AI business ideas you can build with llama.cpp
llama.cpp: Free, open source (MIT) (opens in new tab). Ranked by traction (X, YouTube, Hacker News).
Ideas
Filter ideas
Showing all 4 ideas.
- 1—
Sell local-AI hardware buyers reports on how fast a model will run before they buy
Llmspeed lets people feel how fast a given local language model would generate text on a given machine, before spending money on a GPU or Mac.
Free speed tool; charge about $9 for a buyer’s report comparing hardware with API costs (affiliate links labeled).
Cost $30–$105/monthCould earn $1,800/mo estimateAgent setup Advanced77 traction𝕏 228 X posts/7dY 331 HN pts
- 2—
Sell a private offline transcription and meeting notes app
A phone app that transcribes recordings on the device, separates speakers and turns meetings and lectures into notes, with nothing sent to the cloud.
Charge app users $7/mo for private, offline transcription and meeting notes on their phone.
Cost $30–$90/monthRevenue $1,295/mo self-reported in an online communityAgent setup Advanced41 traction𝕏 96 X posts/7d
- 3—
Sell a private budgeting app whose AI runs on the user’s own computer
Pulse Local is a Mac budgeting app that answers questions about your spending with small models running on your machine, with no account, analytics or crash logs.
Charge privacy-minded people about $4/mo (or a license) for a budgeting app whose AI never leaves their computer.
Cost $30–$80/monthCould earn $800/mo estimateAgent setup Advanced34 traction𝕏 13 X posts/7dY 122 HN pts
- 4▼7
Sell AI transaction categorization as an add-on for self-hosted budget apps
Users of open-source budget apps such as Actual and Firefly III keep asking for AI that categorizes imported bank transactions and answers questions about their data.
Charge self-hosted budget app users about $5/mo for AI categorization that learns from their corrections.
Cost $30–$85/monthCould earn $1,500/mo estimateAgent setup Advanced8 traction𝕏 8 X posts/7d
Rank orders ideas by Traction, highest first. Traction adds up public interest we measure: X posts and engagement over 7 days at full weight, the source YouTube video's views and likes at 0.6, the source Hacker News post's points and comments at 0.8, and how often the topic comes up on Hacker News over the past year at 0.3. Each number is log-scaled, so one viral post can't dominate. ▲3 / ▼3 means places moved since yesterday, and NEW means added in the last 7 days. Full formula.
How the columns work. Cost: monthly tool cost range, sorted by the top of the range. Earning: monthly revenue a source reported, with its label (verified, third-party, self-reported, or self-reported in an online community), otherwise our estimate (customers × price). Agent setup: our Easy / Intermediate / Advanced rubric (how we rate it). Ties keep rank order.
Last updated · How we rank and label numbers
The Weekly top 5.
Get the Weekly top 5 + free download: Top 25 AI businesses with verified revenue (PDF + CSV; revenue verified by TrustMRR via Stripe or RevenueCat). Free, every Friday.
Your free download: Top 25 list (PDF) · Spreadsheet (CSV)