Feature UI

.devcontainer/devcontainer.json

@@ -1,29 +1,29 @@
-{
-  "name": "ai-agent-dev",
-  "build": {
-    "dockerfile": "Dockerfile"
-  },
-
-  // This is where your repo will be mounted inside the container
-  "remoteUser": "vscode",
-  "workspaceFolder": "/workspaces/${localWorkspaceFolderBasename}",
-
-  "customizations": {
-    "vscode": {
-    "settings": {
-      "python.defaultInterpreterPath": "${workspaceFolder}/.venv/bin/python",
-      "python.envFile": "${workspaceFolder}/.env"
-    },
-      "extensions": [
-        "ms-python.python",
-        "ms-python.vscode-pylance",
-        "tamasfe.even-better-toml"
-      ]
-    }
-  },
-
-  "forwardPorts": [7860],
-
-  // Install project in editable mode after the container is built  
-  "postCreateCommand": "rm -rf .venv && uv venv && uv pip install -e . && echo '. $PWD/.venv/bin/activate' >> /home/vscode/.bashrc"
-}
+{
+  "name": "ai-agent-dev",
+  "build": {
+    "dockerfile": "Dockerfile"
+  },
+
+  // This is where your repo will be mounted inside the container
+  "remoteUser": "vscode",
+  "workspaceFolder": "/workspaces/${localWorkspaceFolderBasename}",
+
+  "customizations": {
+    "vscode": {
+    "settings": {
+      "python.defaultInterpreterPath": "${workspaceFolder}/.venv/bin/python",
+      "python.envFile": "${workspaceFolder}/.env"
+    },
+      "extensions": [
+        "ms-python.python",
+        "ms-python.vscode-pylance",
+        "tamasfe.even-better-toml"
+      ]
+    }
+  },
+
+  "forwardPorts": [7860],
+
+  // Install project in editable mode after the container is built  
+  "postCreateCommand": "rm -rf .venv && uv venv && uv pip install -e . && echo '. $PWD/.venv/bin/activate' >> /home/vscode/.bashrc"
+}

.env.dist

@@ -1,19 +1,19 @@
-OPENAI_API_KEY=sk-xxxx
-GITHUB_TOKEN=ghp_xxxx
-# Optional model overrides (defaults work):
-OPENAI_MODEL=gpt-4o
-
-# Software catalog
-SOFTWARE_CATALOG=path/to/your/catalog.jsonl
-
-# Pipeline configuration
-TOP_K=8                # Number of candidates to retrieve
-NUM_CHOICES=3          # Number of tools to recommend
-USE_AGENT=1            # Use pydantic-ai agent (1) or standard pipeline (0)
-
-# Logging configuration
-LOGLEVEL_CONSOLE=WARNING
-LOGLEVEL_FILE=INFO
-FILE_LOG=1
-LOG_DIR=logs
+OPENAI_API_KEY=sk-xxxx
+GITHUB_TOKEN=ghp_xxxx
+# Optional model overrides (defaults work):
+OPENAI_MODEL=gpt-4o
+
+# Software catalog
+SOFTWARE_CATALOG=path/to/your/catalog.jsonl
+
+# Pipeline configuration
+TOP_K=8                # Number of candidates to retrieve
+NUM_CHOICES=3          # Number of tools to recommend
+USE_AGENT=1            # Use pydantic-ai agent (1) or standard pipeline (0)
+
+# Logging configuration
+LOGLEVEL_CONSOLE=WARNING
+LOGLEVEL_FILE=INFO
+FILE_LOG=1
+LOG_DIR=logs
 LOG_PROMPTS=0         # write selector prompt snapshots

.github/copilot-instructions.md

@@ -1,130 +1,130 @@
-# AI Agent — Copilot Instructions
-
-This is a **RAG + VLM imaging tool recommender** that helps users find the right imaging software for their images and tasks. Users drop an image, describe their task, and get ranked software recommendations with demo links.
