Note Beta. This is the initial public release. The agent will sometimes be wrong about an item's identity, grade, value, or potential errors, and it can miss valuable details. Improving it takes ongoing testing, tuning, and development. Use the results as a fast first read, not a final answer.
AgentCollectibles is an AI agent that grades your collectibles from photos. Take a picture of the front and back of a coin or card. Submit it. The agent tells you what it is, estimates the grade, flags potential errors or varieties it finds, and gives you a value estimate. If an error has a backstory, the agent may include it.
#Why we built this
Getting an item professionally graded is expensive and slow. Before you pay to send an item to PSA, PCGS, CGC, etc., you usually want to know whether it's worth it.
The usual alternative — walking into a local shop — has its own problems. The person at the counter is relying on personal knowledge and whatever research they happen to do. Sometimes they're great. Sometimes they're not. And sometimes they're not motivated to be honest with you: they might quote you face value on a coin without mentioning it's a known error variety, miss the story that turns an ordinary-looking item into a sought-after collectible, or simply not catch what they're looking at.
That's the gap AgentCollectibles is built to close.
#What the agent does
Identifies what it is — Denomination, year, mint mark, set, parallel, variant — the agent tells you what you have.
Estimates the grade — Condition assessment using the standard grading scale for the category.
Flags potential errors and varieties — Doubled dies, off-center strikes, miscuts, printing errors — details that may make the item noteworthy.
May include the backstory — Historical context or what makes a variety sought after — when there's a story to tell.
Estimates market value — Based on recent comparable sales when available.
Works on a phone — Snap front and back. Submit. That's it.
Today, it doesn't certify authenticity or replace a grading company's determination. The goal is to give you an honest, fast first read on what you have, so you can decide what to do with it.
#Why is this possible now?
Grading items from a photo is difficult to do well. AI and computer vision have gotten a lot better in the last couple of years — new AI models keep being released, giving us more opportunities to test and improve the agent.
That gives us two ways to work toward better grading:
New models can help improve accuracy. When a new, smarter AI model is released, we need to benchmark it against known grades, test it, and tune the agent to see whether it improves the results.
Your feedback helps guide development. When you send an item in for professional grading and report back, that comparison helps us see where the agent gets it right or wrong. We use that evidence to guide further testing, tuning, and development.
#The vision
You pull out your phone. You take a picture of the front and the back of an item. You submit it. The agent gives you a first read on what it is, what it might be worth, and any potential errors or varieties. If an error has a backstory, the agent may include it.
That's what this product does today — with the caveat that it's beta, and improving it takes continued development.
#Next
Getting Started — Create an account and grade your first item.
Grading — Bulk Scan for triage, Individual Grade for the deep read.
Studio — Get an item ready to publish.
Collection — Your public profile and private inventory.
Settings — Profile, billing, and usage.
Pricing — Plans, and how tokens are billed.
#AI capabilities, bugs, and issues
AI is moving fast, and we're continuing to develop and test the agent — but it will still be wrong sometimes. That's the honest reality of where AI technology is right now for this type of use case, and it's why this is shipping as beta.
How to think about it:
Use the agent as a tool, not a verdict. The practical value is being able to check items immediately without mailing each one in or carrying them into a shop. Even a directionally correct grade can save you time and money, and gives you an idea of what might be worth paying for a professional grade.
We track accuracy against real datasets. The system is benchmarked, and we're working to push consistency and accuracy higher.
New AI models can help improve accuracy. That takes benchmarking, testing, and tuning the system to see which changes improve its grading.
Tip Pro tip: If the agent doesn't detect an error you're pretty sure of, try running it again. A second pass on a fresh photo can catch what the first one missed.
#Usage reimbursement
If you get an item graded professionally and the result doesn't match the agent's grade, email us, and we'll refund the usage charged for the Agent Grade back to your account.
Email support@agentcollectibles.com and include:
The grade AgentCollectibles returned
The exact images you submitted
A picture of the item with the professional grade to compare against
That feedback is the highest-signal data we get. It helps us see where the agent gets it wrong and what to work on next.