EAFIT Challenge — Example Agent

An example AI agent with GitHub MCP integration, deployed as a Verifiable Service on the Verana ecosystem. This is part of the EAFIT Challenge.

Architecture

This example agent is built with the hologram-generic-ai-agent-vs container, which provides a ready-to-use AI chatbot with DIDComm messaging, verifiable credential authentication, and MCP tool integration.

You can find other agent examples (GitHub Agent, Wise Agent, etc.) in the hologram-verifiable-services repository.

This agent is a child service of the EAFIT Challenge organization. It:

1. Receives a Service credential from the organization (proves it's a legitimate service)
2. Uses the Avatar credential definition from avatar.eafit.testnet.verana.network to authenticate users via AnonCreds proof requests
3. Provides GitHub MCP tools to authenticated users via encrypted DIDComm chat

Repository Structure

├── config.env              # Service configuration (ports, org URLs, credDef, etc.)
├── deployment.yaml         # Helm chart values for K8s deployment
├── agent-pack.yaml         # Chatbot agent pack (prompts, menus, MCP config)
├── common/
│   └── common.sh           # Shared shell helpers
├── docker/
│   └── docker-compose.yml  # Local development stack
├── scripts/
│   ├── setup.sh            # Local setup (VS Agent + ngrok + Service credential)
│   └── start.sh            # Start Docker Compose stack
├── docs/
│   └── README.md           # User-facing guide
└── .github/
    └── workflows/
        └── deploy.yml      # GitHub Actions workflow for K8s deployment

Local Development

Prerequisites

• Docker and Docker Compose
• ngrok (authenticated)
• curl, jq
• An OpenAI API key

The setup script connects to the deployed EAFIT organization at admin.organization.eafit.testnet.verana.network to obtain the Service credential. No local organization instance is required.

Quick Start

# 1. Set up the VS Agent (deploys container, gets Service credential)
source config.env
./scripts/setup.sh

# 2. Start the full stack (chatbot + redis + postgres)
export NGROK_DOMAIN=<your-ngrok-domain>
export OPENAI_API_KEY=sk-...
./scripts/start.sh

Note: if you don't want to use an OPENAI_API_KEY, you can configure any other LLM, refer to agent pack schema for available options.

Kubernetes Deployment (GitHub Actions)

The .github/workflows/deploy.yml workflow deploys the agent to the shared EAFIT Challenge K8s cluster.

Required GitHub Secrets

Secret	Description
OVH_KUBECONFIG	Kubeconfig for the K8s cluster
K8S_NAMESPACE	Target namespace (ideally, use your team name)
EXAMPLE_AGENT_OPENAI_API_KEY	OpenAI API key for the chatbot
EXAMPLE_AGENT_POSTGRES_PASSWORD	PostgreSQL password
EXAMPLE_AGENT_MCP_CONFIG_ENCRYPTION_KEY	Encryption key for MCP user configs (generate with openssl rand -hex 32)
EXAMPLE_AGENT_WALLET_KEY	VS Agent wallet encryption key (generate with openssl rand -base64 32)
EXAMPLE_AGENT_VSAGENT_DB_PASSWORD	VS Agent internal DB password

Deployment

Run the workflow from the GitHub Actions tab with step all to deploy and obtain credentials.

The agent will be available at the URL configured in AGENT_PUBLIC_URL (see below).

Configuration

Key settings in config.env:

• AGENT_PUBLIC_URL — Public URL of the deployed agent. For student teams, use the convention: https://<agentname>.agents.<team_name>.teams.eafit.testnet.verana.network
• CREDENTIAL_DEFINITION_ID — AnonCreds credDef from the EAFIT Avatar service (hardcoded)
• ORG_VS_PUBLIC_URL — Public URL of the EAFIT organization agent
• SERVICE_NAME — Display name shown in the Service credential
• OPENAI_MODEL — LLM model for the chatbot

License

Apache-2.0