RAG API

The RAG (Retrieval-Augmented Generation) pipeline lets you index your codebase and give agents deep contextual understanding of your project.

Index Code

Upload code to the RAG pipeline for vectorization.

POST /rag/index
Authorization: Bearer <token>
Content-Type: application/json

Request Body:

{
  "files": [
    {
      "path": "src/auth.ts",
      "content": "export function authenticate(token: string) { ... }"
    },
    {
      "path": "src/database.ts",
      "content": "export class Database { ... }"
    }
  ],
  "repository": "my-project"
}

Field	Type	Required	Description
files	array	✅	Array of {path, content} objects
repository	string	❌	Repository identifier for grouping

Response (200):

{
  "indexed": 2,
  "vectors": 15,
  "repository": "my-project",
  "message": "2 files indexed (15 vectors)"
}

---

Search Code

Query the indexed codebase using semantic search.

POST /rag/search
Authorization: Bearer <token>
Content-Type: application/json

Request Body:

{
  "query": "How does authentication work?",
  "repository": "my-project",
  "limit": 5
}

Field	Type	Required	Description
query	string	✅	Natural language search query
repository	string	❌	Filter by repository
limit	number	❌	Max results (default: 5, max: 20)

Response (200):

{
  "results": [
    {
      "path": "src/auth.ts",
      "content": "export function authenticate(token: string) {\n  const decoded = jwt.verify(token, SECRET);\n  return decoded;\n}",
      "score": 0.92,
      "chunk": 3
    },
    {
      "path": "src/middleware/auth.ts",
      "content": "export const requireAuth = async (req, res, next) => {\n  const token = req.headers.authorization?.split(' ')[1];\n  ...\n}",
      "score": 0.87,
      "chunk": 1
    }
  ],
  "query": "How does authentication work?",
  "total": 2
}

---

Using RAG with Agents

Combine RAG context with agent execution for codebase-aware analysis:

# 1. Index your codebase
find src -name "*.ts" | while read f; do
  echo "{\"path\":\"$f\",\"content\":$(cat "$f" | jq -Rs .)}"
done | jq -s '{files: .}' | \
curl -X POST https://api.lunaos.ai/rag/index \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d @-

# 2. Search for relevant context
CONTEXT=$(curl -s -X POST https://api.lunaos.ai/rag/search \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"query": "authentication flow"}' \
  | jq -r '.results[].content' | head -c 4000)

# 3. Run agent with RAG context
curl -N -X POST https://api.lunaos.ai/agents/execute \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d "{\"agent\": \"code-review\", \"context\": \"$CONTEXT\"}"

How RAG Works

1. Chunking: Files are split into semantic chunks (functions, classes, blocks)
2. Embedding: Each chunk is converted to a vector using Workers AI
3. Storage: Vectors are stored in Cloudflare Vectorize
4. Retrieval: Queries are embedded and matched against stored vectors using cosine similarity
5. Augmentation: Retrieved chunks are injected into the agent's context for grounded responses