fix(openai): enhance API compatibility for frontend integration (Issue #113)

[Michael-A-Kuykendall]

Summary

• Enhances OpenAI API compatibility to resolve frontend integration issues
• Fixes Open WebUI and AnythingLLM detection/integration problems
• Adds explicit Content-Type headers and standardized response formats
• Resolves Issue Open WebUI and AnythingLLM integration compatibility issues #113: Open WebUI and AnythingLLM integration compatibility issues

Problem Fixed

Popular LLM frontends (Open WebUI, AnythingLLM) could not detect or properly integrate with shimmy despite CURL requests working fine. The issue was subtle compatibility gaps in response headers and model metadata format.

Technical Enhancements

🔧 Response Headers

• All JSON endpoints: Explicit Content-Type: application/json headers
• SSE Streaming: Proper text/event-stream, no-cache, keep-alive headers
• Error responses: Consistent headers for all error scenarios

📊 Enhanced Model Structure

• Optional fields: permission, root, parent for OpenAI spec compliance
• Stable timestamps: Fixed Jan 1, 2022 timestamp for consistent responses
• Proper serialization: Fields omitted when None for clean JSON

🚨 Improved Error Handling

• Consistent format: All errors follow OpenAI error specification
• Proper status codes: 404 for model not found, 500 for server errors
• Detailed messages: Helpful error information for debugging

📡 Streaming Compatibility

• Complete headers: All required SSE headers for frontend streaming
• Proper format: OpenAI-compatible chunk format maintained

Changes Made

Core API Responses (src/openai_compat.rs)

// Before: Implicit headers, minimal model fields
Json(ModelsResponse { ... })

// After: Explicit headers, complete model structure
(
    [(header::CONTENT_TYPE, "application/json")],
    Json(ModelsResponse { ... })
)

Enhanced Model Structure

#[derive(Debug, Serialize, Deserialize)]
pub struct Model {
    pub id: String,
    pub object: String,
    pub created: u64,
    pub owned_by: String,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub permission: Option<Vec<serde_json::Value>>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub root: Option<String>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub parent: Option<String>,
}

Comprehensive Error Responses

let error_response = serde_json::json!({
    "error": {
        "message": format!("Model '{}' not found. Available models: {:?}", req.model, available_models),
        "type": "invalid_request_error",
        "param": "model",
        "code": "model_not_found"
    }
});

Test Coverage

• ✅ test_enhanced_model_structure(): Validates optional field serialization
• ✅ test_frontend_error_response_format(): Error format compatibility
• ✅ test_frontend_models_endpoint_response(): Models endpoint structure
• ✅ test_open_webui_anythingllm_compatibility(): Frontend-specific tests
• ✅ 27 OpenAI compatibility tests passing

Frontend Impact

• 🎯 Open WebUI: Should now detect shimmy as valid model provider
• 🎯 AnythingLLM: Should automatically recognize shimmy endpoints
• 🎯 General: Any OpenAI-compatible frontend gets better compatibility
• 🎯 Streaming: SSE responses work correctly with frontend clients

Validation

The changes maintain 100% backward compatibility while adding the missing pieces that frontends require for proper integration. All existing functionality preserved.

🤖 Generated with Claude Code

[Michael-A-Kuykendall]

Closing this PR as the clean version was merged in PR #123 (commit 37f67ac). This branch had contaminated git history with unsigned commits.