Docs
How Reasonlayer works — building agentic workflows, running them in production, and keeping a human in the loop where it counts.
Getting started
- 01
Describe the workflow
Chat with AI in our web app, or drive it from your own desktop agent like Claude Cowork or ChatGPT Desktop. Say what you want in plain English — the agent asks for the details it's missing.
- 02
Review the plan
We turn the conversation into a visual representation of the workflow: the steps, the tools it touches, and where a human needs to weigh in. You edit it before anything runs.
- 03
Connect your tools
Authorize the integrations the workflow needs. Every connection is scoped — the agent gets the access the workflow requires and nothing beyond it.
- 04
Run it
Fire it from a trigger or put it on a schedule. Runs are logged end to end, so you can audit what the agent did, what it decided, and what it cost.
Core concepts
Sandboxed agents
Every agent runs in a cloud sandbox that behaves like a real computer — files, a browser, a shell. The sandbox is the security boundary, so an agent can work like a person without the blast radius of one.
Human in the loop
Pause a run for approval, route an output to the right reviewer, and escalate when something is out of bounds. Review interfaces are built for the task, not a generic approve/reject button.
Permissions and access control
Invite your team with different levels of access. A reviewer can review work without getting access to the rest of the system, and every agent action is bound by the permissions you grant it.
Models and cost
Use any model, per step. Cost and latency are tracked per run so you can keep tuning which model does which part of the work.
Integrations
Hundreds of connectors for the apps your operations already run on, plus a browser and filesystem for the ones without an API worth using.
Auditability
Non-deterministic output needs a paper trail. Every run keeps its inputs, reasoning, tool calls, and human decisions so you can answer what happened and why.
Keep reading
- FAQ — the questions we get asked most.
- Forward deployed — have our team build the first workflows with you.
- Blog — notes on running agents in production.
Need something more specific?
Full API and SDK reference is on the way. Until then, tell us what you're building and we'll walk you through it.