-
-## Architecture Overview
-
-The system follows a two-stage pipeline:
-
-1. **Retrieval Stage** (`retriever/`, `api/pipeline.py`): Fast text search using BGE-M3 embeddings + CrossEncoder reranker. No LLM calls. Returns top-K candidates.
-
-2. **Selection Stage** (`generator/`): Single VLM call (OpenAI GPT-4o/mini) that sees the image + candidates + metadata and returns ranked recommendations with accuracy scores.
-
-### Key Components
-
-- **`api/pipeline.RAGImagingPipeline`**: Main orchestrator. Handles file validation, metadata extraction, retrieval, and VLM selection.
-- **`retriever/embedders.py`**: FAISS vector index with BGE-M3 + CrossEncoder reranker for candidate retrieval.
-- **`generator/generator.VLMToolSelector`**: Vision-language model that selects best tools from candidates.
-- **`utils/image_meta.py`**: Robust metadata extraction for DICOM, NIfTI, TIFF stacks with medical imaging focus.
-- **`utils/tags.py`**: Control tags system for query refinement (`[EXCLUDE:tool1|tool2]`, `[NO_RERANK]`, `[REFINE]`).
-
-## Data Flow Patterns
-
-### Input Processing
-- Files validated via `utils/file_validator.py` (size limits, format checks)
-- Images converted to PNG previews for VLM via `utils/previews.py`
-- Metadata extracted preserving original format info (critical for format compatibility matching)
-- Format tokens added to retrieval query (e.g. `format:DICOM format:NIfTI`)
-
-### Retrieval Query Construction
-```python
-# Clean task text + format tokens from uploaded files
-query = f"{clean_task} format:{ext_tokens}"  # e.g. "segment lungs format:DICOM"
-```
-
-### VLM Selection Input
-The VLM receives:
-- **Text**: User task + candidate table + original file metadata
-- **Image**: PNG preview (converted from any format) 
-- **Metadata**: Original extension, dimensions, file info (crucial for IO compatibility)
-
-## Critical Patterns
-
-### Error Handling
-- **Graceful degradation**: If image conversion fails, continue text-only
-- **Robust metadata**: All metadata extraction wrapped in try/catch with sensible defaults
-- **File validation**: Early validation prevents downstream errors
-
-### Control Tags System
-Users can control behavior via tags in their queries:
-- `[EXCLUDE:toolname1|toolname2]` - Exclude specific tools from results
-- `[NO_RERANK]` - Skip CrossEncoder reranker (faster, less accurate)
-- `[REFINE]` - Force clarification turn for alternatives
-
-### Conversation Flow
-- **Complete**: Normal success with tool recommendations
-- **Needs Clarification**: VLM asks followup questions when task is ambiguous
-- **Terminal No-Tool**: No suitable tools found with explanation
-
-## Development Workflows
-
-### Running the App
-```bash
-# Install with pip using pyproject.toml
-pip install -e ".[dev]"
-
-# Configure .env with OPENAI_API_KEY and SOFTWARE_CATALOG path
-ai_agent ui  # Launches Gradio on port 7860
-```
-
-### Testing
-- **`tests/full_test.py`**: End-to-end pipeline tests driven by `tests/data/test_data.json`
-- Uses test doubles for VLM calls to avoid API costs
-- Run with: `pytest tests/`
-
-### Change Documentation
-- **`CHANGELOG.md`**: Follow [Keep a Changelog](https://keepachangelog.com/) format
-- Use semantic versioning with sections: Added, Changed, Deprecated, Removed, Fixed, Security
-- Update CHANGELOG.md for ALL user-facing changes before merging PRs
-- Format: `### Added\n- New feature description` under version heading
-- Version entries: `## [x.y.z] - YYYY-MM-DD`
-
