aurarouter/auraconfig.yaml at main · AuraCoreDynamics/aurarouter · 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
# --- SYSTEM SETTINGS ---
system:
  log_level: INFO
  default_timeout: 120.0
  active_analyzer: aurarouter-default

# --- HARDWARE INVENTORY (mobile1) ---
models:
  # ----- Ollama (local HTTP) -----
  local_3070_qwen:
    provider: ollama
    endpoint: http://localhost:11434/api/generate
    model_name: qwen2.5-coder:7b
    hosting_tier: on-prem
    cost_per_1m_input: 0.0
    cost_per_1m_output: 0.0
    parameters:
      temperature: 0.1
      num_ctx: 4096

  # ----- llama.cpp (embedded, no Ollama needed) -----
  # Download a GGUF first:  aurarouter download-model --repo Qwen/Qwen2.5-Coder-7B-Instruct-GGUF --file qwen2.5-coder-7b-instruct-q4_k_m.gguf
  # local_llama_qwen:
  #   provider: llamacpp
  #   model_path: "C:/models/qwen2.5-coder-7b-q4_k_m.gguf"
  #   parameters:
  #     n_ctx: 4096
  #     n_gpu_layers: -1      # -1 = all layers to GPU
  #     n_batch: 512
  #     temperature: 0.1
  #     max_tokens: 2048

  # ----- Ollama on another machine -----
  # local_3090_deepseek:
  #   provider: ollama
  #   endpoint: http://192.168.1.50:11434/api/generate
  #   model_name: deepseek-coder-v2:lite
  #   parameters:
  #     num_ctx: 16384

  # ----- OpenAPI-compatible (vLLM, LocalAI, LM Studio, etc.) -----
  # my_openapi:
  #   provider: openapi
  #   endpoint: http://localhost:8000/v1
  #   model_name: my-model
  #   # api_key: YOUR_API_KEY

# --- ROLES & ROUTING ---
# The router iterates each list until one model succeeds.
roles:
  # Who decides intent?
  router:
    - local_3070_qwen

  # Who creates the architectural plan?
  reasoning:
    # - local_3090_deepseek
    - local_3070_qwen

  # Who writes the code?
  coding:
    - local_3070_qwen

  # Who validates output quality? (Optional — enables closed-loop execution)
  reviewer:
    - local_3070_qwen

# --- EXECUTION SETTINGS ---
execution:
  max_review_iterations: 3   # Max review-correct cycles per task (0 = disable review)

# --- UNIFIED ARTIFACT CATALOG ---
# Models, services, and analyzers in one registry.
# Models can also be registered here (new-style), alongside the legacy
# 'models' section above.
catalog:
  aurarouter-default:
    kind: analyzer
    display_name: AuraRouter Default
    description: Intent classification with complexity-based triage routing
    provider: aurarouter
    analyzer_kind: intent_triage
    capabilities:
      - code
      - reasoning
      - review
      - planning
    role_bindings:
      simple_code: coding
      complex_reasoning: reasoning
      review: reviewer