Add OpenAI provider support

AGENTS.md

@@ -1,6 +1,6 @@
 # Inspamity — Agent Guidelines
 
-AI-powered spam detection tool that integrates with rspamd using the Anthropic Claude API.
+AI-powered spam detection tool that integrates with rspamd using AI providers (Anthropic Claude, OpenAI).
 
 ## Project Structure
 
@@ -9,7 +9,10 @@ AI-powered spam detection tool that integrates with rspamd using the Anthropic C
 ├── cli_toolbox.py                 # CLI tool for testing emails
 ├── email_utils/
 │   ├── config.py                  # Shared config loader (system/local)
-│   ├── anthropic_spam_check.py    # Anthropic API integration
+│   ├── prompts.py                 # Shared SYSTEM_PROMPT constant
+│   ├── ai_spam_check.py           # Provider dispatcher (check_spam_with_ai)
+│   ├── anthropic_spam_check.py    # Anthropic provider implementation
+│   └── openai_spam_check.py       # OpenAI provider implementation
 │   └── process_email.py           # Email parsing and formatting
 ├── rspamd/
 │   └── external_ai_test.lua       # rspamd Lua plugin
@@ -35,19 +38,27 @@ pip install -e ".[dev]"
 - **Quotes**: double quotes (enforced by ruff)
 - **Imports**: sorted by ruff (`I` rule), stdlib → third-party → local
 
+## Before Submitting a PR
+
+- Run `ruff check . && ruff format --check . && pytest` — all must pass
+- Check if `README.md` needs updates to reflect your changes (new features, changed config, updated commands)
+
 ## Testing
 
 - **Framework**: pytest 9.0.3
 - **Run tests**: `pytest -v`
 - **All changes must pass**: `ruff check . && ruff format --check . && pytest`
 - Tests live in `tests/` and mirror the module structure
-- Mock external APIs — never make real API calls in tests. Use `unittest.mock.patch` for the Anthropic client and `monkeypatch` for config paths
+- Mock external APIs — never make real API calls in tests. Use `unittest.mock.patch` for API clients and `monkeypatch` for config paths
 - Use `tmp_path` for temporary files, `monkeypatch` for config isolation
 
 ## Architecture Notes
 
+- **Provider dispatch**: `ai_spam_check.py` reads the `provider` setting from config and dispatches to the appropriate provider module. Provider imports are lazy (inside if/elif branches) so only the selected SDK is loaded.
+- **Adding a new provider**: Create `email_utils/<provider>_spam_check.py` with a `check_spam_with_<provider>(email_content: str) -> dict[str, Any]` function, add an elif branch in `ai_spam_check.py`, and add the config section.
+- **System prompt** is shared across providers via `email_utils/prompts.py`. Do not duplicate it in provider modules.
 - **Config loading** is centralized in `email_utils/config.py`. Don't duplicate config logic elsewhere. Config is read from `/etc/inspamity/config.ini` (system) or `config.ini` in the project root (local), in that priority order.
 - **Entry points** (`email_ai_interface.py`, `cli_toolbox.py`) live at the project root because the rspamd Lua plugin references them by absolute path (`/usr/local/inspamity/`). Do not move them into a package.
 - **Production deployment** uses a venv at `/usr/local/inspamity/.venv/` — the Lua script calls that venv's Python directly.
-- The Anthropic API **temperature** parameter is only passed when explicitly set in config. Do not hardcode it.
-- `check_spam_with_ai()` must always return a dict with `is_spam`, `confidence`, and `reason` keys, even on error.
+- The **temperature** parameter is only passed to the API when explicitly set in config. Do not hardcode it.
+- All provider functions must always return a dict with `is_spam`, `confidence`, and `reason` keys, even on error.

