Opportunities › AI model guardrails and prompt protection › Proposed app
Kerbline
One set of AI safety rules that follows your prompts to any hosted model
- Score
- 36
- confidence 55
- Evidence
- 8
- items on the page
- Rivals
- 2
- named in sources
- First version
- 40 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: Writing and testing AI safety rules, auditing decisions, and managing policies demand careful work at a desk. Sample screens on a computer, in the proposed brand. Their numbers are sample data.
Today
All rules active128 blocked
About this appAI proposal
Kerbline sits between your apps and the AI models you buy. You write your safety rules once: block prompt injection, hide private data, check answers before they reach users. Kerbline applies them to every request, whichever provider answers.
It is built for enterprise teams that use hosted models and do not self-host. Switching providers does not mean rebuilding your protection. Rules, test cases and logs stay with you, not with a model vendor.
Instead of stitching together many separate guardrail projects, you get one policy editor, one test bench and one audit trail. Every rule change can be tried against past traffic before it goes live.
- Tagline
- Write your AI safety rules once. Keep them when the model changes.
- Voice
- PlainCalmPrecise
- Why this name
- A kerb keeps traffic on the road without choosing the vehicle. Kerbline keeps requests in bounds whichever model drives. Short, pronounceable.
In the first version
Members say which features they would use. It tells PWAS what people want; it changes no score.
Write rules once, use on any model
Define input and output policies a single time and route them to any hosted model endpoint, so a provider switch keeps the same protection.
No member yet
Prompt attack screening
Checks incoming prompts for injection and jailbreak patterns and lets you choose to block, rewrite or flag each one.
No member yet
Output checks before delivery
Scan model answers for unsafe content, leaked secrets and personal data, then hold, redact or pass them on.
No member yet
Replay before you ship a rule
Run a new or changed rule against saved past requests and see what it would have blocked before it goes live.
No member yet
Brand identityAI proposal
Check the name against trademark registers before using it.
Palette
#0F5C55Deep Teal · primary#F4F1EAPaper · surface#1C2B2AInk Green · text#E0742BSignal Orange · accent#3B6FA0Slate Blue · data#B3362DBrick · alert
Typography
Aa
Kerbline
PricingAI proposal
The material gives no guardrail competitor prices. Fireworks.ai's verified page shows usage-based plans from $0.10 to $3, so checks are metered by volume. These tiers are a starting point to test.
Starter
Free
- 1 app
- 1 model connection
- 10,000 checks a month
- Basic rule library
- 7-day log
Team
$49 a month
- 5 apps
- 3 model connections
- 200,000 checks a month
- Replay testing
- 90-day log
Business
$249 a month
- Unlimited apps
- 10 model connections
- 2 million checks a month
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
- Partners and integrations. List Kerbline as a drop-in gateway for hosted model platforms and AI frameworks, with ready-made connectors and shareable rule packs that teams import from guides and repositories.
- Do they become customers?
- Engineers who find it via a connector are often the buyers' teams, but budget sits with security leads. Test whether a developer trial turns into a paid Team plan within 30 days.
- First customers
- Platform or security leads at mid-size software companies (200 to 2,000 staff) that ship an AI feature on hosted models and have no self-hosted setup.
- Where to reach them
- Hacker News and dev forums: Publish an open rule pack and a write-up on keeping safety rules across providers; invite readers to a trial.
- Direct outreach on LinkedIn: Message platform and security leads at companies with public AI features; offer a free replay audit of their prompts.
- AI engineering meetups: Give short demos of switching providers while keeping the same rules, then collect sign-ups.
- First milestone
- Five paying Team-plan companies in one vertical within 90 days, each routing live traffic through Kerbline.
Why people already pay for this
- Public claimMost enterprises are not self-hosting AI models due to expensive hardware costs
- Public claimThere are hundreds of different guardrail projects and approaches in the market
- Public claimEnterprises have a need for model choice and want guardrails independent of vendor
- Public claimPrompt protection and guardrails are a central concern in AI deployment
- Verified
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 · Request proxyEdge proxy that forwards model callsRules must run in the request path with low delay, in front of any provider.
- This app · Content checksPattern rules plus a small classifier serviceDetects injection, personal data and unsafe output beyond simple matching.
RisksAI inference
- TechnicalSitting in the request path adds delay and a new point of failure. Customers may reject it unless it is fast and can fail open or closed by their choice.
- AIDetection of injection and unsafe output is never complete. False alarms annoy users and misses damage trust, so claims must stay modest and measured.
- CommercialEvidence shows hundreds of guardrail projects exist. Buyers may assemble free ones, or model providers may bundle their own controls.
- DataPrompts and answers can hold sensitive data. Enterprises will ask where logs live and how long, which raises security review effort.
- PlatformProviders may change API formats or terms, and routing traffic through a third party may need their approval or break features like streaming.
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.