Sourcerer: the model reads the thrift find, plain code decides the price

Description and use

Sourcerer enables offline pricing of thrift items using a local Gemma model and simple arithmetic, without internet or API keys.

Problem

Resellers need a quick, offline way to determine the maximum price they should pay for a thrifted item, accounting for fees, shipping, supplies, and desired margin, without relying on internet or API keys.

Outcome

Sourcerer enables offline, instant pricing for thrift items, reducing decision time and avoiding costly mistakes, with privacy guarantees and no external API calls.

Workflow

• 1. Photograph or type the item description.
• 2. Local Gemma model (gemma3:4b) reads the image or text.
• 3. Packages/shared calculates eBay fees, shipping, supplies, and desired margin.
• 4. Verdict card displays MAX OFFER, EST. RESALE, and profit.
• 5. User can edit asking price and see instant re‑pricing.
• 6. App logs provider execution and privacy status.

Evidence

Photograph an item, or type what you see; a local open‑weight model reads it, and plain TypeScript decides the most you should pay — after eBay's fees, shipping, supplies and the margin you actually want.

Source

No internet, no API keys, no per‑scan cost.

Source

The app still boots from the service‑worker cache and prices from her own sold history, tagged estimate so it can never pass for a live comp.

Source

Hardware, for every timing claim in this post: 8-core CPU, 15.4 GB RAM, an NVIDIA GTX 1650 with 3.6 GB of VRAM — enough to hold part of the model, not all of it, so most inference runs on CPU.

Source