EdgeKit

PlatformManaged

The agentic reader revenue platform

A team of agents that acquire, retain and engage readers, toward the goal you set, inside the guardrails you write, measured against your own rules.

Managed today: EdgeKit’s team runs each agent’s loop with you. You approve and deploy every change, and automation takes over each step as it ships.The managed service

AgentsIllustrative

Today

Testing one new offer takes nine steps. A person does eight of them.

Counted on EdgeKit as it ships today, not on somebody else’s stack. With the managed agents, your team writes the goal and approves; EdgeKit does the rest.

  1. 01Write the offerToday · YouManaged · You, or we propose it
  2. 02Write the rule that shows itToday · YouManaged · We propose it
  3. 03Set up the test and its segmentToday · YouManaged · We do
  4. 04DeployToday · YouManaged · You: approving is the deploy
  5. 05Wait for trafficToday · YouManaged · We move traffic toward the goal
  6. 06Copy the events to your warehouseToday · A scheduleManaged · A schedule
  7. 07Write the SQL that reads the resultToday · YouManaged · We do, against the holdout
  8. 08Pick the winnerToday · YouManaged · We do, and show you why
  9. 09Edit the rule and deploy againToday · YouManaged · We propose; you deploy

From rules to goals

Stop writing a rule for every reader. Write the goal.

Your rules don’t go away. They become the holdout every agent is measured against, and the guardrails it can’t cross.

Rules

  • Your team writes a rule for every case it can think of
  • One test at a time, read in your own SQL
  • Everyone a rule matches sees the same wall

Goals

  • Your team writes the goal and the limits
  • Every arm tested at once, measured against your holdout
  • Traffic moves to the arms that earn your goal, segment by segment

How an agent works

Goal, guardrails, test, propose, deploy, measure

One loop of the Acquisition agent. Traffic moves between the arms you deployed, and anything new waits for you.

Acquisition agent · one loopIllustrative
Goal
Yield per 1,000 readers
Guardrails
Price floor £4Discount ≤ 50%Page views ≥ 3 / readerExploration ≤ 20%

Share of traffic

  • Meter of 5, then offer30%
  • Registration wall at 330%
  • Trial offer at 230%
  • Holdout · your current rules10%
ProposalAdd an annual offer at the trial pointWaiting for your deploy

You set the goal: yield per 1,000 readers.

Control

Nothing moves outside what you deployed

Nothing new goes live without your deploy

Traffic moves only between the arms you approved. A new offer, price or rule is a proposal in your console, and approving it is a deploy, in the same audit log as every change your team makes.

A permanent holdout

Every agent runs beside a holdout on your current rules, so the lift it claims is measured on your traffic, against your baseline, never against a benchmark.

A stop-loss you set

If the goal falls further below the holdout than you allow, traffic goes back to your rules. Today EdgeKit’s team watches for it and moves it back; a stop-loss that trips on its own is coming soon.

What it never does, in writing

Each agent’s page lists what it will never do: no price below your floor, no wall in front of a subscriber, no message without consent.

Measured on your readers only

Each organization’s results come from its own readers, in its own stack. Never pooled, never resold.

Every proposal shows its evidence

A proposal arrives with the data behind it, and our analysts answer “why” with the query attached.

Data

Decided at the edge, fed by your warehouse

Every decision is made inside the request, on the same reader record as your paywall, so the choice is made before the page renders. Your warehouse feeds it: scores your data team computes arrive as facts, and every event and decision goes back out.

  • Decides in the request, at the edge
  • Your own propensity and churn scores, read by every rule
  • Every decision lands in Snowflake or BigQuery
Your score, end to end

Beyond the four agents

Your journalism in your readers’ AI assistants

AI Access & Answers

Your journalism inside your readers’ AI assistants

Readers connect Claude, ChatGPT or any MCP assistant to your site and ask questions answered from your reporting, with every article behind the same rules as your paywall.

Explore AI Access & Answers

Beside, not instead

Agents decide. The tools you already run deliver.

Your email platform sends

Braze, SendGrid or Resend delivers every message. The agents decide which message each segment gets.

Your CMS stays

Your editors keep their tools and their pinned slots. Agents fill only the slots you hand them.

Your billing provider stays

Stripe or Recurly keeps charging. Agents pick offers and prices inside your floors; your billing provider bills.

Platform

See the agents on your own readers

Book a demo