modelweaver/modelweaver.example.yaml at main · kianwoon/modelweaver · 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
97
98
99
100
101
102
# ModelWeaver Configuration
# Run 'npx modelweaver init' to auto-configure this file AND
# ~/.claude/settings.json for seamless Claude Code integration.

# Hedging and timing configuration
hedging:
  speculativeDelay: 1000   # ms before starting backup providers (lower = faster fallback)
  cvThreshold: 0.5         # latency variance threshold for multi-copy hedging
  maxHedge: 4              # max concurrent copies per request

# Smart routing: classify request complexity and route to appropriate tier
# Disabled by default. When enabled, analyzes the last user message to determine
# if a more capable (or simpler) model tier should handle the request.
# Requires routing entries named "tier1", "tier2" (and optionally "tier3").
# smartRouting:
#   enabled: true
#   escalationThreshold: 2  # minimum cumulative score to trigger tier override
#   patterns:
#     "1":                   # Tier 1 — best models for complex tasks
#       - pattern: "architect|design system|from scratch"
#         score: 3
#       - pattern: "debug|troubleshoot|investigate|root cause"
#         score: 2
#       - pattern: "analyze.*reason|explain why|step.by.step"
#         score: 2
#     "2":                   # Tier 2 — good models for moderate tasks
#       - pattern: "explain|summarize|compare"
#         score: 2
#       - pattern: "write.*test|refactor|review"
#         score: 2
#
# When smartRouting is enabled, add these routing entries (coexist with named tiers):
# routing:
#   tier1:                    # Smart routing tier 1 — routed for complex requests
#     - provider: anthropic
#       model: claude-opus-4-20250514
#   tier2:                    # Smart routing tier 2 — routed for moderate requests
#     - provider: anthropic
#       model: claude-sonnet-4-20250514
#   tier3:                    # Smart routing tier 3 — default / passthrough
#     - provider: anthropic
#       model: claude-haiku-4-5-20251001

server:
  port: 3456
  host: localhost

providers:
  anthropic:
    baseUrl: https://api.anthropic.com
    apiKey: ${ANTHROPIC_API_KEY}
    rateLimitBackoffMs: 1000  # delay (ms) before retrying after 429/503 when no Retry-After header
  openrouter:
    baseUrl: https://openrouter.ai/api
    apiKey: ${OPENROUTER_API_KEY}
    rateLimitBackoffMs: 1000  # delay (ms) before retrying after 429/503 when no Retry-After header
  # Example: provider-specific transient error patterns
  # some-provider:
  #   retryableErrorPatterns:
  #     - "network error"
  #     - "code:1234"
  #     - "please contact customer service"

# Exact model name routing (checked FIRST, before tier patterns)
modelRouting:
  "glm-5-turbo":
    - provider: anthropic
  "MiniMax-M2.7":
    - provider: openrouter
  # Distribution example: spread traffic across providers by weight
  # "claude-sonnet-4":
  #   - provider: anthropic
  #     weight: 70
  #   - provider: openrouter
  #     weight: 30

# Tier-based routing (fallback)
routing:
  sonnet:
    - provider: anthropic
      model: claude-sonnet-4-20250514
      # Context management — reduce tokens sent to provider:
      # compressOldTurns: 10   # keep last 10 turns verbatim, skeleton older ones
      # maxContextMessages: 60 # hard limit on total messages (runs after compression)
      # toolResultLimit: 8000  # max chars per tool_result block
    - provider: openrouter
      model: anthropic/claude-sonnet-4
  opus:
    - provider: anthropic
      model: claude-opus-4-20250514
    - provider: openrouter
      model: anthropic/claude-opus-4
  haiku:
    - provider: anthropic
      model: claude-haiku-4-5-20251001
    - provider: openrouter
      model: anthropic/claude-haiku-4

tierPatterns:
  sonnet: ["sonnet", "3-5-sonnet", "3.5-sonnet"]
  opus: ["opus", "3-opus", "3.5-opus"]
  haiku: ["haiku", "3-haiku", "3.5-haiku"]