Use Case

A copilot for anything

Embed Tactus agents inside your app: a chat UI, real tools, and explicit human checkpoints when the agent is about to do something that matters.

This walkthrough focuses on two Human-in-the-Loop interactions: an approval before side effects, and a structured inputs form when the agent needs missing details.

Acme Copilot
Chat UI + Tactus runtime + HITL components

I can help you draft responses, look up context via tools, and propose actions. You stay in control of side effects.

Draft a short reply to the customer and send it.

Here’s a draft. Before I send anything, I’ll ask for an explicit approval.

Acme CopilotAcme Copilot42s ago
Send this reply to the customer? (Tool: send_email)

Done. Next, I can open a replacement ticket — I just need two details.

Acme CopilotAcme Copilot42s ago
Create a replacement ticket (Tool: create_ticket)
Acme CopilotAcme Copilot42s ago
What is the order number?
These are the same components used by the Tactus IDE and documented in Resources → Components.

What this unlocks in your application

  • A familiar chat surface for users (copilot UX), backed by a real procedure with tools and guardrails.
  • Asynchronous human-in-the-loop: the agent can run independently, then pause to ask for a decision only when it hits a checkpoint (approval, missing input, high-risk side effect).
  • Centralized control over where the agent is allowed to act, what it can call, and when a human must confirm.

Architecture: Sidecar Chat

Host ApplicationChat UITactus RuntimeHuman CheckpointsTools & APIs

User interacts with the embedded Chat UI in your application.

Chat UI + async queued HITL

A copilot is more than “AI chat in your app.” You get the chat surface, plus an asynchronous interface paradigm: the agent can run on its own and only ask for human input at the moments that matter. Those requests queue up and the workflow resumes the moment the human responds.

Closely supervised
The chat surface. Great when a human is present to steer and supervise each step.
HumanInput QueueAgent
Asynchronous queued HITL
The unlock. The agent runs independently and queues approvals/inputs when needed, so throughput isn't gated on constant supervision.
HumanInput QueueAgentHuman Queue
Where to start

Use the IDE to play with these interactions quickly, then embed the same components in your app when you are ready.

Want runnable procedures too?

Use cases show the UX and the guardrails. Examples give you runnable code with embedded specs.