-### Environment Management
-- **uv**: Fast Python package manager used in `tools/image/Dockerfile`
-- Creates isolated `.venv` environments for reproducible builds
-- Dockerfile uses `uv venv && uv pip install -e .` pattern for container builds
-
-### Logging & Debugging
-- Set `LOG_PROMPTS=1` to save VLM prompts + images to `logs/`
-- File logs in `logs/app_YYYYMMDD.log` with structured JSON events
-- Console/file log levels configurable via `.env`
-
-## Project Conventions
-
-### Schema Patterns
-- **Pydantic models** in `generator/schema.py` with robust field validation and aliasing for catalog compatibility
-- **Enum-based** conversation states and tool reasons for type safety
-- **Field normalization**: Dimensions (2D/3D/4D), modalities (CT/MRI/XR), file formats via validators
-
-### Catalog Integration
-- Software catalog in JSONL format following schema.org SoftwareSourceCode structure
-- **Runnable examples**: Links to HuggingFace Spaces, notebooks, web demos
-- **Supporting data**: Format compatibility info used for matching
-
-### Module Boundaries
-- `api/`: Pipeline orchestration, no UI dependencies
-- `generator/`: Pure VLM logic, no retrieval dependencies  
-- `retriever/`: Pure vector search, no generation dependencies
-- `utils/`: Shared utilities, no business logic
-- `ui/`: Gradio interface only
-
-### Configuration
-- Environment-based config via `.env` (API keys, model names, catalog paths)
-- Sensible defaults for all settings
-- No hardcoded paths or credentials
-
-## Medical Imaging Context
-
-This tool specializes in medical/scientific imaging:
-- **Modalities**: CT, MRI, X-ray, Ultrasound, PET, SPECT, Microscopy
-- **Formats**: DICOM, NIfTI, TIFF stacks, standard images
-- **Dimensions**: 2D images, 3D volumes, 4D timeseries
-- **Tasks**: Segmentation, registration, analysis, visualization
-
-The VLM selection considers format compatibility as a primary factor - tools supporting the user's input format are strongly preferred.
-
-## Security Notes
-- Only makes external calls to OpenAI VLM API (with user image preview)
-- Never uploads user data to third-party tool demos
-- Returns links only; user chooses whether to visit demos
+# AI Agent — Copilot Instructions
+
+This is a **RAG + VLM imaging tool recommender** that helps users find the right imaging software for their images and tasks. Users drop an image, describe their task, and get ranked software recommendations with demo links.
+
+## Architecture Overview
+
+The system follows a two-stage pipeline:
+
+1. **Retrieval Stage** (`retriever/`, `api/pipeline.py`): Fast text search using BGE-M3 embeddings + CrossEncoder reranker. No LLM calls. Returns top-K candidates.
+
+2. **Selection Stage** (`generator/`): Single VLM call (OpenAI GPT-4o/mini) that sees the image + candidates + metadata and returns ranked recommendations with accuracy scores.
+
+### Key Components
+
+- **`api/pipeline.RAGImagingPipeline`**: Main orchestrator. Handles file validation, metadata extraction, retrieval, and VLM selection.
+- **`retriever/embedders.py`**: FAISS vector index with BGE-M3 + CrossEncoder reranker for candidate retrieval.
+- **`generator/generator.VLMToolSelector`**: Vision-language model that selects best tools from candidates.
+- **`utils/image_meta.py`**: Robust metadata extraction for DICOM, NIfTI, TIFF stacks with medical imaging focus.
+- **`utils/tags.py`**: Control tags system for query refinement (`[EXCLUDE:tool1|tool2]`, `[NO_RERANK]`, `[REFINE]`).
