Opportunities › Security controls for autonomous agent actions › Proposed app
Gatewick
A policy firewall that checks every agent action before it reaches production
- Score
- 38
- confidence 47
- Evidence
- 4
- items on the page
- Rivals
- 3
- named in sources
- First version
- 60 d
- estimate
- Type
- Desktop and mobile
- PWA: on computers and phones alike
- Market since
- 2023
- Emerging market · AI estimate
ScreensConcept
Desktop and mobile: DevOps teams configure policies and rules at their desk, but must approve urgent agent actions from anywhere including their phone during incidents. Sample screens on a computer and a phone, in the proposed brand. Their numbers are sample data.
Pending approval
Destructive1 waiting
About this appAI proposal
Gatewick sits between your AI agents and your production systems. It reads requests at the application protocol level, such as SQL or Kubernetes API calls, and applies your rules: allow, deny, hold for a human, or send to a second model for review.
It is for DevOps and infrastructure teams who let agents diagnose and fix incidents, but do not want to choose between read-only access and approving every step by hand. Rules apply per cluster or environment, and every decision is logged.
Unlike general LLM gateways or MCP proxies, it is built around low-level protocols and awkward real setups, such as a database reached through a tunnel inside Kubernetes. The mobile-friendly console lets on-call staff approve or reject actions from a phone.
- Tagline
- Let agents fix production. Keep your hand on the gate.
- Voice
- CalmPreciseDirect
- Why this name
- Gate plus wick: a gate that controls passage and a short fuse you can cut before harm. Two plain syllable groups, easy to say on an incident call.
In the first version
Members say which features they would use. It tells PWAS what people want; it changes no score.
Protocol-aware rules
Write rules on what a request does, such as a DELETE without a WHERE clause or a pod exec, not just on which host or tool it calls.
No member yet
Human approval queue
Destructive actions pause and go to on-call staff. Approve or reject from your phone with the full request shown.
No member yet
Second-model review
Route risky actions to a separate LLM reviewer first. Its verdict and reasoning are stored beside the request.
No member yet
Tunnel handling
Inspect traffic carried through Kubernetes port-forwards or exec sessions, so a tunneled database is policed like a direct one.
No member yet
Brand identityAI proposal
Check the name against trademark registers before using it.
Palette
#14232BDeep Slate · primary#F4F1EAPaper · surface#1B2428Ink · text#E08A1ESignal Amber · accent#2F7D5BGate Green · data: allowed#B3382CStop Red · data: denied
Typography
Aa
Gatewick
PricingAI proposal
The material gives no competitor prices. Tiers follow the suggested per-cluster or environment model and are starting points to test, not evidence-based figures.
Trial
Free
- 1 environment
- 2 agents
- Allow and deny rules
- 7-day audit log
Team
$199 a month
- Up to 3 environments
- 10 agents
- Approval queue on mobile
- LLM review
- 90-day audit log
Business
$599 a month
- Up to 10 environments
- 50 agents
- Dry-run replay
- 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. Gatewick sits inside a company's infrastructure and is not visible to outsiders. Reach comes from sales, open policy examples, technical writing and talks, not from product use.
- Do they become customers?
- Approvers added to the queue are colleagues already in the team, so invites could help seat growth, but not new customers. Test whether published policy templates bring inbound trials.
- First customers
- Platform and SRE teams of 5 to 30 engineers at companies already running AI incident-response agents against Kubernetes and Postgres in production.
- Where to reach them
- Hacker News and dev forums: Share a write-up on policing a tunneled database for agents, plus a free sample policy pack, and talk with commenters.
- SRE and Kubernetes communities: Offer hands-on pilot setups in community chats and meetups to teams already trying incident agents.
- Direct outreach: Contact platform leads who have written about agent incidents or access, offering a free dry-run on their logs.
- First milestone
- Five teams run Gatewick in front of a production environment, and at least three pay for the Team tier within 90 days.
Why people already pay for this
- VerifiedHacker News post received 112 points and 31 comments
- Public claimAI agents are being deployed in production to autonomously diagnose and fix problems with access to critical systems.
- Public claimExisting security solutions for agent access control lack the capability to handle low-level protocol controls and complex real-world scenarios.
- Public claimOrganizations need to control agent actions at the protocol level, require human and LLM review of destructive operations, and maintain audit logs.
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 · Proxy runtimeLong-running network proxy service (Go or Rust)Protocol parsing and tunnel handling need persistent TCP connections, which Vercel functions cannot hold.
- This app · Protocol parsersSQL parser and Kubernetes API/exec stream decodersRules must read what a request does, not just where it goes.
RisksAI inference
- TechnicalParsing Postgres and Kubernetes protocols, including tunneled streams, is hard to get right. A parser gap could let a harmful action through, or block legitimate fixes.
- CommercialTeams may feel a read-only default or existing gateways are enough, or build a proxy in-house. Willingness to pay per environment is untested.
- PlatformAgent frameworks and cloud vendors may add native action controls, shrinking the gap this product fills.
- AIA reviewing LLM can be fooled by crafted input or be wrong. It must support, not replace, hard rules and human approval.
- DataThe proxy sees sensitive production traffic and queries. Storing it in audit logs needs redaction, access control and careful retention.
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.