diff --git a/CLAUDE.md b/CLAUDE.md
new file mode 100644
index 0000000..e90eda0
--- /dev/null
+++ b/CLAUDE.md
@@ -0,0 +1,94 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+VLM Run Python SDK - Official Python SDK for the VLM Run API platform. Provides structured data extraction from images, documents, videos, and audio using vision-language models.
+
+## Development Commands
+
+```bash
+# Install for development
+pip install -e ".[test]"
+
+# Run all tests
+pytest -sv tests
+
+# Run a single test file
+pytest -sv tests/test_client.py
+
+# Run a specific test
+pytest -sv tests/test_client.py::test_client_init
+
+# Lint and format code
+make lint
+# Or directly: pre-commit run --all-files
+
+# Build package
+make dist
+```
+
+## Architecture
+
+### Client Structure (`vlmrun/client/`)
+
+The SDK uses a resource-based pattern where `VLMRun` is the main client that exposes specialized resources:
+
+- **VLMRun** (`client.py`) - Main entry point, initializes all resources on construction after validating the API key via `/health` endpoint
+- **APIRequestor** (`base_requestor.py`) - Handles HTTP requests with retry logic (tenacity), timeout handling, and error mapping to custom exception types
+- **Resource Classes** - Each API domain has its own resource class:
+  - `ImagePredictions`, `DocumentPredictions`, `VideoPredictions`, `AudioPredictions` (`predictions.py`)
+  - `Agent` (`agent.py`) - Also provides OpenAI-compatible completions interface via `client.agent.completions`
+  - `Files`, `Hub`, `Models`, `Datasets`, `Finetuning`, `Feedback`, `Executions`, `Artifacts`
+
+### Predictions Pattern
+
+`FilePredictions(route)` is a factory function that creates prediction classes for document/audio/video. These inherit from `SchemaCastMixin` which handles automatic type casting of responses to Pydantic models based on domain schemas.
+
+Key methods: `generate()` (use domain), `execute()` (use named model), `schema()` (auto-generate schema).
+
+### CLI Structure (`vlmrun/cli/`)
+
+Built with Typer. Entry point is `vlmrun/cli/cli.py`. Subcommands are in `vlmrun/cli/_cli/`:
+- `chat.py` - Interactive chat with Orion agent
+- `config.py` - API key and base URL configuration
+- `generate.py`, `files.py`, `predictions.py`, `hub.py`, `models.py`
+
+### Type System (`vlmrun/client/types.py`)
+
+All API responses are typed with Pydantic models. Key types:
+- `PredictionResponse` - Standard response for all predictions
+- `GenerationConfig` - Configuration for generation (prompt, response_model, json_schema, temperature, etc.)
+- `MarkdownDocument/MarkdownPage/MarkdownTable` - Structured document extraction results
+
+### Common Utilities (`vlmrun/common/`)
+
+- `image.py` - Image encoding, EXIF handling
+- `video.py` - Video frame extraction
+- `pdf.py` - PDF processing
+- `utils.py` - Helper functions like `remote_image()`
+
+## Configuration
+
+Environment variables:
+- `VLMRUN_API_KEY` - API key (required)
+- `VLMRUN_BASE_URL` - Override default API URL (default: `https://api.vlm.run/v1`)
+- `VLMRUN_CACHE_DIR` - Override cache directory (default: `~/.vlmrun/cache`)
+
+## Optional Dependencies
+
+Install based on needed functionality:
+- `pip install vlmrun[cli]` - CLI with Typer/Rich
+- `pip install vlmrun[video]` - Video processing (numpy)
+- `pip install vlmrun[doc]` - PDF processing (pypdfium2)
+- `pip install vlmrun[openai]` - OpenAI SDK for chat completions API
+- `pip install vlmrun[all]` - All optional dependencies
+
+## Testing
+
+Tests use pytest with fixtures in `tests/conftest.py`. Tests require `VLMRUN_API_KEY` environment variable. Test files mirror the client structure (e.g., `test_predictions.py`, `test_agent.py`).
+
+## Code Style
+
+Uses ruff for linting and black for formatting. Pre-commit hooks enforce style on commit. Configuration is in `pyproject.toml` and `.pre-commit-config.yaml`.
diff --git a/vlmrun/cli/cli.py b/vlmrun/cli/cli.py
index 5d3139b..002bda7 100644
--- a/vlmrun/cli/cli.py
+++ b/vlmrun/cli/cli.py
@@ -117,7 +117,8 @@ def main(
     if ctx.resilient_parsing or is_help_requested():
         return
 
-    if ctx.invoked_subcommand is not None:
+    # Skip credential check for config commands (needed to set API key)
+    if ctx.invoked_subcommand is not None and ctx.invoked_subcommand != "config":
         check_credentials(ctx, api_key, base_url)
         ctx.obj = VLMRun(api_key=api_key, base_url=base_url)
 
diff --git a/vlmrun/version.py b/vlmrun/version.py
index 7225152..43a1e95 100644
--- a/vlmrun/version.py
+++ b/vlmrun/version.py
@@ -1 +1 @@
-__version__ = "0.5.2"
+__version__ = "0.5.3"