One run, start to finish
Follow each run from the first step through approvals to the result in its Activity tab.
Features
A fleet of agents on your schedule
Describe the goal. Subagents split the work, run side by side, and report back inside the limits you set.
Built to run in parallel
One goal. Subagents side by side.
A run hands parts of the work to subagents, runs them at the same time, and brings back one result.
Parallel subagents
A run can hand work to several subagents that work at the same time.
Runs on a schedule
Set a cron or interval schedule in your time zone, or start a run by hand.
You set the limits
Pick the tools and which actions need approval. Set spend, time, and step limits for each run.
Cost on every run
Each run shows its cost, tokens, and time, split by model, subagent, and tool.
Works with your stack
Everything a run needs
Connected accounts
Agents use the tools your team connects, with more than one account per tool.
Model providers
Pick a model per chat or set team defaults. Bring your own keys if you want.
Templates
Start from a template and change its steps, tools, and schedule.
Action permissions
Allow, deny, or require approval for each action. Runs wait until a person decides.
Run activity
See what is running, waiting for approval, paused at a limit, or failed.
Usage and cost
See spend by model and export it as CSV.
Agent factory questions, answered
How workflows run, who stays in control, and what each run costs.
Want to design your first agent workflow? Talk to us
1What is an AI Agent Factory?
A place to turn a goal into a workflow that runs again and again, with the tools, schedule, approvals, and limits you set.
2Can we start with one workflow?
Yes. Start from one template or one goal, read the first results, then add more.
3Can agents use our existing tools?
Yes. Agents use the accounts your team connects. Actions that change data wait for approval.
4How do humans stay in control?
You choose which actions are allowed, denied, or need approval. A run waits until someone approves.
5Can different jobs use different models?
Yes. Pick a model per chat, set team defaults, and choose the harness per team, project, or chat.
6How are failed runs handled?
A failed run keeps its activity and steps, so you can see what happened and run the task again.
7How does a successful run become reusable?
The workflow keeps its steps, tools, and schedule, so the next run follows the same plan.
8How do we keep costs in check?
Each run shows its cost. Usage groups spend by model, and a run pauses when it reaches its spend limit.
Put your first agents to work
Pick a template or describe a goal, set the limits, and start the first run.