+
+## Data Flow Patterns
+
+### Input Processing
+- Files validated via `utils/file_validator.py` (size limits, format checks)
+- Images converted to PNG previews for VLM via `utils/previews.py`
+- Metadata extracted preserving original format info (critical for format compatibility matching)
+- Format tokens added to retrieval query (e.g. `format:DICOM format:NIfTI`)
+
+### Retrieval Query Construction
+```python
+# Clean task text + format tokens from uploaded files
+query = f"{clean_task} format:{ext_tokens}"  # e.g. "segment lungs format:DICOM"
+```
+
+### VLM Selection Input
+The VLM receives:
+- **Text**: User task + candidate table + original file metadata
+- **Image**: PNG preview (converted from any format) 
+- **Metadata**: Original extension, dimensions, file info (crucial for IO compatibility)
+
+## Critical Patterns
+
+### Error Handling
+- **Graceful degradation**: If image conversion fails, continue text-only
+- **Robust metadata**: All metadata extraction wrapped in try/catch with sensible defaults
+- **File validation**: Early validation prevents downstream errors
+
+### Control Tags System
+Users can control behavior via tags in their queries:
+- `[EXCLUDE:toolname1|toolname2]` - Exclude specific tools from results
+- `[NO_RERANK]` - Skip CrossEncoder reranker (faster, less accurate)
+- `[REFINE]` - Force clarification turn for alternatives
+
+### Conversation Flow
+- **Complete**: Normal success with tool recommendations
+- **Needs Clarification**: VLM asks followup questions when task is ambiguous
+- **Terminal No-Tool**: No suitable tools found with explanation
+
+## Development Workflows
+
+### Running the App
+```bash
+# Install with pip using pyproject.toml
+pip install -e ".[dev]"
+
+# Configure .env with OPENAI_API_KEY and SOFTWARE_CATALOG path
+ai_agent ui  # Launches Gradio on port 7860
+```
+
+### Testing
+- **`tests/full_test.py`**: End-to-end pipeline tests driven by `tests/data/test_data.json`
+- Uses test doubles for VLM calls to avoid API costs
+- Run with: `pytest tests/`
+
+### Change Documentation
+- **`CHANGELOG.md`**: Follow [Keep a Changelog](https://keepachangelog.com/) format
+- Use semantic versioning with sections: Added, Changed, Deprecated, Removed, Fixed, Security
+- Update CHANGELOG.md for ALL user-facing changes before merging PRs
+- Format: `### Added\n- New feature description` under version heading
+- Version entries: `## [x.y.z] - YYYY-MM-DD`
+
+### Environment Management
+- **uv**: Fast Python package manager used in `tools/image/Dockerfile`
+- Creates isolated `.venv` environments for reproducible builds
+- Dockerfile uses `uv venv && uv pip install -e .` pattern for container builds
+
+### Logging & Debugging
+- Set `LOG_PROMPTS=1` to save VLM prompts + images to `logs/`
+- File logs in `logs/app_YYYYMMDD.log` with structured JSON events
+- Console/file log levels configurable via `.env`
+
+## Project Conventions
+
+### Schema Patterns
+- **Pydantic models** in `generator/schema.py` with robust field validation and aliasing for catalog compatibility
+- **Enum-based** conversation states and tool reasons for type safety
+- **Field normalization**: Dimensions (2D/3D/4D), modalities (CT/MRI/XR), file formats via validators
+
+### Catalog Integration
+- Software catalog in JSONL format following schema.org SoftwareSourceCode structure
+- **Runnable examples**: Links to HuggingFace Spaces, notebooks, web demos
+- **Supporting data**: Format compatibility info used for matching
+
+### Module Boundaries
+- `api/`: Pipeline orchestration, no UI dependencies
+- `generator/`: Pure VLM logic, no retrieval dependencies  
+- `retriever/`: Pure vector search, no generation dependencies
+- `utils/`: Shared utilities, no business logic
+- `ui/`: Gradio interface only
+
+### Configuration
+- Environment-based config via `.env` (API keys, model names, catalog paths)
+- Sensible defaults for all settings
+- No hardcoded paths or credentials
+
+## Medical Imaging Context
+
+This tool specializes in medical/scientific imaging:
+- **Modalities**: CT, MRI, X-ray, Ultrasound, PET, SPECT, Microscopy
+- **Formats**: DICOM, NIfTI, TIFF stacks, standard images
+- **Dimensions**: 2D images, 3D volumes, 4D timeseries
+- **Tasks**: Segmentation, registration, analysis, visualization
+
+The VLM selection considers format compatibility as a primary factor - tools supporting the user's input format are strongly preferred.
+
+## Security Notes
+- Only makes external calls to OpenAI VLM API (with user image preview)
+- Never uploads user data to third-party tool demos
+- Returns links only; user chooses whether to visit demos
 - Prompt logging is optional and local-only