Opportunities › Local large language models on mobile devices › Proposed app
Carryweight
Check if a big on-device model fits an iPhone, and log how it really runs
- Score
- 53
- confidence 46
- Evidence
- 4
- items on the page
- Rivals
- 1
- named in sources
- First version
- 25 d
- estimate
- Type
- Desktop app
- PWA: runs in the browser, installs on Mac and Windows
- Market since
- 2023
- Emerging market · AI estimate
ScreensConcept
Desktop app: Developers estimate model fit, run intensive tests on devices, and analyze detailed performance logs across builds—work best done at a desk with full dashboards and reports. Sample screens on a computer, in the proposed brand. Their numbers are sample data.
Fit estimate
Estimate, not measuredFits, tight
About this appAI proposal
Carryweight is a web app for mobile developers who want large mixture-of-experts models to run on iPhones and other Apple devices without cloud APIs. Enter a model and a target device. It estimates RAM and storage needs when expert weights are streamed from storage, and says whether the plan looks realistic.
Open the test page on a real phone and Carryweight records load time, tokens per second and memory headroom. Results go into a shared log by model, quantization and device, so a team can compare runs instead of guessing.
It does not run models itself. It sits beside whatever streaming runtime you use and gives you the numbers and a shareable report to decide what to ship.
- Tagline
- Know what your phone can carry before you build.
- Voice
- PlainPrecisePractical
- Why this name
- A model has weight, and a phone has a limit on what it can carry. The name says the product weighs one against the other.
In the first version
Members say which features they would use. It tells PWAS what people want; it changes no score.
Fit estimator
Pick a model, quantization and device. See estimated resident RAM, storage footprint and expert-streaming read load before you download anything.
No member yet
Device test page
Open a link on a real iPhone to capture storage read speed and the memory and compute details the browser exposes. No install needed.
No member yet
Run log
Record tokens per second, first-token delay and crashes per model build and device. Filter and compare runs across your team.
No member yet
Runtime-agnostic imports
Paste or upload benchmark output from your runtime. Fields map to a common format, so different tools can be compared.
No member yet
Brand identityAI proposal
Check the name against trademark registers before using it.
Palette
#12332EDeep Teal · primary#F6F1E7Paper · surface#1B2422Ink · text#F2B544Amber · accent#4F7F6BMoss · data
Typography
Aa
Carryweight
ABCDEFGHIJKLMNOPQRSTUVWXYZ 0123456789
Know what your phone can carry before you build.
PricingAI proposal
The material gives no competitor prices (Qwen's are not stated). These tiers are a starting point to test with developers.
Free
Free
- Fit estimator
- 1 project
- 20 logged runs
Developer
$19 a month
- Unlimited runs
- Shareable reports
- Regression alerts
- Import from any runtime
Team
$79 a month
- Up to 10 seats
- Shared device library
- Cloud cost comparison
- Priority support
Would you pay Developer · $19 a month?
Your answer tells PWAS which prices people accept. It changes no score.
No answers yet.
What it could be worthEstimate
What a software business making this much sells for today, from real deals. Pick a monthly revenue.
If it makes
Worth about
$194,000
Sale range $132,000 to $263,000
$120,000 a year in revenue
From 87 software businesses listed and sold on Empire Flippers in the last year, revenue vetted by the broker: they sell for about 19× their monthly revenue, most between 13× and 26×.
Measured on profit, the middle deal is 32× monthly net profit: about 2.7 years of profit.
An estimate, not an offer or financial advice. A real price depends on profit, growth, churn, how much of the work is the owner's, and the buyer.
How it would growAI proposal
How the product would reach new customers through its own use, and how the first ones would be found. A plan to test, not a result.
- Built-in loop
- Shared outputs. Developers publish fit reports and run comparisons by link to clients, product owners and open-source readers. Each report shows the Carryweight name and the measured numbers behind a decision.
- Do they become customers?
- Report readers are often product managers or other developers who face the same question and could sign up. Test whether readers start their own projects, not only whether they open the link.
- First customers
- Independent iOS developers and small app studios of 1 to 10 people who are trying to add a local large language model to an existing iPhone app.
- Where to reach them
- Open-source repo communities: Offer the fit estimator in issues and discussions of on-device inference projects, where people ask whether a model will fit.
- iOS developer forums: Post measured fit tables for popular models on common iPhones, with a link to the free tool.
- First milestone
- Ten developers log at least 20 real device runs each, and five of them pay for the Developer tier within 60 days.
Why people already pay for this
- VerifiedGitHub repository gained 658 stars in 70 days
- Public claimLarge Qwen Mixture-of-Experts models can run locally on Apple devices including iPhone through on-device weight streaming
- AI inferenceExisting cloud-based LLM inference services create demand for on-device alternatives
- VerifiedHacker News post received 312 points and 141 comments
Tech stack
Every proposal builds on the same PWA base, installable from the browser on any phone or computer. Dashed boxes are what this app adds.
- AppNext.js and TypeScript
- PWAManifest and service worker
- UITailwind CSS
- Data and authSupabase Postgres
- PaymentsStripe
- EmailResend
- HostingVercel
- This app · Model metadataCurated table of model sizes and expert layoutsThe fit estimator needs parameter counts, expert sizes and quantization data, kept up to date by hand or script.
- This app · File storageObject storage for uploaded benchmark logsTeams upload raw runtime output that is parsed and kept for comparison.
RisksAI inference
- TechnicalA browser cannot see exact RAM, thermal state or Metal details on iOS, so the test page gives limited numbers. Real accuracy may need data imported from a native runtime.
- CommercialDevelopers may use a free spreadsheet or the runtime's own benchmarks. The material shows interest in the runtime (658 stars in 70 days), not in a paid tool beside it.
- DataFit estimates depend on model layouts and runtime behaviour that change fast. Stale numbers would erode trust quickly.
- PlatformSafari limits storage, background work and memory for web pages, which restricts what the test page can measure on iPhone.
Everything marked AI proposal or AI inference was written by a model from the evidence on the opportunity page. The proposed name must be checked against trademark registers before it is used.