feat/vision_slover

config/agents.yaml

@@ -55,3 +55,16 @@ co_writer:
 narrator:
   temperature: 0.7
   max_tokens: 4000
+
+# =============================================================================
+# Vision Solver Module - Image analysis and GeoGebra visualization
+# =============================================================================
+# Agents: vision_solver_agent (with bbox, analysis, ggbscript, reflection stages)
+# This module handles:
+# - BBox: Visual element detection with pixel coordinates
+# - Analysis: Geometric semantic analysis
+# - GGBScript: GeoGebra command generation
+# - Reflection: Command validation and correction
+vision_solver:
+  temperature: 0.3
+  max_tokens: 12000

config/main.yaml

@@ -1,5 +1,5 @@
 system:
-  language: en
+  language: zh
 paths:
   user_data_dir: ./data/user
   knowledge_bases_dir: ./data/knowledge_bases
@@ -10,6 +10,7 @@ paths:
   research_output_dir: ./data/user/research/cache
   research_reports_dir: ./data/user/research/reports
   solve_output_dir: ./data/user/solve
+  vision_solver_output_dir: ./data/user/vision_solver
 tools:
   rag_tool:
     kb_base_dir: ./data/knowledge_bases
@@ -66,6 +67,23 @@ solve:
       max_iterations: 3
     precision_answer_agent:
       enabled: true
+vision_solver:
+  enabled: true
+  max_image_size: 10MB
+  geogebra:
+    xmin: -10
+    xmax: 10
+    ymin: -8
+    ymax: 8
+  stages:
+    bbox:
+      temperature: 0.3
+    analysis:
+      temperature: 0.3
+    ggbscript:
+      temperature: 0.3
+    reflection:
+      temperature: 0.3
 research:
   planning:
     rephrase:

docs/ref/wiki/geogrebra-llms.md

@@ -0,0 +1,7 @@
+## This page describes the GeoGebra Apps API to interact with GeoGebra apps. Please see GeoGebra Apps Embedding on how to embed our apps into your web pages.
+
+- https://geogebra.github.io/docs/reference/en/GeoGebra_Apps_API/
+
+## This page lists all parameters that can be used to configure GeoGebra Apps. Please see GeoGebra Apps Embedding to learn how to embed GeoGebra apps into your pages and where these parameters can be used.
+
+- https://geogebra.github.io/docs/reference/en/GeoGebra_App_Parameters/

scripts/install_all.sh

@@ -78,23 +78,23 @@ if [ ! -f "$REQUIREMENTS_FILE" ]; then
     exit 1
 fi
 
-print_info "Using Python: $(which python)"
+print_info "Using Python: $(which python3)"
 print_info "Requirements file: $REQUIREMENTS_FILE"
 
 # Function to install with uv (faster and better resolver)
 install_with_uv() {
     # Check if uv is available
     if ! command -v uv &> /dev/null; then
         print_info "Installing uv for better dependency resolution..."
-        if python -m pip install uv; then
+        if python3 -m pip install uv; then
             print_success "uv installed successfully"
         else
             return 1
         fi
     fi
 
     print_info "Using uv for faster dependency resolution..."
-    if python -m uv pip install -r "$REQUIREMENTS_FILE"; then
+    if python3 -m uv pip install -r "$REQUIREMENTS_FILE"; then
         return 0
     else
         return 1
@@ -107,7 +107,7 @@ install_with_pip_staged() {
 
     # Stage 1: Core dependencies
     print_info "Stage 1/3: Installing core dependencies..."
-    if python -m pip install \
+    if python3 -m pip install \
         "python-dotenv>=1.0.0" \
         "PyYAML>=6.0" \
         "tiktoken>=0.5.0" \
@@ -137,15 +137,15 @@ install_with_pip_staged() {
     print_info "Stage 3/4: Installing raganything (includes lightrag-hku, this may take a while)..."
     if ! python -m pip install "raganything>=0.1.0"; then
         print_warning "Standard install failed, trying with --no-deps..."
-        python -m pip install "raganything>=0.1.0" --no-deps || print_warning "raganything installation had issues"
+        python3 -m pip install "raganything>=0.1.0" --no-deps || print_warning "raganything installation had issues"
     fi
 
     # Stage 4: docling (alternative parser for Office/HTML documents)
     print_info "Stage 4/4: Installing docling (document parser for Office/HTML)..."
     python -m pip install "docling>=2.31.0" || print_warning "docling installation had issues (optional, can be skipped)"
 
     # Optional deps
-    python -m pip install "perplexityai>=0.1.0" "dashscope>=1.14.0" 2>/dev/null || true
+    python3 -m pip install "perplexityai>=0.1.0" "dashscope>=1.14.0" 2>/dev/null || true
 
     return 0
 }
@@ -165,7 +165,7 @@ else
     else
         # Strategy 3: Direct pip as last resort
         print_warning "Staged installation had issues, trying direct pip install..."
-        if python -m pip install -r "$REQUIREMENTS_FILE"; then
+        if python3 -m pip install -r "$REQUIREMENTS_FILE"; then
             print_success "Backend dependencies installed successfully"
         else
             print_error "Backend dependencies installation failed"
@@ -318,7 +318,7 @@ ALL_OK=true
 print_info "Checking backend key packages..."
 
 check_python_package() {
-    if python -c "import $1" 2>/dev/null; then
+    if python3 -c "import $1" 2>/dev/null; then
         print_success "  ✓ $1"
         return 0
     else
@@ -332,15 +332,15 @@ check_python_package uvicorn || ALL_OK=false
 check_python_package openai || ALL_OK=false
 
 # Check lightrag_hku (import name is lightrag)
-if python -c "import lightrag" 2>/dev/null; then
+if python3 -c "import lightrag" 2>/dev/null; then
     print_success "  ✓ lightrag_hku"
 else
     print_error "  ✗ lightrag_hku not installed"
     ALL_OK=false
 fi
 
 # Check raganything
-if python -c "import raganything" 2>/dev/null; then
+if python3 -c "import raganything" 2>/dev/null; then
     print_success "  ✓ raganything"
 else
     print_error "  ✗ raganything not installed"

src/agents/vision_solver/__init__.py

@@ -0,0 +1,27 @@
+"""Vision Solver Agent Module.
+
+This module implements image analysis and GeoGebra visualization for math problems.
+It follows a four-stage pipeline:
+1. BBox - Visual element detection with pixel coordinates
+2. Analysis - Geometric semantic analysis
+3. GGBScript - Generate GeoGebra drawing commands
+4. Reflection - Validate and fix commands
+"""
+
+from src.agents.vision_solver.models import (
+    AnalysisOutput,
+    BBoxOutput,
+    GGBScriptOutput,
+    ImageAnalysisState,
+    ReflectionOutput,
+)
+from src.agents.vision_solver.vision_solver_agent import VisionSolverAgent
+
+__all__ = [
+    "VisionSolverAgent",
+    "BBoxOutput",
+    "AnalysisOutput",
+    "GGBScriptOutput",
+    "ReflectionOutput",
+    "ImageAnalysisState",
+]