Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.llmgrid.ai/llms.txt

Use this file to discover all available pages before exploring further.

​
Overview

The Agents screen lets you register and manage agents that can be invoked by applications and workflows. Agents expose skills (what they can do), declare capabilities (how they behave), and can optionally be made discoverable in the AI Hub.

​
Agents List

The main Agents page shows a list of all agents in the selected tenant.

​
Empty state

If there are no agents, the page displays a prompt indicating that no agents are found and you can create one to get started.

​
Primary action

  • Add New Agent — opens the agent creation dialog.

​
Add New Agent

Selecting Add New Agent opens a multi-section form used to define identity, endpoints, skills, and behaviors.

​
Agent Name (required)

A unique, machine-readable identifier for the agent. Example
  • customer-support-agent
Use a short, stable name since this is commonly referenced by clients and automation.

​
Basic Information (Required)

​
Display Name (required)

A human-friendly name shown across the console and (if published) in discovery surfaces. Example
  • Customer Support Agent

​
Description (required)

A clear summary of what the agent does. Guidelines
  • Describe core jobs (e.g., “triages tickets”, “answers policy questions”, “generates summaries”)
  • Mention expected inputs/outputs at a high level
  • Avoid internal implementation details

​
URL (required)

The base endpoint where the agent is hosted. Example
  • https://agent.example.com
This endpoint is invoked when the agent is called.

​
Version

Agent implementation version used for lifecycle tracking. Example
  • 1.0.0

​
Protocol Version

The protocol version the agent adheres to for compatibility. Example
  • 1.0

​
Skills (Required)

Skills define what the agent can do and how clients should call it.

​
Add Skill

Select Add Skill to create one or more skills. Notes
  • At least one skill is required to create an agent.
  • Skills should be narrowly scoped and named clearly to reduce ambiguity.

​
Capabilities

Capabilities describe runtime behaviors supported by the agent.

​
Streaming

When enabled, the agent can stream partial responses.

​
Push Notifications

When enabled, the agent can send asynchronous updates (useful for long-running workflows).

​
State Transition History

When enabled, the agent can emit or retain state-change history across interactions, supporting traceability and multi-step flows. Enable only what you need to keep integrations simpler and easier to govern.

​
Optional Settings

​
Icon URL

A URL pointing to an icon representing the agent (useful for UI and discovery views). Example
  • https://example.com/icon.png

​
Documentation URL

A link to documentation that explains this agent’s skills and usage. Recommended contents
  • Skill list with purpose and examples
  • Request/response shape guidelines
  • Authentication and access prerequisites
  • Known limitations and error handling
Example
  • https://docs.example.com

​
Supports Authenticated Extended Card

Indicates whether the agent supports authenticated rich UI interactions (where applicable).

​
Advanced Parameters

This section provides additional optional parameters used to control invocation behavior.

​
Model (Optional)

Optionally associate a default model or alias with the agent. Notes
  • If unspecified, model selection can be determined by routing or the invocation context.
  • If specified, ensure the model is available and permitted for intended callers.

​
Make Public

Controls whether the agent is eligible to be listed in the AI Hub for discovery.
Making an agent public enables discovery only. Authentication, access controls, routing rules, and guardrails still apply.

​
Create or Cancel

  • Create Agent — validates required fields and saves the agent.
  • Cancel — closes the dialog without saving.
After creation, the agent appears in the Agents list and can be published through AI Hub → Agent Hub (if enabled).

​
Best Practices

  • Use a stable Agent Name and a descriptive Display Name
  • Keep skills small, explicit, and easy to test
  • Enable only the capabilities your client needs (streaming, push, state history)
  • Provide a Documentation URL for every agent intended for broad usage
  • Test the agent end-to-end before making it discoverable
  • AI Hub — publish agents for discovery
  • MCP Servers — provide tools agents can use
  • Models — manage and govern available models
  • Playground — validate behavior before rollout
  • Virtual Keys — control authentication and operational limits