Created dockerfile + updated streaming

.gitignore

@@ -328,4 +328,5 @@ tests/__pycache__/
 docs/start.md
 docs/typescript-translation-plan.md
 
-# Note: test scripts in tests/ directory should be tracked in git
+# Note: test scripts in tests/ directory should be tracked in git
+fix_env.patch

CLAUDE.md

@@ -0,0 +1,150 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+This is an OpenAI-compatible API gateway for Claude Code CLI. It wraps the Claude Code CLI and exposes its functionality through a FastAPI server with OpenAI-compatible endpoints, enabling integration with tools like Cursor, Cline, Roo Code, and Open WebUI.
+
+## Core Architecture
+
+### Request Flow
+1. Client sends OpenAI-compatible request to `/v1/chat/completions`
+2. FastAPI endpoint validates and parses the request
+3. `ClaudeManager` spawns a Claude Code CLI process with the prompt
+4. Claude Code CLI runs to completion and outputs stream-json format
+5. Output is parsed and converted to OpenAI-compatible format
+6. Response is returned as streaming SSE or non-streaming JSON
+
+### Key Components
+
+**ClaudeProcess (claude_manager.py:18-228)**
+- Manages a single Claude Code CLI process
+- Spawns process with `--output-format stream-json --dangerously-skip-permissions`
+- Runs from `src/` directory to use existing Claude Code authentication
+- Process runs to completion (not interactive) and all output is captured at once
+- Output is parsed line-by-line as JSON messages and queued
+
+**ClaudeManager (claude_manager.py:230-343)**
+- Creates and manages multiple Claude Code processes
+- Enforces concurrent session limits
+- Does NOT store completed processes to avoid "max sessions" errors
+- Each request creates a fresh process
+
+**Chat Endpoint (api/chat.py:30-302)**
+- Handles `/v1/chat/completions` requests
+- Extracts user prompt from last user message
+- Creates session and project directory
+- Spawns Claude process and handles streaming/non-streaming responses
+- Updates session manager with token usage
+
+**Streaming (utils/streaming.py)**
+- `OpenAIStreamConverter` converts Claude's stream-json output to OpenAI SSE format
+- Looks for `{"type":"assistant","message":{"content":[{"type":"text","text":"..."}]}}` messages
+- Extracts text content and emits as SSE chunks
+- Sends `data: [DONE]` to complete stream
+
+**Response Creation (utils/streaming.py:331-431)**
+- For non-streaming: collects all Claude messages and extracts assistant content
+- Handles both array format `[{"type":"text","text":"..."}]` and string format
+- Always ensures content is non-empty with fallback messages
+- Returns OpenAI-compatible response with basic usage stats
+
+### Critical Implementation Details
+
+1. **Claude CLI Invocation**: Always use exact flags `--output-format stream-json --verbose --dangerously-skip-permissions` and run from the `src/` directory where Claude Code is authenticated
+
+2. **Process Lifecycle**: Claude Code CLI completes immediately and is not interactive. All output is captured at once via `communicate()`. Don't store processes after completion.
+
+3. **Session IDs**: Claude Code generates its own session ID in the first message. Extract and use this ID instead of the generated one.
+
+4. **Output Parsing**: Parse each line as JSON. Look for `type` field to identify message types (assistant, result, error, etc.)
+
+5. **Content Extraction**: Assistant messages have nested structure: `message.content[]` where each item has `type:"text"` and `text` field
+
+6. **Authentication**: Uses existing Claude Code authentication in working directory. No API key configuration in code.
+
+## Common Commands
+
+### Development
+```bash
+make install          # Install dependencies
+make start            # Start dev server with reload
+make start-prod       # Start production server
+make test             # Run pytest tests
+make test-real        # Run end-to-end curl tests
+make clean            # Clean Python cache
+```
+
+### Testing Individual Components
+```bash
+# Test Claude Code CLI directly
+claude -p "hello" --output-format stream-json --dangerously-skip-permissions
+
+# Test API endpoint
+curl -X POST http://localhost:8000/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{"model":"claude-3-5-haiku-20241022","messages":[{"role":"user","content":"hello"}]}'
+
+# Test with streaming
+curl -X POST http://localhost:8000/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{"model":"claude-3-5-haiku-20241022","messages":[{"role":"user","content":"hello"}],"stream":true}'
+```
+
+### Port Management
+```bash
+make kill PORT=8000   # Kill process on specific port
+```
+
+## Supported Claude Models
+
+- `claude-opus-4-20250514` - Claude Opus 4 (Most powerful)
+- `claude-sonnet-4-20250514` - Claude Sonnet 4 (Latest Sonnet)
+- `claude-3-7-sonnet-20250219` - Claude Sonnet 3.7 (Advanced)
+- `claude-3-5-haiku-20241022` - Claude Haiku 3.5 (Fast, default)
+
+Models are mapped in `models/claude.py` and validated before use.
+
+## Configuration
+
+Key settings in `core/config.py`:
+- `claude_binary_path`: Auto-detected from PATH or npm global install
+- `project_root`: `/tmp/claude_projects` - workspace for Claude processes
+- `default_model`: `claude-3-5-haiku-20241022`
+- `max_concurrent_sessions`: 10
+- `streaming_timeout_seconds`: 300
+
+## Limitations
+
+- Claude Code has ~25k token input limit (less than normal API)
+- Context auto-compacts beyond 100k
+- Runs in bypass mode to avoid tool permission prompts
+- Linux/Mac only (use WSL on Windows)
+- Claude Code must be authenticated in the current directory before starting API
+
+## Docker Deployment
+
+Uses Ubuntu 22.04 base with Node.js 18+ for Claude Code CLI. Runs as non-root `claudeuser`. Two authentication modes:
+
+1. **API Key**: Set `ANTHROPIC_API_KEY` env var
+2. **Claude Max**: Set `USE_CLAUDE_MAX=true` and authenticate interactively via `docker exec -it claude-code-api claude`
+
+## OpenAI Compatibility
+
+Implements subset of OpenAI Chat Completions API:
+- POST `/v1/chat/completions` - Create completion (streaming and non-streaming)
+- GET `/v1/models` - List available models
+- GET `/health` - Health check
+
+Extension fields:
+- `project_id`: Claude Code project context
+- `session_id`: Continue existing conversation
+
+## Debugging Tips
+
+- Check logs for "DEBUG: Claude stderr" and "DEBUG: Claude stdout" messages
+- Use `/v1/chat/completions/debug` endpoint to test request validation
+- Verify Claude Code works: `claude --version` and `claude -p "test" --output-format stream-json`
+- Ensure Claude Code is authenticated: run `claude` in the project directory first
+- Check that `src/` directory exists and is accessible