Endpoints Convertion

[PixelShober]

Nele.AI bietet folgende Endpoints an: https://api.aieva.io/docs#/
Ich möchte nun diese auch in Codex CLI oder Cursor etc nutzen.
Somit müssen eine Übersetzungschicht zwischen den Endpoints hergestellt werden.

Ich möchte ein fertiges Programm zum ausführen (.exe) und ungern Bibliothek etc für den Endnutzer manuell einbinden also würde ich eher in Richtung C# anstatt Python tendieren, wenn du dies für sinnvoll erachtest aber gib mir dazu gerne nochmal deine Meinung.

[PixelShober]

Status-Update (Endpoints Convertion)

Implementiert:

• OpenAI-kompatible Endpoints: /v1/models, /v1/chat/completions (sync + stream), /v1/responses (sync), /v1/audio/transcriptions, /v1/images/generations
• Konfiguration via appsettings.json/appsettings.Local.json (BaseUrl, ApiKey, DefaultChatModel, Urls), wird immer aus dem EXE-Ordner geladen
• Default-Model-Fallback, wenn im Request kein model gesetzt ist
• Root-Endpoint zeigt "API not provided" wenn kein Key konfiguriert ist
• README mit Anforderungen, Setup und Beispielen

Commits:

• 898062d
• 2912af7
• a8cdc35

[PixelShober]

Plan\n\nApproach\n- Inspect /v1/chat/completions streaming path and current SSE formatting/headers.\n- Reproduce Cline 422 with a minimal streamed request; compare with OpenAI-compatible format.\n- Adjust request/response mapping to ensure valid streaming chunks and proper error handling.\n- Add/update tests for streaming behavior if feasible.\n\nFiles likely to change\n- Proxy/* (controller/endpoint handlers for chat completions)\n- Proxy/* (streaming helpers / DTOs)\n- Tests/* (streaming-related tests)\n\nRisks\n- Breaking existing non-streaming responses or model routing.\n- Client expectations (Cline) differ from current stream format.\n\nTest plan\n- dotnet restore\n- dotnet build\n- dotnet test\n- Manual: run proxy, send /v1/chat/completions with stream=true, verify SSE event chunks + [DONE] and Cline connection works.

[PixelShober]

✅ Understood
❓ Questions

1. Can you paste the exact request Cline sends to /v1/chat/completions (headers + JSON body), or a minimal curl that reproduces the 422?
2. Which upstream model/provider is configured in appsettings (BaseUrl/ApiKey/DefaultChatModel), and does that upstream support streaming for this model?
3. Should the streaming response strictly follow OpenAI SSE format (data: {chunk} + data: [DONE]) or a different variant expected by Cline?

🧠 Assumptions I can make (only if I approve)
I can adjust the streaming format/headers to match OpenAI SSE conventions used by Cline.
I can add/adjust tests for streaming behavior.

⛔ Blocked until reply

[PixelShober]

1. Please search this up yourself in the docu of cline or another code agent for visual studio
2. I configured Claude Sonnet 4.5, but I want to make permanent streaming an option to enable
3. I really want to bring Nele AI API as close to Open AI API as possible so I can use it as an Open AI Custom Provider in any Code agent this is the most important goal.

[PixelShober]

Plan (Tools limit fix)\n\nApproach\n- Normalize tools in /v1/chat/completions and trim function.description to Nele limit (default 1000).\n- Add config to override max length.\n- Log count of trimmed tool descriptions.\n\nFiles likely to change\n- Proxy/Program.cs\n- Proxy/appsettings.json\n- README.md\n\nRisks\n- Over-trimming tool descriptions could reduce tool call quality.\n\nTest plan\n- dotnet restore\n- dotnet build\n- dotnet test\n- Manual: send tools with >1000 description and verify 200 response.

[PixelShober]

Kurzupdate zu #1:

• Was/Warum: Tools-Description-Trim (<=1000) fuer Copilot/Nele, Knowledge-Endpunkte + Image-Attachments + web_search_results Mapping + Logging
• Key Files: Proxy/Program.cs, Proxy/appsettings.json, README.md, Clients/NeleDesktop/MainWindow.xaml, Tests/NeleDesktop.UITests/UiBehaviorTests.cs
• Tests: dotnet restore .\NeleOpenAIProxy.sln; dotnet build .\NeleOpenAIProxy.sln -c Release; dotnet test .\NeleOpenAIProxy.sln -c Release
• Functional: n/a (API-Smoketest empfohlen)
• Risiko/Rollback: gering; Rollback per Revert von Commit 9edcb86
• PR: Add ChatGPT-style voice transcription and image attachments #6

[PixelShober]

Plan (kurz):\n- Rollen aus chat/completions loggen (nur Rollenliste).\n- Ungueltige Rollen auf erlaubte Werte mappen (developer->system, tool/function->assistant).\n- Tests: dotnet restore/build/test (Release).

[PixelShober]

Kurzupdate zu #1:

• Was/Warum: Rollenliste im Proxy loggen + Rollen normalisieren (developer->system, tool/function->assistant) zur Vermeidung von 422
• Key Files: Proxy/Program.cs
• Tests: dotnet restore .\NeleOpenAIProxy.sln; dotnet build .\NeleOpenAIProxy.sln -c Release; dotnet test .\NeleOpenAIProxy.sln -c Release
• Risiko/Rollback: gering; Rollback per Revert von Commit 9c5913c
• PR: Endpoints Convertion #7