Module: Legion::LLM::Settings

Extended by:
Legion::Logging::Helper
Defined in:
lib/legion/llm/settings.rb

Class Method Summary collapse

Class Method Details

.arbitrage_defaults ⇒ Object 



160
161
162
163
164
165
166
167
168
 
# File 'lib/legion/llm/settings.rb', line 160

def self.arbitrage_defaults
  {
    enabled:            false,
    prefer_cheapest:    true,
    quality_floor:      0.7,
    cost_table_refresh: 86_400,
    cost_table:         {}
  }
end

.batch_defaults ⇒ Object 



170
171
172
173
174
175
176
177
 
# File 'lib/legion/llm/settings.rb', line 170

def self.batch_defaults
  {
    enabled:          false,
    window_seconds:   300,
    max_batch_size:   100,
    eligible_intents: %w[batch background low_priority]
  }
end

.budget_defaults ⇒ Object 



140
141
142
143
144
145
146
 
# File 'lib/legion/llm/settings.rb', line 140

def self.budget_defaults
  {
    session_max_tokens:  nil,
    session_warn_tokens: nil,
    daily_max_tokens:    nil
  }
end

.confidence_defaults ⇒ Object 



66
67
68
69
70
71
72
73
74
75
 
# File 'lib/legion/llm/settings.rb', line 66

def self.confidence_defaults
  {
    bands: {
      low:       0.3,
      medium:    0.5,
      high:      0.7,
      very_high: 0.9
    }
  }
end

.context_curation_defaults ⇒ Object 



222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
 
# File 'lib/legion/llm/settings.rb', line 222

def self.context_curation_defaults
  {
    enabled:               true,
    mode:                  'heuristic',
    llm_assisted:          false,
    llm_model:             nil,
    tool_result_max_chars: 2000,
    thinking_eviction:     true,
    exchange_folding:      true,
    superseded_eviction:   true,
    dedup_enabled:         true,
    dedup_threshold:       0.85,
    target_context_tokens: 40_000
  }
end

.conversation_defaults ⇒ Object 



238
239
240
241
242
243
244
245
 
# File 'lib/legion/llm/settings.rb', line 238

def self.conversation_defaults
  {
    summarize_threshold: 50_000,
    target_tokens:       20_000,
    preserve_recent:     10,
    auto_compact:        true
  }
end

.daemon_defaults ⇒ Object 



77
78
79
80
81
82
 
# File 'lib/legion/llm/settings.rb', line 77

def self.daemon_defaults
  {
    url:     'http://127.0.0.1:4567',
    enabled: true
  }
end

.debate_defaults ⇒ Object 



262
263
264
265
266
267
268
269
270
271
272
273
 
# File 'lib/legion/llm/settings.rb', line 262

def self.debate_defaults
  {
    enabled:                  false,
    gaia_auto_trigger:        false,
    default_rounds:           1,
    max_rounds:               3,
    advocate_model:           nil,
    challenger_model:         nil,
    judge_model:              nil,
    model_selection_strategy: 'rotate'
  }
end

.default ⇒ Object 



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
 
# File 'lib/legion/llm/settings.rb', line 9

def self.default
  model_override = ENV.fetch('ANTHROPIC_MODEL', nil)
  {
    enabled:                   true,
    connected:                 false,
    pipeline_enabled:          true,
    pipeline_async_post_steps: true,
    max_tool_rounds:           200,
    default_model:             model_override,
    default_provider:          nil,
    system_baseline:           system_baseline_default,
    providers:                 providers,
    routing:                   routing_defaults,
    budget:                    budget_defaults,
    confidence:                confidence_defaults,
    discovery:                 discovery_defaults,
    gateway:                   gateway_defaults,
    daemon:                    daemon_defaults,
    prompt_caching:            prompt_caching_defaults,
    arbitrage:                 arbitrage_defaults,
    batch:                     batch_defaults,
    scheduling:                scheduling_defaults,
    rag:                       rag_defaults,
    embedding:                 embedding_defaults,
    conversation:              conversation_defaults,
    telemetry:                 telemetry_defaults,
    context_curation:          context_curation_defaults,
    debate:                    debate_defaults,
    provider_layer:            provider_layer_defaults,
    tool_trigger:              tool_trigger_defaults
  }
end

.discovery_defaults ⇒ Object 



101
102
103
104
105
106
107
 
# File 'lib/legion/llm/settings.rb', line 101

def self.discovery_defaults
  {
    enabled:         true,
    refresh_seconds: 60,
    memory_floor_mb: 2048
  }
end

.embedding_defaults ⇒ Object 



201
202
203
204
205
206
207
208
209
210
211
212
213
214
 
# File 'lib/legion/llm/settings.rb', line 201

def self.embedding_defaults
  {
    dimension:         1024,
    enforce_dimension: true,
    provider_fallback: %w[ollama bedrock openai],
    provider_models:   {
      ollama:  'mxbai-embed-large',
      azure:   'text-embedding-3-small',
      bedrock: 'amazon.titan-embed-text-v2:0',
      openai:  'text-embedding-3-small'
    },
    ollama_preferred:  %w[mxbai-embed-large nomic-embed-text bge-large snowflake-arctic-embed]
  }
end

.gateway_defaults ⇒ Object 



148
149
150
151
152
153
154
155
156
157
158
 
# File 'lib/legion/llm/settings.rb', line 148

def self.gateway_defaults
  {
    enabled:            true,
    endpoint:           nil,
    api_key:            nil,
    timeout_seconds:    30,
    model_policy:       {},
    headers:            {},
    fallback_to_direct: true
  }
end

