AI Automation

Build AI automation agents that run multi-step workflows per customer. Each customer gets their own environment with persistent state, crash recovery, and isolated execution.

The problem

AI automation workflows are multi-step and long-running. Scrape data, process it, call APIs, generate reports, send emails. If step 7 of 10 fails, you don't want to restart from step 1. And each customer has different data, different configs, different state — you need isolation.

How OnCell solves it

Each customer gets their own cell with a durable execution journal. Your agent runs steps, and each successful step is recorded. If the agent crashes, it resumes from the last successful step — not from the beginning. State persists across runs.

const cell = await oncell.cells.create({
  customerId: "customer-acme",
  agent: automationAgentCode,
});

// Inside the agent — crash-proof workflow:
const data = await ctx.journal.durable("scrape", () => scrapeWebsite(url));
const processed = await ctx.journal.durable("process", () => processData(data));
const report = await ctx.journal.durable("report", () => generateReport(processed));
await ctx.journal.durable("email", () => sendEmail(report));
// If it crashes at step 3, it replays steps 1-2 from cache and continues from 3

What your agent gets

Durable execution — journal-based crash recovery. Each step is cached. Restart resumes from the last success.

Per-customer isolation— each customer's workflow runs independently. Different configs, different data, different state.

Persistent storage — store scraped data, generated reports, intermediate results. Available across runs.

Database — track workflow state, customer preferences, run history.

Streaming — send progress updates to your dashboard in real-time.

Pause / resume — workflows that run periodically. Cell sleeps between runs, wakes in 200ms.

Example architectures

Lead enrichment — agent scrapes company data, enriches with APIs, stores results. Per-customer pipeline with different sources.

Content generation — agent researches topics, generates drafts, iterates on feedback. Each client has their own content history.

Data pipeline — agent pulls from APIs, transforms data, generates reports. Crash-proof, runs nightly per customer.