directives.
-
Step 2 · tools
Generate designs
Calls an image-generation tool for each directive, then a fulfillment partner's mockup API for product photos.
-
Step 3 · Haiku → Sonnet
Audit images
Haiku 5.5 screens every image for print issues and trademark or likeness risk; Sonnet 5.5 reviews anything flagged.
-
Step 4 · Sonnet 5.5
Write listings
Titles, tags and descriptions for each marketplace the seller uses, as schema-valid JSON.
-
Step 5 · Haiku 5.5
Audit listings
Checks each listing against marketplace rules — title length, tag count, banned and trademarked terms — and sends failures back to step 4.
-
Human review
The seller approves
Designs and listings land in a review queue. The seller edits, approves or rejects — nothing goes live on its own.
-
Then · nightly
Sync & re-check
Approved products sync through Printful, Gelato or Prodigi. Every night the agent re-checks live listings against trend velocity and suggests updates.
Which model does what
- Claude Sonnet 5.5 — planner, writer, reviewer
- Clusters niches, plans concepts, writes listings and the daily digest, and makes the final call on anything flagged. It follows fine-grained creative direction like "Y2K, but quieter".
- Claude Haiku 5.5 — high-volume classification & auditing
- Built to triage 10,000+ raw signals a day across five regions, flags trademark risk, and audits every image and listing before a person sees it.
Tools the agent calls
- Trend board lookup
- Image generation
- Partner mockup APIs
- Public trademark search
- Listing schema validator
- Catalog sync (Printful, Gelato, Prodigi)
agent/plan_concepts.py
prompt-cached
# Shop profile + brand rules are cached; only the niche changes per call
response = client.messages.create(
model="claude-sonnet-5-5",
max_tokens=4096,
system=[{
"type": "text",
"text": shop_profile,
"cache_control": {"type": "ephemeral"},
}],
tools=[generate_image, render_mockup, trademark_search],
output_config={"format": CONCEPT_SCHEMA},
messages=[{"role": "user",
"content": niche_cluster}],
)Why we need Claude API credits
This only works with an LLM, every single day
Reading tens of thousands of messy signals, deciding which ones are the same niche, judging saturation, spotting trademark risk and matching niches to each seller's shop is language work. No rule-based script can do it, and it has to run again every day because trends move every day. That makes the free radar a large, continuous API workload — and credits go to the free community tier first.
| Task | Tokens per run | Runs / month | Monthly tokens |
|---|---|---|---|
| Signal triage Haiku 5.5 · batch of 50 signals (10,000 signals/day) | ~4k in + ~1k out ≈ 5k | 200/day × 30 = 6,000 | ≈ 30.0M |
| Niche clustering & scoring Sonnet 5.5 · per region, twice a day | ~60k in + ~8k out ≈ 68k | 5 × 2/day × 30 = 300 | ≈ 20.4M |
| Trademark risk flag Haiku 5.5 · per candidate niche | ~1.5k in + ~0.3k out ≈ 1.8k | 500/day × 30 = 15,000 | ≈ 27.0M |
| Daily digests & board copy Sonnet 5.5 · per region | ~20k in + ~4k out ≈ 24k | 5 × 30 = 150 | ≈ 3.6M |
| Personal niche alerts Haiku 5.5 · per seller, weekly | ~2.5k in + ~0.5k out ≈ 3k | 5,000 × 4 = 20,000 | ≈ 60.0M |
| Total, free community tier (year one) | ≈ 141.0M / month | ||
Projection at our 12-month target (25,000 sellers with weekly alerts): shared scan 30.0M + 20.4M + 27.0M + 3.6M = 81.0M, plus alerts 25,000 × 4 × 3k = 300.0M ≈ 381M tokens / month for the free tier alone. Target, not a measurement.
- The shared scan runs once per region for everyone; personal alerts grow with every seller who joins — which is why usage climbs with the community.
- The Design Agent (paid, later) is heavier per seller. Example: a full-shop audit and refresh of 300 listings at ~6k input + ~1.5k output tokens each ≈ 2.25M tokens per sync, before images and nightly re-checks.
- Prompt caching keeps repeated context (shop profiles, rules) cheap, but every step above is still a real model call.
Roadmap
The next 24 months
Free tool first, paid agent after — and a trend analyst in every indie seller's pocket by the end of 2027.
-
Now · Q4 2026
Open Trend Radar prototype
Daily scan, niche clustering and scoring agents; early-access list for the first digests.
-
Next · H1 2027
Public launch, open source, 5 regions
Free digests and web board for the US, UK, EU, Canada and Australia, personal niche alerts, and the public repository under the MIT license.
-
Later · H2 2027
Design Agent (paid)
Concepts, designs, mockups, listings, audits and catalog sync through Printful, Gelato and Prodigi, with continuous re-checks.
25,000
Year-one target: sellers on the free radar
5 regions
Target by mid-2027: regional radars
Morning → afternoon
Goal: trend to approved product the same day
Goals and targets, not current results.
FAQ
Straight answers
Is the trend radar really free?
Yes. Open Trend Radar is free forever for individual sellers, and the code will be open source at launch. The paid product is the Design Agent, which does the per-shop work.
Is it available today?
Not yet. We're building it now. Join the early-access list and we'll email you when the first digest goes out. The dashboard on this page is a preview with sample data.
How do you handle trademarks?
Every niche, design and listing gets an automated trademark-risk flag, and anything risky goes to human review. Flags reduce risk but are not legal advice — publishing decisions stay with you.
Who will own the designs the Design Agent makes?
You will. Designs and listings you approve are yours to sell. We won't resell or reuse your designs or shop data.
Is my data used to train AI models?
No. We use Anthropic's commercial API, which does not train on customer inputs or outputs by default.
Catch the next trend before it peaks.
Get the free daily radar the day it launches.