Opportunities › Local model deployment for inference cost reduction › Proposed app
Stillrun
Test a local model against your paid API, then move traffic over safely
- Score
- 53
- confidence 56
- Evidence
- 6
- items on the page
- Rivals
- 2
- named in sources
- First version
- 45 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: Technical leaders configure model serving, compare API responses side by side, and track migration metrics through control panels. Sample screens on a computer, in the proposed brand. Their numbers are sample data.
Hardware fit
Ready to serve18.7 tok/s
About this appAI proposal
Stillrun is a control panel for teams that want to move inference from paid model APIs to open models running on their own hardware. It scans your machines, suggests a model that fits, and sets up a serving endpoint that accepts the requests your apps already send.
Before any traffic moves, Stillrun replays a sample of your real prompts against both the paid API and the local model, then scores the answers side by side on your own checks: extracted fields, defect flags, format rules. You move one workload at a time and can roll back at once.
It is for CTOs and small engineering teams with capable hardware and a growing per-token bill, who have seen local models do well in tests but have not yet run them in production.
- Tagline
- Prove it locally. Then switch the bill off.
- Voice
- CalmPrecisePractical
- Why this name
- A quiet server that keeps running: the name suggests steady, owned hardware doing the work without a meter running.
In the first version
Members say which features they would use. It tells PWAS what people want; it changes no score.
Hardware scan and model fit
Reads GPU, memory and disk on each machine and suggests open models and quantization levels that fit, with expected tokens per second.
No member yet
Side-by-side replay
Replays sampled prompts against your paid API and the local model and shows both answers next to each other.
No member yet
Your own pass/fail checks
Define checks such as required fields, planted defects or exact formats. Get a match rate per workload before you switch.
No member yet
Drop-in endpoint
Serves local models behind an API shape your apps already use, so a switch is a base URL and key change.
No member yet
Brand identityAI proposal
Check the name against trademark registers before using it.
Palette
#1F6F5BPine · primary#F4F1EAOat · surface#1F2A24Charcoal Green · text#D9822BCopper · accent#3B5B8CSlate Blue · data
Typography
Aa
Stillrun
ABCDEFGHIJKLMNOPQRSTUVWXYZ 0123456789
Prove it locally. Then switch the bill off.
PricingAI proposal
The material gives no competitor prices, only that frontier APIs bill per token and local models cost $0 per token. These tiers are a starting point to test, priced by nodes, not tokens.
Trial
Free
- 1 node
- 1 workload
- 200 replay prompts a month
- Savings ledger
Team
$299 a month
- Up to 3 nodes
- 10 workloads
- Unlimited replays
- Traffic shifting and fallback
- Email alerts
Scale
$999 a month
- Up to 15 nodes
- Unlimited workloads
- Audit export
- Priority support
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
- No loop in the product. Stillrun runs inside a company's own infrastructure and is not seen by outsiders. Customers are reached through content, community posts and direct outreach, plus shareable replay reports used internally.
- Do they become customers?
- Replay reports are shared with colleagues and finance, who may become users, but this is untested. Test whether a report shared internally brings a second team into a paid plan.
- First customers
- CTOs and lead engineers at companies of 20 to 200 people that already own high-spec GPU machines and spend heavily on a frontier model API for extraction or classification.
- Where to reach them
- Hacker News and Show HN: Publish a worked replay of a local model on a real extraction task and invite teams to try their own.
- Self-hosting AI communities: Join forums and chat groups where teams discuss running open models, and offer free trial replays.
- Direct outreach to CTOs: Message engineering leaders who mention rising inference bills and offer a one-workload replay.
- First milestone
- Five teams complete a replay on their own prompts, and at least two pay to shift one workload within 60 days.
Why people already pay for this
- AI inferenceCompanies currently pay per-token costs to frontier model APIs even when local models can match the quality
- Public claimLocal open-source models now match frontier model performance on business-critical tasks
- Public claimOperational complexity delays migration even when local models prove equivalent
- Public claimSignificant portion of current inference spend could be shifted to local models
- AI inferenceLocal model performance on real-world business tests matches frontier models
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 · Node agentSmall agent installed on customer machinesScans hardware, runs the model server and reports health to the dashboard without exposing the machine.
- This app · Model servingOpen-source inference serverRuns open models on local GPUs behind an API compatible with common client libraries.
RisksAI inference
- TechnicalSupporting varied GPUs, drivers and operating systems is hard. The first version may need to target one or two hardware setups.
- CommercialTeams may use free open-source serving tools directly, and the value rests on replay and checks being worth paying for.
- DataReplaying real prompts sends customer data to a paid API a second time, so teams need control over sampling and redaction.
- AIParity seen in a few tests may not hold on a team's own tasks. Checks can pass on samples yet miss rare failures.
- PlatformPaid API providers may change terms on comparing or reusing their outputs, or cut prices, which would shrink the savings.
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.