Execution · trust
Every run leaves a record you can check.
Delegation needs proof. Chainabit keeps the plan, the steps, the tool calls and the outputs of every agent run — and waits for your approval where you told it to.
The problem
You cannot delegate to something you cannot audit.
Assistants that act without a trail produce results nobody can verify and that change between runs. Teams respond by not delegating at all.
How it works
Describe it once. Review the result.
The plan is explicit
Before executing, an agent lays out the steps it intends to take.
Sensitive steps wait
Actions that write to external tools or spend credits can require an approval, per policy.
The run is recorded
Each step, tool call, input and output is stored with the run.
Inspect any result
Open a run to see how the output was produced, retry a step, or send feedback.
Capabilities
Execution history
Approval gates
A pending approval pauses the run and shows exactly what will happen if you say yes.
Run history
A searchable list of every agent run with status, duration and outcome.
Structured outputs
Results are typed — tasks, documents, messages, artifacts — so they can be reused and measured.
Policies
Set what an agent or twin may do on its own and what requires you.
Questions
Yes, per session or per agent. Auto mode runs without approvals; policies still apply to sensitive actions.
The run records the failure and its cause. You can retry the step, adjust the input, or let the agent recover where it can.
Run history stays with the workspace. Organization plans add usage reporting across members.
In an isolated sandbox with its own filesystem, separate from your workspace data.
Yes. Approval gates and policies can pause sensitive actions until you review them, while lower-risk work can continue within the boundaries you set.
Start with one recurring task.
Create a free account, describe one thing you do every week, and let Chao take it from there.