Skip to main content

AI Chat

AI Chat provides a production-ready conversational layer for interacting with your agents. It supports real-time responses, tool execution, and operational controls while keeping chat flows readable and auditable.

Key Capabilities​

Multi-protocol chat connectivity​

Use the same chat experience with different backend protocols, including WebSocket and HTTP-based transports. This helps you standardize the frontend experience while keeping flexibility in backend architecture.

Flexible chat surfaces​

Choose the interaction pattern that fits your product:

  • Embedded chat panels in application pages.
  • Sidebar chat for notebook or document workflows.
  • Floating chat for low-disruption assistance.
  • Standalone chat views for dedicated agent tasks.

Streaming and tool execution​

AI Chat supports token streaming and structured tool calls so users can:

  • See progressive responses in real time.
  • Trigger agent tools directly from chat.
  • Track tool-call execution and outputs.
  • Mix narrative responses with structured artifacts.

Model, skills, and tools controls​

When enabled by policy, chat can expose controls for:

  • Model selection.
  • MCP tool availability.
  • Skills enable/disable state.
  • Runtime-aware execution options.

These controls help teams adapt behavior per workflow without rebuilding UI.

Operational indicators and safety​

AI Chat includes status signals and controls for safer operation:

  • Runtime and execution state indicators.
  • Sandbox status visibility.
  • Interrupt support for long-running execution.
  • Approval-aware flows for sensitive actions.

Conversation quality and continuity​

Designed for real production workflows:

  • Conversation persistence by runtime/session.
  • Prompt suggestions for fast onboarding.
  • Empty-state guidance and branded headers.
  • Error banners and inline recovery patterns.

Typical Workflow​

  1. Connect chat to an agent endpoint.
  2. Enable the required protocol and runtime settings.
  3. Configure visible controls (model, tools, skills) by role.
  4. Add status indicators and approval gates where needed.
  5. Persist conversation history for continuity and audit.

Best Practices​

  • Start with a minimal chat surface, then progressively expose advanced controls.
  • Keep tool execution explicit and observable in the UI.
  • Use approval gates for actions that affect external systems or sensitive data.
  • Pair chat operations with evaluations and monitoring for reliable production behavior.

Next Steps​