FeaturesAI AssistIndex

AI-Assisted Features

Use your own LLM keys to enhance descriptions, generate tools, scenarios, and full MCP or agent card assets.

Overview

A²D supports Bring Your Own Key (BYOK) for AI: you configure one or more LLM providers (OpenAI, Anthropic, Ollama, etc.) at the organization level. API keys are stored encrypted. Once a default configuration is set and tested, AI-assisted actions become available in the UI.

Configuring AI

  1. Open Settings (or your organization’s AI configuration entry point).
  2. Add a new LLM configuration: choose provider, enter API key, optional base URL, and model.
  3. Test connection to verify the key and endpoint.
  4. Set one configuration as Default so it is used for all AI features.

Required: Set CREDENTIALS_ENCRYPTION_KEY in your environment (e.g. in .env.local) so the application can encrypt and store API keys. See your deployment or .env.example for details.

What You Can Generate

  • Enhance description — Improve name and description for an MCP server or A2A agent using the current default model.
  • Generate tools — Propose MCP tools from a short description.
  • Generate scenarios — Propose mock scenarios (condition/response pairs) for MCP or A2A.
  • Generate full asset — Generate a full MCP server or agent card from a short description (e.g. “A CRM agent that can retrieve contacts, accounts and orders”). Prefer shorter descriptions to avoid very long outputs and JSON parsing issues.
  • Generate skills — Propose agent card skills from context.
  • Generate REST endpoints — Propose HTTP operations and paths for a REST API from a short description.
  • Generate REST scenarios — Propose mock scenarios for REST endpoints.
  • Generate full REST API — Propose a full REST API (endpoints, schemas, scenarios) from a brief prompt.

If the model returns invalid JSON, you may see an error such as “Expected ’,’ or ’}’ after property value”. Try a shorter description or try again. Operators can set DEBUG_AI_JSON=1 to log response length and a snippet around the parse error position for troubleshooting.

Providers

Supported providers include OpenAI, Anthropic (Claude), Google Gemini, Azure OpenAI, Groq, Mistral, OpenRouter, Together, Ollama, Heroku, and others. Local and gateway providers (e.g. Ollama) are useful for development without exposing cloud keys.

System Prompts

AI behavior is driven by system prompts stored in the database. Organizations can override global prompts. Prompt templates use placeholders such as {{name}}, {{description}}, and {{context}}. See your administrator or implementation docs for customizing prompts.

Next Steps

  • MCP Servers — Create or edit servers and use AI to enhance descriptions or generate tools.
  • Agent Cards — Use AI to generate skills or full agent cards.
  • Design Rules — Ensure generated content meets compliance.

Configure a default LLM in Settings to unlock AI-assisted features across the platform.

OverviewOrganization & Members