Module: Legion::LLM::Settings

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

Class Method Summary collapse

Class Method Details

.api_defaults ⇒ Object 



419
420
421
422
423
424
425
426
427
 
# File 'lib/legion/llm/settings.rb', line 419

def self.api_defaults
  {
    auth: {
      enabled:      false,
      api_keys:     [],
      pass_through: false
    }
  }
end

.arbitrage_defaults ⇒ Object 



288
289
290
291
292
293
294
295
296
 
# File 'lib/legion/llm/settings.rb', line 288

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

.batch_defaults ⇒ Object 



298
299
300
301
302
303
304
305
 
# File 'lib/legion/llm/settings.rb', line 298

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

.budget_defaults ⇒ Object 



268
269
270
271
272
273
274
 
# File 'lib/legion/llm/settings.rb', line 268

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

.claude_cli_defaults ⇒ Object 



157
158
159
160
161
162
 
# File 'lib/legion/llm/settings.rb', line 157

def self.claude_cli_defaults
  {
    settings_path: '~/.claude/settings.json',
    config_path:   '~/.claude.json'
  }
end

.compliance_defaults ⇒ Object 



442
443
444
445
446
447
448
449
450
451
452
453
 
# File 'lib/legion/llm/settings.rb', line 442

def self.compliance_defaults
  {
    classification_scan:   false,
    encrypt_audit:         false,
    phi_block_cloud:       false,
    cloud_providers:       %w[bedrock anthropic openai gemini azure],
    redact_pii:            false,
    redaction_placeholder: '[REDACTED]',
    strict_hipaa:          false,
    default_level:         :public
  }
end

.confidence_defaults ⇒ Object 



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

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

.config_value(config, key, default = nil) ⇒ Object 



62
63
64
65
66
67
68
69
70
71
72
 
# File 'lib/legion/llm/settings.rb', line 62

def self.config_value(config, key, default = nil)
  return default unless config.respond_to?(:key?)

  string_key = key.to_s
  return config[string_key] if config.key?(string_key)

  symbol_key = key.to_sym if key.respond_to?(:to_sym)
  return config[symbol_key] if symbol_key && config.key?(symbol_key)

  default
end

.context_curation_defaults ⇒ Object 



363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
 
# File 'lib/legion/llm/settings.rb', line 363

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 



379
380
381
382
383
384
385
386
 
# File 'lib/legion/llm/settings.rb', line 379

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

.current_settings ⇒ Object 



127
128
129
130
131
132
133
134
135
 
# File 'lib/legion/llm/settings.rb', line 127

def self.current_settings
  settings = Legion::Settings[:llm]
  return settings if settings.is_a?(Hash)

  {}
rescue StandardError => e
  handle_exception(e, level: :warn, handled: true, operation: 'llm.settings.current_settings')
  {}
end

.daemon_defaults ⇒ Object 



199
200
201
202
203
204
 
# File 'lib/legion/llm/settings.rb', line 199

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

.debate_defaults ⇒ Object 



406
407
408
409
410
411
412
413
414
415
416
417
 
# File 'lib/legion/llm/settings.rb', line 406

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 



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
 
# File 'lib/legion/llm/settings.rb', line 11

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,
    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,
    api:                       api_defaults,
    compliance:                compliance_defaults,
    skills:                    skills_defaults,
    claude_cli:                claude_cli_defaults
  }
end

.discovery_defaults ⇒ Object 



224
225
226
227
228
229
230
 
# File 'lib/legion/llm/settings.rb', line 224

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

.embedding_defaults ⇒ Object 



329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
 
# File 'lib/legion/llm/settings.rb', line 329

def self.embedding_defaults
  {
    dimension:                    1024,
    enforce_dimension:            true,
    provider_fallback:            %w[ollama bedrock openai],
    provider_models:              {
      bedrock:   'amazon.titan-embed-text-v2:0',
      anthropic: nil,
      openai:    'text-embedding-3-small',
      gemini:    'text-embedding-004',
      azure:     'text-embedding-3-small',
      ollama:    'mxbai-embed-large'
    },
    ollama_preferred:             %w[mxbai-embed-large nomic-embed-text bge-large snowflake-arctic-embed],
    ollama_context_chars:         {
      'mxbai-embed-large'      => 1400,
      'bge-large'              => 1400,
      'snowflake-arctic-embed' => 1400,
      'nomic-embed-text'       => 24_000
    },
    ollama_default_context_chars: 1400,
    prefix_registry:              {
      'nomic-embed-text'  => { document: 'search_document: ', query: 'search_query: ' },
      'mxbai-embed-large' => { query: 'Represent this sentence for searching relevant passages: ' }
    }
  }
end

.enterprise_privacy? ⇒ Boolean

Returns:

  • (Boolean) 


116
117
118
119
120
121
122
123
124
125
 
# File 'lib/legion/llm/settings.rb', line 116

def self.enterprise_privacy?
  if Legion::Settings.respond_to?(:enterprise_privacy?)
    Legion::Settings.enterprise_privacy?
  else
    ENV['LEGION_ENTERPRISE_PRIVACY'] == 'true'
  end
rescue StandardError => e
  handle_exception(e, level: :warn, handled: true, operation: 'llm.settings.enterprise_privacy')
  ENV['LEGION_ENTERPRISE_PRIVACY'] == 'true'
end

