v0.5.0: Dark mode and one visual system

Components for humans who can't let agents run unchecked.
AI agents can now send emails, execute trades, and deploy code. When actions are irreversible, oversight stops being optional.
Implement reliable human oversight today.
Oversight in action
Agents act. Humans supervise.
As AI transitions from generating text to executing irreversible actions in the real world, the primary bottleneck is no longer intelligence; it's oversight. Depute provides the UI primitives required to build safe, auditable agentic systems where humans remain in the loop.
- PlanCard
- ApprovalGate
- ConfidenceMeter
- RunControls
- ToolTrace
- ArtifactCard
- AutomationBiasAlert
- OrchestratorView
- AgentRoster
- SwarmMonitor
- DelegationGate
- TaskQueue
- +6 more
- BindingApproval
- StateDiff
- TransactionReceipt
- RollbackTimeline
- CapabilityMatrix
- +4 more
- ax-depute@latest add <component>
- Copy source into your project
- You own the code
Install one component
1. Copy the source
npx ax-depute@latest add approval-gate2. Import the tokens once
@import "./styles/depute.css";3. Render it
import { ApprovalGate } from './components/ApprovalGate';
Component showcase
ApprovalGate
Stop a high-risk action until a person approves, rejects, or stages it.
AutomationBiasAlert
Add friction when approvals start looking like rubber stamps.
BudgetMeter
Show session spend against a limit, with a visible burn rate.
ToolTrace
Read every tool call, including input, output, and failure.
DecisionRecord
Keep a receipt of who approved an action and why.
SessionOverview
Roll up what the agent touched so the blast radius stays visible.