README.md

@@ -93,6 +93,37 @@ After this, enable debugging in the config.ini
 
 ## ⚙️ Configuration
 
+Copy `config.ini.default` to `config.ini` (or `/etc/inspamity/config.ini` for system-wide) and edit it:
+
+```ini
+[settings]
+# AI provider: anthropic or openai
+provider = anthropic
+
+[anthropic]
+api_key = your_api_key_here
+model = claude-haiku-4-5-latest
+temperature = 0.0
+timeout = 20.0
+
+[openai]
+api_key = your_api_key_here
+model = gpt-5.4-mini
+temperature = 0.0
+timeout = 20.0
+```
+
+### Supported Providers
+
+| Provider | Default Model | Config Section |
+|----------|--------------|----------------|
+| Anthropic | `claude-haiku-4-5-latest` | `[anthropic]` |
+| OpenAI | `gpt-5.4-mini` | `[openai]` |
+
+Set `provider` in `[settings]` to switch between them. Only the selected provider's API key is required.
+
+### rspamd Integration
+
 By default, the rspamd Lua script is configured to:
 - Run after all other checks (type postfilter)
 - Skip emails already marked as spam

cli_toolbox.py

@@ -43,7 +43,7 @@ def main() -> None:
         # Check for spam using AI if requested
         if args.check_ai:
             try:
-                from email_utils.anthropic_spam_check import check_spam_with_ai
+                from email_utils.ai_spam_check import check_spam_with_ai
 
                 # Send formatted email content to AI
                 result = check_spam_with_ai(output)

config.ini.default

@@ -1,12 +1,21 @@
+[settings]
+# AI provider to use: anthropic or openai
+provider = anthropic
+# Enable debug mode to save copies of processed emails and AI output
+debug_mode = false
+# Directory where debug files will be stored
+debug_directory = /var/local/inspamity
+
 [anthropic]
 api_key = your_api_key_here
 model = claude-haiku-4-5-latest
 # Temperature for AI responses (0.0 = deterministic). Remove to use model default.
 temperature = 0.0
 timeout = 20.0
 
-[settings]
-# Enable debug mode to save copies of processed emails and AI output
-debug_mode = false
-# Directory where debug files will be stored
-debug_directory = /var/local/inspamity
+[openai]
+api_key = your_api_key_here
+model = gpt-5.4-mini
+# Temperature for AI responses (0.0 = deterministic). Remove to use model default.
+temperature = 0.0
+timeout = 20.0

email_ai_interface.py

@@ -5,7 +5,7 @@
 from datetime import datetime
 from pathlib import Path
 
-from email_utils.anthropic_spam_check import check_spam_with_ai
+from email_utils.ai_spam_check import check_spam_with_ai
 from email_utils.config import load_config
 from email_utils.process_email import format_email_content, get_email_content
 

email_utils/ai_spam_check.py

@@ -0,0 +1,35 @@
+from typing import Any
+
+from email_utils.config import load_config
+
+
+def check_spam_with_ai(email_content: str) -> dict[str, Any]:
+    """Check if an email is spam using the configured AI provider.
+
+    Dispatches to the appropriate provider based on the 'provider'
+    setting in config.ini (defaults to 'anthropic').
+    """
+    try:
+        config = load_config()
+        provider = config.get("settings", "provider", fallback="anthropic")
+
+        if provider == "anthropic":
+            from email_utils.anthropic_spam_check import check_spam_with_anthropic
+
+            return check_spam_with_anthropic(email_content)
+        elif provider == "openai":
+            from email_utils.openai_spam_check import check_spam_with_openai
+
+            return check_spam_with_openai(email_content)
+        else:
+            return {
+                "is_spam": "no",
+                "confidence": 0,
+                "reason": f"Unknown AI provider: {provider}",
+            }
+    except Exception as e:
+        return {
+            "is_spam": "no",
+            "confidence": 0,
+            "reason": f"Error checking spam with AI: {e}",
+        }

email_utils/anthropic_spam_check.py

@@ -4,26 +4,10 @@
 import anthropic
 
 from email_utils.config import load_config
+from email_utils.prompts import SYSTEM_PROMPT
 
