Step 34: ASP.NET server — OpenAI-compatible API endpoints #84

Summary

Add an OpenAI-compatible API server to dotLLM via ASP.NET Minimal API endpoints in DotLLM.Server.

Roadmap Step: 34
Dependencies: Step 16 (Chat templates ✅), Step 17 (Streaming ✅), Step 42 (Tool calling ✅)

Acceptance Criteria

• /v1/chat/completions — non-streaming and SSE streaming, chat template formatting, tool calling
• /v1/completions — raw prompt completion, non-streaming and streaming
• /v1/models — list loaded model info
• /v1/tokenize — encode text to tokens (extension endpoint)
• /v1/detokenize — decode tokens to text (extension endpoint)
• /health and /ready — health and readiness probes
• Model loading at startup via --model CLI arg
• CPU and GPU device support via --device arg
• Sequential request processing via semaphore (batching is Step 35)
• OpenAI Python SDK compatibility verified
• response_format support (json_object, json_schema)
• tools / tool_choice support with tool call detection
• Request/response DTOs with snake_case JSON serialization

Key Files

New

• src/DotLLM.Server/Models/ — OpenAI-compatible request/response DTOs
• src/DotLLM.Server/Endpoints/ — Minimal API endpoint handlers
• src/DotLLM.Server/RequestConverter.cs — DTO ↔ engine type conversion
• src/DotLLM.Server/ServerState.cs — concurrency control, readiness
• src/DotLLM.Server/ServerOptions.cs — startup configuration

Modified

• src/DotLLM.Server/Program.cs — model loading, DI, endpoint registration
• src/DotLLM.Server/DotLLM.Server.csproj — add project references

References

• docs/SERVER.md