One agent operating system for every agent you run.

A job is a prompt, a trigger and an outcome contract: what done looks like, how to check it, and when to stop. Jobs run once, on an interval, on a cron schedule, by hand, after another job succeeds, when an email arrives, or when a webhook fires.

• Schedules, webhooks and inbound email
• Chains: run this after that succeeds
• Priority levels, so urgent work goes first
• Budgets that stop runaway spend

Agents read and write tables: the keyword plan, the content log, incidents, funnel steps. Connectors fill tables from Search Console, GA4 and more, and charts update as rows arrive.

• Tables agents and people both edit
• Connector-fed tables that refresh themselves
• Charts on any table
• A change history on every row

Risky actions become tasks with Approve and Reject on them. Pull requests, deploys, published posts and held emails land in one review queue, so you can clear a morning’s work in a few minutes.

• Autonomy levels you choose
• One queue for everything waiting on you
• Edit before you approve
• Nothing runs for the first time without your yes

Each agent works towards a few criteria a machine can check. After every run, a separate evaluator reads the record of what the agent actually did and decides whether the work moved forward. The agent’s own summary is never taken on trust.

• Machine-checkable goal criteria
• An evaluator that is not the worker
• Failures that name the external fix they need
• Progress you can read in Pulse

With the Contextely plugin, agents remember what they learned about each product. Chat builds agents, jobs and tables from a plain description, and every job it creates carries an outcome contract. Templates provision a whole crew in one step.

• Product-scoped memory
• Chat that creates real agents and jobs
• Templates you can change freely
• Plugins for mail, connections and memory