Merge pull request #20 from rostilos/0.1.2-rc

0.1.2 rc

Author: rostilos

README.md

@@ -6,23 +6,23 @@
 
 CodeCrow supports multiple version control systems with varying levels of integration. Below is the current feature matrix:
 
-| Feature | Bitbucket | GitHub | GitLab |
-| :--- | :---: | :---: | :---: |
-| PR Analysis | + | + | - |
-| Branch Analysis | + | + | - |
+| Feature                | Bitbucket | GitHub | GitLab |
+|:-----------------------| :---: | :---: | :---: |
+| PR Analysis            | + | + | - |
+| Branch Analysis        | + | + | - |
 | Task Context Retrieval | - | - | - |
-| /ask | + | + | - |
-| /analyze | + | + | - |
-| /review | + | + | - |
-| Continuous Analysis | + | + | - |
-| RAG Pipeline | + | + | - |
+| /ask                   | + | + | - |
+| /analyze               | + | + | - |
+| /summarize             | + | + | - |
+| Continuous Analysis    | + | + | - |
+| RAG Pipeline           | + | + | - |
 
 ## Key Features
 
 - **Context-Aware Reviews**: Powered by a custom RAG (Retrieval-Augmented Generation) pipeline using Qdrant vector storage.
 - **Incremental Analysis**: Only scans changed code to keep feedback fast and cost-efficient.
 - **Multi-Tenant Architecture**: Securely manage multiple teams and projects from a single dashboard.
-- **Interactive Commands**: Command CodeCrow directly from PR comments using `/ask`, `/analyze`, and `/review`.
+- **Interactive Commands**: Command CodeCrow directly from PR comments using `/ask`, `/analyze`, and `/summarize`.
 
 ## Documentation
 

deployment/config/mcp-client/.env.sample

@@ -6,5 +6,60 @@ RAG_API_URL=http://host.docker.internal:8001
 CODECROW_API_URL=http://codecrow-web-application:8081
 
 # Platform MCP JAR path
-PLATFORM_MCP_JAR=/app/codecrow-platform-mcp-1.0.jar\
-INTERNAL_API_SECRET=codecrow-internal-secret-change-me
+PLATFORM_MCP_JAR=/app/codecrow-platform-mcp-1.0.jar
+INTERNAL_API_SECRET=codecrow-internal-secret-change-me
+
+# === LLM Settings ===
+# Temperature for code review (0.0 = deterministic, 0.1-0.3 = more creative)
+LLM_TEMPERATURE=0.0
+
+# === Direct Review Mode ===
+# When rawDiff is provided in request, bypasses MCP agent for faster review
+# This is controlled by pipeline-agent sending rawDiff field
+# No configuration needed - automatic when rawDiff is present
+
+# === Diff Processing Limits (matching MCP server LargeContentFilter) ===
+# Maximum file size in bytes (default: 25KB - same as LargeContentFilter.DEFAULT_SIZE_THRESHOLD_BYTES)
+DIFF_MAX_FILE_SIZE=25600
+# Maximum files to include in review (default: 100)
+DIFF_MAX_FILES=100
+# Maximum total diff size in bytes (default: 500KB)
+DIFF_MAX_TOTAL_SIZE=500000
+# Maximum lines per file (default: 1000)
+DIFF_MAX_LINES_PER_FILE=1000
+
+# === LLM Reranking (for large PRs) ===
+# Enable LLM-based reranking for large PRs
+LLM_RERANK_ENABLED=true
+# Minimum files to trigger LLM reranking (default: 20)
+LLM_RERANK_THRESHOLD=20
+# Max RAG items to rerank via LLM
+LLM_RERANK_MAX_ITEMS=15
+
+# === RAG Context Settings ===
+# Minimum relevance score for RAG results (0.0-1.0)
+RAG_MIN_RELEVANCE_SCORE=0.7
+# Default number of RAG chunks to retrieve
+RAG_DEFAULT_TOP_K=15
+# Cache TTL in seconds
+RAG_CACHE_TTL_SECONDS=300
+# Max cache entries
+RAG_CACHE_MAX_SIZE=100
+
+# === Context Budget (token allocation percentages, must sum to 1.0) ===
+CONTEXT_BUDGET_HIGH_PRIORITY_PCT=0.30
+CONTEXT_BUDGET_MEDIUM_PRIORITY_PCT=0.40
+CONTEXT_BUDGET_LOW_PRIORITY_PCT=0.20
+CONTEXT_BUDGET_RAG_PCT=0.10
+
+# === Prompt Logging (Debug) ===
+# Enable prompt logging for debugging
+PROMPT_LOG_ENABLED=true
+# Log full prompts to file
+PROMPT_LOG_TO_FILE=true
+# Also log to console (verbose)
+PROMPT_LOG_TO_CONSOLE=false
+# Directory for prompt logs
+PROMPT_LOG_DIR=/tmp/codecrow_prompts
+# Max log files to keep (auto-cleanup)
+PROMPT_LOG_MAX_FILES=50

deployment/config/rag-pipeline/.env.sample

@@ -5,7 +5,7 @@ QDRANT_COLLECTION_PREFIX=codecrow
 # OpenRouter Configuration
 # Get your API key from https://openrouter.ai/
 OPENROUTER_API_KEY=sk-or-v1-your-api-key-here
-OPENROUTER_MODEL=openai/text-embedding-3-small
+OPENROUTER_MODEL=qwen/qwen3-embedding-8b
 
 #Additional Environment Variables
 HOME=/tmp
@@ -14,6 +14,10 @@ TRANSFORMERS_CACHE=/tmp/.transformers_cache
 HF_HOME=/tmp/.huggingface
 LLAMA_INDEX_CACHE_DIR=/tmp/.llama_index
 
+# Free Plan Limits (Optional - set to 0 to disable limits)
+RAG_MAX_CHUNKS_PER_INDEX=70000
+RAG_MAX_FILES_PER_INDEX=40000
+
 # Alternative OpenRouter models (use full format with provider prefix):
 # OPENROUTER_MODEL=openai/text-embedding-3-large  # Higher quality, more expensive
 # OPENROUTER_MODEL=openai/text-embedding-ada-002  # Legacy model