%YAML 1.1
# Continue (VS Code) – Konfiguration für den zentralen LocalAI-Server
# Ziel:   ~/.continue/config.yaml   (Windows: %USERPROFILE%\.continue\config.yaml)
# Secret: ~/.continue/.env  mit  LOCALAI_API_KEY=<Key aus der .env des Servers>
#
# apiBase je nach Standort anpassen (nur EINE Zeile unten aktiv lassen):
#   auf dem Server selbst:  http://127.0.0.1:8800/v1
#   im Heim-LAN:            http://192.168.178.72:8800/v1
#   über NetBird:           http://100.73.82.219:8800/v1   (oder NetBird-DNS-Name)

---
name: LocalAI
version: 1.0.0
schema: v1

# gemeinsame Verbindungsdaten (YAML-Anker, braucht den %YAML-1.1-Header oben)
model_defaults: &localai
  provider: openai
  apiBase: http://192.168.178.72:8800/v1
  apiKey: ${{ secrets.LOCALAI_API_KEY }}
  requestOptions:
    timeout: 1800   # Sekunden; erster Request lädt ggf. ein Modell

models:
  # --- Chat / Agent / Edit ------------------------------------------------
  - name: Qwen3.5 35B-A3B
    <<: *localai
    model: qwen3.5-35b-a3b
    roles: [chat, edit, apply, summarize]
    capabilities: [tool_use]
    defaultCompletionOptions:
      contextLength: 122880
      maxTokens: 8192

  - name: GLM-4.7 Flash
    <<: *localai
    model: glm-4.7-flash
    roles: [chat, edit, apply]
    capabilities: [tool_use]
    defaultCompletionOptions:
      contextLength: 122880
      maxTokens: 8192

  - name: Qwen3.5 9B (schnell)
    <<: *localai
    model: qwen3.5-9b
    roles: [chat, edit, apply]
    capabilities: [tool_use]
    defaultCompletionOptions:
      contextLength: 28672
      maxTokens: 4096

  # --- Tab-Autocomplete (Fill-in-the-Middle) ------------------------------
  - name: Qwen2.5-Coder 1.5B (Autocomplete)
    <<: *localai
    model: qwen2.5-coder-1.5b
    roles: [autocomplete]
    useLegacyCompletionsEndpoint: true   # /v1/completions statt Chat
    promptTemplates:
      autocomplete: "<|fim_prefix|>{{{prefix}}}<|fim_suffix|>{{{suffix}}}<|fim_middle|>"
    defaultCompletionOptions:
      contextLength: 8192
      maxTokens: 256
      temperature: 0.1
      stop: ["<|endoftext|>", "<|fim_prefix|>", "<|fim_suffix|>", "<|fim_middle|>", "<|file_sep|>", "<|fim_pad|>", "<|repo_name|>"]
    autocompleteOptions:
      debounceDelay: 250
      maxPromptTokens: 2048
      onlyMyCode: true

  # --- Embeddings für @codebase / Indexing --------------------------------
  - name: Nomic Embed
    <<: *localai
    model: nomic-embed
    roles: [embed]

context:
  - provider: code
  - provider: codebase
  - provider: diff
  - provider: terminal
  - provider: problems
  - provider: folder
  - provider: file
  - provider: currentFile