Opportunities › Self-hosted LLM serving and load balancing › Proposed app
Hearthline
One control panel for your self-hosted LLM and vision model servers
- Score
- 49
- confidence 45
- Evidence
- 4
- 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: Ops teams manage model fleet distribution, monitor nodes, and configure access controls—all technical work done at a desk with multiple screens. Sample screens on a computer, in the proposed brand. Their numbers are sample data.
Fleet
1 node needs attention6 of 7 up
About this appAI proposal
Hearthline is a small control plane for teams that run language and vision models on their own machines. It sits in front of ggml-based model servers, spreads requests across CPU and GPU nodes, and shows what each node and model is doing. You install one lightweight agent per machine and point it at the panel.
It is built for engineering teams at small companies up to enterprises that cannot send prompts to a cloud API for privacy, data-location or latency reasons. The material describes existing options as many moving parts to deploy and maintain. Hearthline aims for fewer pieces: one agent, one router, one dashboard.
You can run it fully on your own network, or let us host the control panel while your models stay on your hardware. This is a proposal. Whether teams will pay for it is what a first test should show.
- Tagline
- Fewer moving parts between your models and your users.
- Voice
- PlainCalmPractical
- Why this name
- A hearth is the steady fire at the centre of a home. A line carries requests to it. The name suggests your own place, kept running and warm.
In the first version
Members say which features they would use. It tells PWAS what people want; it changes no score.
One agent per machine
Install a single agent on a CPU or GPU host. It starts the ggml model server, reports health and joins the pool. No separate orchestration layer to learn.
No member yet
Load balancing across mixed hardware
Route requests by model, queue depth and node type, so a GPU box takes heavy jobs and CPU nodes take small ones.
No member yet
Language and vision models together
Serve text and image-capable models from the same pool, with one endpoint and one set of keys for your apps.
No member yet
Live fleet view
See nodes, loaded models, queue length, latency and errors on one phone-friendly screen. Installable as a PWA.
No member yet
Brand identityAI proposal
Check the name against trademark registers before using it.
Palette
#1F5C5ADeep Teal · primary#F6F1E7Oat · surface#1F2A2ESlate Ink · text#D9622BEmber · accent#3C7A9EHarbor Blue · data
Typography
Aa
Hearthline
ABCDEFGHIJKLMNOPQRSTUVWXYZ 0123456789
Fewer moving parts between your models and your users.
PricingAI proposal
The material gives no competitor prices. The tiers are a starting point to test, following the monetization idea of managed hosting, support and enterprise features. Inference as a service is not included.
Community
Free
- Up to 2 nodes
- Fleet view and basic routing
- Community support
Team
$99 a month
- Up to 10 nodes
- Rollouts with rollback
- Keys, quotas and email alerts
- Hosted control panel
Business
$399 a month
- Up to 50 nodes
- Audit log and single sign-on
- Multi-model routing rules
- Email support with response target
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. Hearthline runs inside a company's own network and serves internal apps, so its use does not expose it to new buyers. Growth comes from open documentation, guides and direct outreach to teams.
- Do they become customers?
- Nothing spreads through use. Test whether engineers who find it through guides and open-source channels move from the free tier to a paid Team plan within 30 days.
- First customers
- Platform or ML engineers at companies of 20 to 200 people that already run ggml-based models on their own servers because of privacy or data-location rules.
- Where to reach them
- Open-source communities: Answer serving and deployment questions in GitHub discussions and forums for ggml-based tools, and link a short setup guide.
- Technical guides: Publish step-by-step guides on mixing CPU and GPU nodes behind one endpoint, with working configs.
- Direct outreach: Message engineers who open issues about multi-node serving and offer a free setup call.
- First milestone
- Five teams run Hearthline on at least three nodes for 30 days, and two of them pay for the Team plan.
Why people already pay for this
- VerifiedGitHub repository gained 1675 stars in 894 days
- Public claimOpen-source serving platform addresses complexity in existing LLM deployment solutions
- Public claimMarket need exists for both CPU and GPU-based model serving
- AI inferenceRepository has substantial adoption signal with 1675 stars and 98 forks
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 cross-platform agent binaryRuns beside the ggml model server on each host, reports health and applies routing and rollout commands.
- This app · Request routerLightweight streaming HTTP proxyMust forward streamed model replies with low delay; this does not fit in serverless functions.
RisksAI inference
- CommercialTeams may stay with free open-source serving tools and see no reason to pay for a panel. The only verified signal is repository growth (1675 stars in 894 days), which is not buyer demand.
- TechnicalStreaming proxy, health checks and safe rollouts across mixed CPU and GPU hosts are hard to make reliable. A routing bug could take down a customer's production AI.
- PlatformThe product depends on the ggml ecosystem and its server interfaces, which can change quickly and may break the agent.
- CommercialDocker Model Runner and llm-d are named competitors and may add the same simple-deployment features, so the gap may close.
- DataBuyers choose self-hosting for privacy. A hosted control panel must prove that no prompts leave their network, or they will not trust it.
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.