YulYens_AI/config.yaml at main · YulYen/YulYens_AI · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
# Language: German (de) or English (en)
language: "de"

# --- Core settings ---------------------------------------------------------------
core:
  # Backend selection: use "ollama" for real LLMs, "dummy" for quick echo tests
  backend: "ollama"
  # Default model name at startup
  model_name: "ministral-3:8b" # e.g. "gptoss20b-q4ks", "mistral7b-local", or "leo-hessianai-13b-chat.Q5" or  "ministral-3:8b"
  warm_up: false
  # Base URL for Ollama (protocol + host + port), e.g. http://127.0.0.1:11434
  ollama_url: "http://127.0.0.1:11434"
  include_date: true  # master switch to add the current date to persona system prompts

# --- UI settings ---------------------------------------------------------------
ui:
  type: web   # options: "web", "terminal", or null (= API only)
  experimental:
    broadcast_mode: true  # enable the ensemble broadcast toggle (TerminalUI, WebUI)
  web:
    host: "0.0.0.0"
    port: 7860
    share: false  # enable to publish a public Gradio share link
    share_auth:
      username: ""
      password: ""


# --- Text-to-Speech settings (piper is currently the only provider) ---------------------------------------------------------------
tts:
  enabled: true
  features:
    terminal_auto_create_wav: true  # creates a wav-file in "out" directory for every AI-Answer
  voices:
    default:
      de: "de_DE-thorsten-high"
      en: "en_US-amy-medium"
    personas_de:
      DORIS: "de_DE-kerstin-low"
      POPCORN: "de_DE-pavoque-low"
      LEAH: "de_DE-kerstin-low"
      PETER: "de_DE-thorsten-high"

# --- Wiki settings (including spaCy-Keyword-Finder) -------------------------------------------------------------
wiki:
  # offline uses Kiwix, online calls Wikipedia, false disables wiki snippets
  mode: offline
  # spaCy model variant for keyword extraction: "medium" or "large"
  spacy_model_variant: "large"
  spacy_model_map:
    de:
      medium: "de_core_news_md"
      large:  "de_core_news_lg"
    en:
      medium: "en_core_web_md"
      large:  "en_core_web_lg"
  online_base_url_map:
    de: "https://de.wikipedia.org"
    en: "https://en.wikipedia.org"
  # Port of the local proxy (always required)
  proxy_port: 8042
  # Maximum wiki snippet length in characters (recommend 1200–1600)
  snippet_limit: 1200
  # Maximum number of distinct wiki snippets to inject when multiple entities are detected
  max_wiki_snippets: 2
  timeout_connect: 5.0
  timeout_read: 8.0
  offline:  # only relevant when mode == "offline" (Kiwix-based wiki mode)
    host: "127.0.0.1"  # preferred host for kiwix-serve; faster than localhost
    kiwix_port: 8080 #
    zim_prefix: "wikipedia_de_all_nopic_2025-06"
    autostart: true
    kiwix_exe: "C:/wikipedia-de-offline/kiwix-serve.exe"   # path to kiwix-serve (Windows example)
    zim_path: "C:\\wikipedia-de-offline\\wikipedia_de_all_nopic_2025-06.zim"
    startup_timeout_s: 7

# --- Logging -------------------------------------------------------------------
logging:
  level: "INFO"   # valid levels: DEBUG, INFO, WARNING, ERROR
  to_console: "auto" # "true", "false" or "auto" (auto: false if terminal-ui, otherwise true)
  dir: "logs"
  conversation_prefix: "conversation"

# --- One-shot API (e.g. for tests) ----------------------------------------------
api:
  enabled: true
  host: "127.0.0.1"
  port: 8013

security:
  enabled: true
  guard: BasicGuard
  prompt_injection_protection: true
  pii_protection: false
  output_blocklist: true
  # Input: block injections/PII. Output: mask PII, block secrets.