Step 1: Getting Context (with RAG)

1) Clone + Setup enterprise-ai-infra

From a parent directory:

git clone https://github.com/bitovi/enterprise-ai-infra.git
cd enterprise-ai-infra

Create required secrets/config

cp .env.example .env

Fill in .env:

OPENAI_API_KEY
GITHUB_ACCESS_TOKEN (PAT with repo read access)
TEMPORAL_NAMESPACE
TEMPORAL_HOST_PORT
TEMPORAL_API_KEY

Optional path overrides (if you already have local checkouts):

PATH_EAI_AGENT_WORKER
PATH_EAI_PIPELINE_WORKEzR
PATH_EAI_MCP

Start everything

make up

This runs setup.sh, which clones required repos into ./modules and starts Docker Compose in watch mode.

2) Validate running services

Important local endpoints:

Qdrant: http://localhost:6333
Qdrant dashboard: http://localhost:6333/dashboard
Agent API: http://localhost:3101
Pipeline API: http://localhost:8002
MCP: http://localhost:3111/mcp

3) Ingest code into Qdrant

All ingestion is triggered through the pipeline API.

3a) Ingest an entire GitHub organization

ORG_NAME=your-org

curl -X POST "http://localhost:8002/organization-repository-vectorization/run" \
     -H "Content-Type: application/json" \
     -d '{"organization_name": "'"${ORG_NAME}"'"}'

Optional request body fields:

file_ext_filter (example: .ts)
chunk_size (default 1000)
chunk_overlap (default 0)
repository_names (default '')

Example with filters:

ORG_NAME=your-org

curl -X POST "http://localhost:8002/organization-repository-vectorization/run" \
     -H "Content-Type: application/json" \
     -d '{
       "organization_name": "'"${ORG_NAME}"'",
       "file_ext_filter": ".ts",
       "chunk_size": 1200,
       "chunk_overlap": 100
     }'

3b) Ingest one specific repo

ORG_NAME=your-org
REPO_NAME=your-repo

curl -X POST "http://localhost:8002/github-code-vectorization/run" \
     -H "Content-Type: application/json" \
     -d '{"organization_name": "'"${ORG_NAME}"'", "repository_name": "'"${REPO_NAME}"'"}'

Optional request body fields:

branch_name (default main)
file_ext_filter
chunk_size
chunk_overlap

Ingest a set of repos (repeat single-repo endpoint)

For a selected set, call the single-repo endpoint once per repo:

ORG_NAME=your-org


curl -X POST "http://localhost:8002/organization-repository-vectorization/run" \
     -H "Content-Type: application/json" \
     -d '{
       "organization_name": "'"${ORG_NAME}"'",
       "repository_names": "repo-a,repo-b,repo-c"
     }'

4) Connect Solutions Architect to local MCP

In .vscode/mcp.json in your repo, configure:

{
  "servers": {
    "enterprise-ai": {
      "type": "http",
      "url": "http://localhost:3111/mcp"
    }
  }
}

This points solutions-architect prompts at your locally running enterprise MCP service.

5) Move Prompts into Github Folder

Take both prompt files from bitovi/ai-enablement-prompts/solutions-architect-rag and move them into .github/prompts. This allows them to be used in a command-style format from the chat window.

Other Useful Operations

Stop the infra:

cd enterprise-ai-infra
make down

Reset Qdrant collection data:

cd enterprise-ai-infra
docker exec eai_infra_admin_tools /usr/bin/python ./qdrant_reset.py