Nexus AI Studio
Visual agentic workflow orchestration
- Platform
- Web app
- Built for
- Ops and automation teams
- What I delivered
- Canvas UX, agent debugging, approvals
- Tools
- Figma, Rive, Cursor
Executive summary
A canvas where operations teams compose, test and monitor multi agent AI workflows without writing code.
0.0x
Faster debugging
Verified post-launch
↓0%
Model spend
Verified post-launch
1,800
Teams onboarded
Verified post-launch
The problem
Agent pipelines were opaque. When an agent failed, teams couldn't tell which step, which tool call or which prompt caused it.
Research insights
- 1Trust grew when users could replay a single agent step.
- 2Ops leads wanted approval gates, not full autonomy.
- 3Cost visibility per run changed which models teams picked.
Constraints
Interactive prototype
This is the real studio running a refund agent. Choose a test ticket, press Run test, approve it when it pauses, then open Trace to see and replay each step.
{ "docs": ["refunds_v4.md", "warranty.md"] }Run the workflow to see output1 · Canvas
Pick a test ticket at the top, press Run test and watch each step light up, stream its output and add to the cost meter. Drag nodes to rearrange.
2 · Approvals
Tickets over $200 pause at the human gate. Open Approvals, read the agent's cited reasoning, then approve or reject.
3 · Trace
See every step on a timeline with latency and tokens. Select any step and replay just that one.
Solution walkthrough
Step level replay
Every node keeps inputs, tool calls and outputs, so any step can be re run in isolation.
Human in the loop gates
Drag in approval nodes pause runs and notify owners in Slack.
Live cost meter
Tokens and spend stream beside each node while a run executes.
Next case study
Prism Design System