.prompt_caching_defaults ⇒ Object 



84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
 
# File 'lib/legion/llm/settings.rb', line 84

def self.prompt_caching_defaults
  {
    enabled:             true,
    min_tokens:          1024,
    scope:               'ephemeral',
    cache_system_prompt: true,
    cache_tools:         true,
    cache_conversation:  true,
    sort_tools:          true,
    response_cache:      {
      enabled:     true,
      ttl_seconds: 300,
      spool_dir:   '~/.legionio/data/spool/llm_responses'
    }
  }
end

.provider_layer_defaults ⇒ Object 



247
248
249
250
251
252
253
 
# File 'lib/legion/llm/settings.rb', line 247

def self.provider_layer_defaults
  {
    mode:                 'ruby_llm',
    native_providers:     %w[claude bedrock],
    fallback_to_ruby_llm: true
  }
end

.providers ⇒ Object 



275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
 
# File 'lib/legion/llm/settings.rb', line 275

def self.providers
  {
    bedrock:   {
      enabled:       false,
      default_model: 'us.anthropic.claude-sonnet-4-6',
      api_key:       nil,
      secret_key:    nil,
      session_token: nil,
      bearer_token:  'env://AWS_BEARER_TOKEN_BEDROCK',
      region:        'us-east-2'
    },
    anthropic: {
      enabled:       false,
      default_model: 'claude-sonnet-4-6',
      api_key:       'env://ANTHROPIC_API_KEY'
    },
    openai:    {
      enabled:       false,
      default_model: 'gpt-4o',
      api_key:       ['env://OPENAI_API_KEY', 'env://CODEX_API_KEY']
    },
    gemini:    {
      enabled:       false,
      default_model: 'gemini-2.0-flash',
      api_key:       'env://GEMINI_API_KEY'
    },
    azure:     {
      enabled:       false,
      default_model: nil,
      api_base:      nil,
      api_key:       nil,
      auth_token:    nil
    },
    ollama:    {
      enabled:       false,
      default_model: 'qwen3.5:latest',
      base_url:      'http://localhost:11434'
    }
  }
end

.rag_defaults ⇒ Object 



188
189
190
191
192
193
194
195
196
197
198
199
 
# File 'lib/legion/llm/settings.rb', line 188

def self.rag_defaults
  {
    enabled:                       true,
    full_limit:                    10,
    compact_limit:                 5,
    min_confidence:                0.5,
    utilization_compact_threshold: 0.7,
    utilization_skip_threshold:    0.9,
    trivial_max_chars:             20,
    trivial_patterns:              %w[hello hi hey ping pong test ok okay yes no thanks thank]
  }
end

.routing_defaults ⇒ Object 



109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
 
# File 'lib/legion/llm/settings.rb', line 109

def self.routing_defaults
  {
    enabled:        false,
    tier_priority:  %w[local fleet direct],
    default_intent: { privacy: 'normal', capability: 'moderate', cost: 'normal' },
    tiers:          {
      local: { provider: 'ollama' },
      fleet: {
        queue:           'llm.request',
        timeout_seconds: 30,
        timeouts:        { embed: 10, chat: 30, generate: 30, default: 30 }
      },
      cloud: { providers: %w[bedrock anthropic] }
    },
    health:         {
      window_seconds:               300,
      circuit_breaker:              { failure_threshold: 3, cooldown_seconds: 60 },
      latency_penalty_threshold_ms: 5000,
      budget:                       { daily_limit_usd: nil, monthly_limit_usd: nil }
    },
    escalation:     {
      enabled:           false,
      pipeline_enabled:  true,
      max_attempts:      3,
      quality_threshold: 50
    },
    rules:          [],
    tier_mappings:  []
  }
end

.scheduling_defaults ⇒ Object 



179
180
181
182
183
184
185
186
 
# File 'lib/legion/llm/settings.rb', line 179

def self.scheduling_defaults
  {
    enabled:         false,
    peak_hours_utc:  '14-22',
    defer_intents:   %w[batch background],
    max_defer_hours: 8
  }
end

.system_baseline_default ⇒ Object 



42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
 
# File 'lib/legion/llm/settings.rb', line 42

def self.system_baseline_default
  <<~PROMPT.strip
    You are Legion, an agentic AI partner running on the LegionIO framework.

    LegionIO is a governed, production-oriented cognitive task and orchestration platform.
    Your role is to help the user accomplish real work quickly, directly, and safely.

    Core behavior:
    - Honor user intent and constraints.
    - Prefer execution over prompt ceremony: do the task when possible, don't just describe it.
    - Be concise by default; expand only when the user asks for depth.
    - Be transparent: never claim you ran something you did not run, and never hide uncertainty.
    - Minimize blast radius: make the smallest effective change and preserve existing behavior unless asked otherwise.
    - Do not YOLO risky actions. For destructive, irreversible, security-sensitive, or high-impact actions, pause and get explicit confirmation.
    - When risk or ambiguity is high, ask focused clarifying questions before acting.
    - Validate outcomes when practical, and report what changed and why.
    - Prefer solving work directly in-session; only produce handoff artifacts (including prompts for other AI tools) when the user explicitly asks for that format.

    Trust model:
    - Trust is earned through reliable outcomes, clarity, and safe execution.
    - Speed matters, but never at the expense of integrity or user trust.
  PROMPT
end

.telemetry_defaults ⇒ Object 



216
217
218
219
220
 
# File 'lib/legion/llm/settings.rb', line 216

def self.telemetry_defaults
  {
    pipeline_spans: true
  }
end

.tool_trigger_defaults ⇒ Object 



255
256
257
258
259
260
 
# File 'lib/legion/llm/settings.rb', line 255

def self.tool_trigger_defaults
  {
    scan_depth: 2,
    tool_limit: 10
  }
end