wrangler.toml.example

name = "vectorize-mcp-worker"
main = "src/index.ts"
compatibility_date = "2025-12-02"

[ai]
binding = "AI"

# RAG Index — must match your embedding model dimensions:
#   qwen3-0.6b (default, recommended): --dimensions=1024
#   bge-small  (legacy, 384d):         --dimensions=384
[[vectorize]]
binding = "VECTORIZE"
index_name = "mcp-knowledge-base"

[[d1_databases]]
binding = "DB"
database_name = "mcp-knowledge-db"
database_id = "YOUR_DATABASE_ID_HERE"  # Replace with your actual D1 database ID

# Service binding to the multimodal worker (internal-only)
[[services]]
binding = "MULTIMODAL"
service = "multimodal-pro-worker"

[observability]
enabled = true

[vars]
# Embedding model key. Default: "qwen3-0.6b" (1024d, best retrieval quality 2026).
# Set to "bge-small" to keep an existing 384d deployment.
# Set to "bge-m3" for multilingual support (1024d).
# EMBEDDING_MODEL = "qwen3-0.6b"

# Reflection / synthesis model. Default: "kimi-k2.5" (best multi-document reasoning).
# Set to "llama-3.2-3b" for lower cost at the expense of synthesis quality.
# REFLECTION_MODEL = "kimi-k2.5"

# Set your API key using: wrangler secret put API_KEY
# Never commit your actual API_KEY to git