-SYSTEM_PROMPT = (
-    "You are a spam detection system. Analyze this email and classify it "
-    "as spam or not. Note that legitimate newsletters are not spam.\n\n"
-    "The attached email contains all headers, but is stripped from HTML "
-    "and attachments. It is also truncated if it's too long. At the end "
-    "it contains a summary of attachments, images and links that were "
-    "in the email.\n\n"
-    "Provide your analysis in JSON format with the following structure:\n"
-    "{\n"
-    '  "is_spam": "yes|no",\n'
-    '  "confidence": 0-100,\n'
-    '  "reason": "brief explanation of key factors that led to this '
-    'classification"\n'
-    "}\n\n"
-    "Only output this JSON. Do not output anything else!"
-)
 
-
-def check_spam_with_ai(email_content: str) -> dict[str, Any]:
+def check_spam_with_anthropic(email_content: str) -> dict[str, Any]:
     """Use Anthropic Claude API to check if an email is spam.
 
     Args:
@@ -62,8 +46,3 @@ def check_spam_with_ai(email_content: str) -> dict[str, Any]:
             "confidence": 0,
             "reason": f"Error checking spam with AI: {e}",
         }
-
-
-if __name__ == "__main__":
-    test_content = "From: test@example.com\nSubject: Test\n\nThis is a test email"
-    print(check_spam_with_ai(test_content))

email_utils/openai_spam_check.py

@@ -0,0 +1,50 @@
+import json
+from typing import Any
+
+from openai import OpenAI
+
+from email_utils.config import load_config
+from email_utils.prompts import SYSTEM_PROMPT
+
+
+def check_spam_with_openai(email_content: str) -> dict[str, Any]:
+    """Use OpenAI API to check if an email is spam.
+
+    Args:
+        email_content: Processed and formatted email content.
+
+    Returns:
+        AI analysis result with is_spam, confidence, and reason.
+    """
+    try:
+        config = load_config()
+
+        api_key = config.get("openai", "api_key")
+        model = config.get("openai", "model", fallback="gpt-5.4-mini")
+        timeout = config.getfloat("openai", "timeout", fallback=20.0)
+
+        client = OpenAI(api_key=api_key, timeout=timeout)
+
+        kwargs: dict[str, Any] = {
+            "model": model,
+            "max_completion_tokens": 100,
+            "messages": [
+                {"role": "system", "content": SYSTEM_PROMPT},
+                {"role": "user", "content": email_content},
+            ],
+        }
+
+        # Only pass temperature if explicitly configured
+        if config.has_option("openai", "temperature"):
+            kwargs["temperature"] = config.getfloat("openai", "temperature")
+
+        response = client.chat.completions.create(**kwargs)
+
+        return json.loads(response.choices[0].message.content)
+
+    except Exception as e:
+        return {
+            "is_spam": "no",
+            "confidence": 0,
+            "reason": f"Error checking spam with AI: {e}",
+        }

email_utils/prompts.py

@@ -0,0 +1,24 @@
+SYSTEM_PROMPT = (
+    "You are a spam detection system. Classify the following email as spam or not spam.\n\n"
+    "Spam includes: unsolicited commercial email, phishing, scam messages, "
+    "malware/link bait, and deceptive bulk mail.\n\n"
+    "Not spam includes: personal correspondence, legitimate newsletters and "
+    "mailing lists the recipient likely subscribed to, transactional emails "
+    "(order confirmations, password resets, shipping notifications), and "
+    "automated notifications from known services.\n\n"
+    "The email below has been pre-processed: HTML has been converted to plain text, "
+    "attachments have been removed, and the body may be truncated. A metadata summary "
+    "(images, links, attachments, DKIM/SPF results) is included at the top. "
+    "Note: long URLs and lists may be truncated by our preprocessing "
+    "(marked with [TRUNCATED]) — this is normal and not a spam indicator.\n\n"
+    "Useful signals to consider:\n"
+    "- DKIM/SPF failures on emails claiming to be from well-known organizations\n"
+    "- Suspicious URL domains (e.g. misspelled brands, free hosting sites), "
+    "especially combined with urgency or threats\n"
+    "- Pressure tactics (act now, account suspended, verify immediately)\n"
+    "- Mismatch between the claimed sender identity and the email content/headers\n\n"
+    "Respond with only this JSON object, no other text or markdown:\n"
+    '{"is_spam": "yes" or "no", "confidence": 0-100, "reason": "brief explanation"}\n\n'
+    "Confidence guide: 0-20 very unlikely spam, 30-50 uncertain, "
+    "60-80 likely spam, 85-100 clearly spam."
+)

