Changelog
Everything that shipped, newest first. One entry per release, describing what you can do that you could not do before.
Paste one URL, and the workflow listens before it runs
A trigger can now be a webhook URL you paste into another service, with a listen window that captures a real delivery into the payload library before you publish.
Run a workflow when an email arrives
Gmail, iCloud, GMX and WEB.DE mailboxes can start a workflow. You hand over the address and an app password, and the platform reads the mailbox itself.
Public docs you can read, search and ask questions of
Documentation lives at /docs, with search, an Ask AI box, and a plain markdown copy of every page for agents.
See what you have connected, and disconnect it
Connections get a real list with a Disconnect button, any system can be named, and third-party browser logins now finish properly.
The interview stops guessing what the platform can do
The agent that designs your workflow reads the published capability docs before it answers, and tells you when it could not.
Set a spending cap, and unattended work parks instead of running up a bill
Monthly caps for your workspace, for you and for any single workflow, with a ledger behind them and a notification when one bites.
Files your agent downloads come back to you
A run can collect a file from a browser download and attach it to the run, with caps and an honest stub when a file is too large.
See everything a run did, in one place
One timeline replaces the split feed and step trace, with full tool calls, durations, and the question each gate asked.
Schedules run in your timezone
A schedule carries its own zone, shows its next run in local time, and keeps firing once a day across both daylight saving edges.
Stop a run, and correct a checkpoint instead of ending the run
Cancel one run or every run of a workflow at once, and say no at a gate without throwing away the work behind it.
See what a run cost, in dollars
Each run carries an estimated cost beside its model, priced per model with the cached share of the input billed at the cached rate.
Fire a run yourself, for real or as a rehearsal
A play button runs the published workflow for real, and a test fire carries your own note and payload with your authority behind it.
Choose the model a workflow runs on, and pay with your own key
Set the model and reasoning effort per workflow, and route every agent of yours through your own OpenAI key.
Watch a recording of what the browser did
Every browser session a run opens is recorded, stored on our side, and playable from the run page days afterwards.
Read the exact conversation a run had
The run page serves every turn, tool call and thinking step verbatim, instead of the clipped preview the live feed shows.
Everything a workflow is, in one version history
The written procedure, the steps and your reference files move together under one counter, and a change asks whether to adapt or reset.
Your agents know what day it is
Every agent surface opens with the current UTC time, and a scheduled run is told which occurrence woke it.
One agent watches the run, and a failure opens a conversation to fix it
Codified steps run with no model in the loop until something needs judgment, and a failed run opens a repair interview.
See what every automated step is allowed to reach
Each code step declares its network access and its connections, and the console prints the complete list beside your steps.
Your agent writes the integration for the system you actually use
The interview authors sandboxed code that subscribes to your service's events, so a trigger no longer waits on our backlog.
Workflows that fire when the thing happens, and a Pause button
Event triggers stay armed and repair themselves, Pause turns a workflow off without deleting it, and a pinned payload makes an event test real.
Watch what the agent is doing while it does it
Tool calls now carry their arguments and results, the model's reasoning summaries are shown, and every list says what is busy.
Test run does exactly what you are looking at
Fire the draft or the published version on purpose, get the refusal at the click, and see a trigger that failed before a run existed.
Close the tab and your run keeps going
Agent turns run detached from the browser request now, on a durable log you can reload straight back into.