codevault 1.3.0-beta.7 → 1.4.0

package/.env.example

@@ -1,36 +1,58 @@
-# OpenAI / Nebius API Configuration
-# For Nebius AI (Qwen embeddings), use your Nebius JWT token and base URL
-OPENAI_API_KEY=your-api-key-here
-OPENAI_BASE_URL=https://api.openai.com/v1
+# Embedding API Configuration
+# Using Nebius AI Studio for Qwen embeddings
+CODEVAULT_EMBEDDING_API_KEY=your-nebius-jwt-token-here
+CODEVAULT_EMBEDDING_BASE_URL=https://api.studio.nebius.com/v1
 
 # Embedding Model Configuration
-# For standard OpenAI: text-embedding-3-large
-# For Nebius/Qwen: Qwen/Qwen3-Embedding-8B
-CODEVAULT_OPENAI_EMBEDDING_MODEL=text-embedding-3-large
+# Qwen/Qwen3-Embedding-8B (4096 dimensions, 32K context)
+# Alternatives:
+#   - text-embedding-3-large (OpenAI, 3072 dims, 8K context)
+#   - text-embedding-3-small (OpenAI, 1536 dims, 8K context)
+CODEVAULT_EMBEDDING_MODEL=Qwen/Qwen3-Embedding-8B
 
-# For Ollama (local embeddings)
-CODEVAULT_OLLAMA_MODEL=nomic-embed-text
+# For Ollama (local embeddings, no API key needed)
+# CODEVAULT_OLLAMA_EMBEDDING_MODEL=nomic-embed-text
 
-# Chunking Configuration
-# Standard: 8192 tokens, Qwen: 32000 tokens
-CODEVAULT_MAX_TOKENS=8192
+# Embedding Chunking Configuration
+# Qwen supports 32K tokens per chunk
+CODEVAULT_EMBEDDING_MAX_TOKENS=32000
 
 # Embedding Dimensions
-# OpenAI text-embedding-3-large: 3072
-# Qwen3-Embedding-8B: 4096
-# Ollama nomic-embed-text: 768
-CODEVAULT_DIMENSIONS=3072
-
-# Rate Limiting
-# Adjust based on your API tier
-CODEVAULT_RATE_LIMIT_RPM=10000
-CODEVAULT_RATE_LIMIT_TPM=600000
-
-# Reranking API (Optional)
-# For NVIDIA reranking service
-CODEVAULT_RERANK_API_URL=https://ai.api.nvidia.com/v1/retrieval/nvidia/nv-rerankqa-mistral-4b-v3/reranking
-CODEVAULT_RERANK_API_KEY=your-nvidia-api-key-here
-CODEVAULT_RERANK_MODEL=nvidia/nv-rerankqa-mistral-4b-v3
+# Must match your model: Qwen = 4096, OpenAI large = 3072, small = 1536, Ollama = 768
+CODEVAULT_EMBEDDING_DIMENSIONS=4096
+
+# Embedding API Rate Limiting
+# Nebius AI Studio limits (adjust based on your tier)
+CODEVAULT_EMBEDDING_RATE_LIMIT_RPM=10000
+CODEVAULT_EMBEDDING_RATE_LIMIT_TPM=600000
+
+# Reranking API (Optional - improves search relevance by 5-10%)
+# Novita Qwen3-Reranker - 32K context, great for code
+CODEVAULT_RERANK_API_URL=https://api.novita.ai/openai/v1/rerank
+CODEVAULT_RERANK_API_KEY=your-novita-api-key-here
+CODEVAULT_RERANK_MODEL=qwen/qwen3-reranker-8b
+
+# Alternatives:
+# Cohere - 4K context, $25 free credits
+# CODEVAULT_RERANK_API_URL=https://api.cohere.ai/v1/rerank
+# CODEVAULT_RERANK_API_KEY=your-cohere-key-here
+# CODEVAULT_RERANK_MODEL=rerank-english-v3.0
+
+# Jina AI - 8K context, free tier available
+# CODEVAULT_RERANK_API_URL=https://api.jina.ai/v1/rerank
+# CODEVAULT_RERANK_API_KEY=your-jina-key-here
+# CODEVAULT_RERANK_MODEL=jina-reranker-v2-base-multilingual
+
+# Chat LLM Configuration (for ask_codebase feature)
+# Uses same API as embeddings by default, but can be different
+CODEVAULT_CHAT_API_KEY=your-nebius-jwt-token-here
+CODEVAULT_CHAT_BASE_URL=https://api.studio.nebius.com/v1
+CODEVAULT_CHAT_MODEL=Qwen/Qwen2.5-72B-Instruct
+CODEVAULT_CHAT_MAX_TOKENS=4096
+CODEVAULT_CHAT_TEMPERATURE=0.7
+
+# For Ollama Chat (local, no API key needed)
+# CODEVAULT_OLLAMA_CHAT_MODEL=llama3.1
 
 # Encryption (Optional)
 # 32-byte key encoded as base64 or hex