pyproject.toml

@@ -16,6 +16,7 @@ authors = [
 dependencies = [
     "anthropic>=0.90,<1.0",
     "beautifulsoup4>=4.13",
+    "openai>=2.0,<3.0",
 ]
 
 [project.optional-dependencies]

tests/test_ai_spam_check.py

@@ -0,0 +1,55 @@
+from unittest.mock import patch
+
+from email_utils import config as config_module
+from email_utils.ai_spam_check import check_spam_with_ai
+
+
+def _mock_config(tmp_path, monkeypatch, provider="anthropic"):
+    """Helper to set up a mock config file with provider selection."""
+    lines = f"[settings]\nprovider = {provider}\n"
+    lines += "[anthropic]\napi_key = test\n"
+    lines += "[openai]\napi_key = test\n"
+    config_file = tmp_path / "config.ini"
+    config_file.write_text(lines)
+    monkeypatch.setattr(config_module, "SYSTEM_CONFIG_PATH", tmp_path / "nonexistent.ini")
+    monkeypatch.setattr(config_module, "LOCAL_CONFIG_PATH", config_file)
+
+
+class TestAiSpamCheckDispatcher:
+    @patch("email_utils.anthropic_spam_check.check_spam_with_anthropic")
+    def test_dispatches_to_anthropic(self, mock_anthropic, tmp_path, monkeypatch):
+        _mock_config(tmp_path, monkeypatch, provider="anthropic")
+        mock_anthropic.return_value = {"is_spam": "no", "confidence": 0, "reason": "test"}
+
+        result = check_spam_with_ai("test email")
+        mock_anthropic.assert_called_once_with("test email")
+        assert result["is_spam"] == "no"
+
+    @patch("email_utils.openai_spam_check.check_spam_with_openai")
+    def test_dispatches_to_openai(self, mock_openai, tmp_path, monkeypatch):
+        _mock_config(tmp_path, monkeypatch, provider="openai")
+        mock_openai.return_value = {"is_spam": "yes", "confidence": 80, "reason": "spam"}
+
+        result = check_spam_with_ai("test email")
+        mock_openai.assert_called_once_with("test email")
+        assert result["is_spam"] == "yes"
+
+    def test_unknown_provider_returns_error(self, tmp_path, monkeypatch):
+        _mock_config(tmp_path, monkeypatch, provider="unknown_provider")
+
+        result = check_spam_with_ai("test email")
+        assert result["is_spam"] == "no"
+        assert result["confidence"] == 0
+        assert "Unknown AI provider" in result["reason"]
+
+    def test_defaults_to_anthropic_when_no_provider_set(self, tmp_path, monkeypatch):
+        """When provider is not in config, defaults to anthropic."""
+        config_file = tmp_path / "config.ini"
+        config_file.write_text("[settings]\n[anthropic]\napi_key = test\n")
+        monkeypatch.setattr(config_module, "SYSTEM_CONFIG_PATH", tmp_path / "nonexistent.ini")
+        monkeypatch.setattr(config_module, "LOCAL_CONFIG_PATH", config_file)
+
+        with patch("email_utils.anthropic_spam_check.check_spam_with_anthropic") as mock_anthropic:
+            mock_anthropic.return_value = {"is_spam": "no", "confidence": 0, "reason": "test"}
+            check_spam_with_ai("test")
+            mock_anthropic.assert_called_once()