.gateway_defaults ⇒ Object 



276
277
278
279
280
281
282
283
284
285
286
 
# File 'lib/legion/llm/settings.rb', line 276

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

.global_value(namespace, *keys, default: nil) ⇒ Object 



82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
 
# File 'lib/legion/llm/settings.rb', line 82

def self.global_value(namespace, *keys, default: nil)
  if Legion::Settings.respond_to?(:dig) && keys.any?
    direct = Legion::Settings.dig(namespace, *keys)
    return direct unless direct.nil?
  end

  keys.reduce(self.namespace(namespace)) do |current, key|
    return default unless current.respond_to?(:key?)

    config_value(current, key)
  end
rescue StandardError => e
  handle_exception(e, level: :warn, handled: true, operation: 'llm.settings.global_value', namespace: namespace, keys: keys)
  default
end

.namespace(namespace) ⇒ Object 



74
75
76
77
78
79
80
 
# File 'lib/legion/llm/settings.rb', line 74

def self.namespace(namespace)
  settings = Legion::Settings[namespace]
  settings.is_a?(Hash) ? settings : {}
rescue StandardError => e
  handle_exception(e, level: :warn, handled: true, operation: 'llm.settings.namespace', namespace: namespace)
  {}
end

.prompt_caching_defaults ⇒ Object 



206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
 
# File 'lib/legion/llm/settings.rb', line 206

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',
      spool_threshold_bytes: 8 * 1024 * 1024
    }
  }
end

.provider_layer_defaults ⇒ Object 



388
389
390
391
392
393
394
395
396
 
# File 'lib/legion/llm/settings.rb', line 388

def self.provider_layer_defaults
  {
    mode:             'auto',
    native_providers: %w[
      ollama vllm anthropic openai gemini mlx
      bedrock azure_foundry vertex claude
    ]
  }
end

.rag_defaults ⇒ Object 



316
317
318
319
320
321
322
323
324
325
326
327
 
# File 'lib/legion/llm/settings.rb', line 316

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

.register_defaults! ⇒ Object 



137
138
139
140
141
142
 
# File 'lib/legion/llm/settings.rb', line 137

def self.register_defaults!
  return unless Legion::Settings.respond_to?(:register_library)

  log.debug '[llm][settings] action=register_defaults'
  Legion::Settings.register_library(:llm, default)
end

.routing_defaults ⇒ Object 



232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
 
# File 'lib/legion/llm/settings.rb', line 232

def self.routing_defaults
  {
    enabled:        true,
    tier_priority:  %w[local fleet openai_compat cloud frontier],
    default_intent: { privacy: 'normal', capability: 'moderate', cost: 'normal' },
    tiers:          {
      local:         { provider: 'ollama' },
      fleet:         {
        queue:           'llm.fleet',
        routing_style:   :shared_lane,
        timeout_seconds: 30,
        timeouts:        { embed: 10, chat: 30, generate: 30, default: 30 }
      },
      openai_compat: {
        gateways: []
      },
      cloud:         { providers: %w[bedrock azure gemini] },
      frontier:      { providers: %w[anthropic openai] }
    },
    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:           true,
      pipeline_enabled:  true,
      max_attempts:      3,
      quality_threshold: 0
    },
    rules:          [],
    tier_mappings:  []
  }
end

.scheduling_defaults ⇒ Object 



307
308
309
310
311
312
313
314
 
# File 'lib/legion/llm/settings.rb', line 307

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

.set_value(*keys, value:) ⇒ Object 



98
99
100
101
102
103
104
105
106
107
 
# File 'lib/legion/llm/settings.rb', line 98

def self.set_value(*keys, value:)
  target = current_settings
  return value unless target.is_a?(Hash)

  assign_value(target, keys, value)
  value
rescue StandardError => e
  handle_exception(e, level: :warn, handled: true, operation: 'llm.settings.set_value', keys: keys)
  value
end

.skills_defaults ⇒ Object 



429
430
431
432
433
434
435
436
437
438
439
440
 
# File 'lib/legion/llm/settings.rb', line 429

def self.skills_defaults
  {
    enabled:           true,
    auto_inject:       true,
    on_demand:         true,
    max_active_skills: 1,
    directories:       ['.legion/skills', '~/.legionio/skills'],
    auto_discover:     { claude: false, codex: false },
    enabled_skills:    [],
    disabled_skills:   []
  }
end

.system_baseline_default ⇒ Object 



164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
 
# File 'lib/legion/llm/settings.rb', line 164

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 



357
358
359
360
361
 
# File 'lib/legion/llm/settings.rb', line 357

def self.telemetry_defaults
  {
    pipeline_spans: true
  }
end

.tool_trigger_defaults ⇒ Object 



398
399
400
401
402
403
404
 
# File 'lib/legion/llm/settings.rb', line 398

def self.tool_trigger_defaults
  {
    scan_depth:       10,
    tool_limit:       25,
    local_tool_limit: 10
  }
end

.transport_connected? ⇒ Boolean

Returns:

  • (Boolean) 


109
110
111
112
113
114
 
# File 'lib/legion/llm/settings.rb', line 109

def self.transport_connected?
  return true if global_value(:transport, :connected) == true

  transport = namespace(:transport)
  config_value(transport, :connected) == true
end

.value(*keys, default: nil) ⇒ Object 



47
48
49
50
51
52
53
54
55
56
57
58
59
60
 
# File 'lib/legion/llm/settings.rb', line 47

def self.value(*keys, default: nil)
  missing = Object.new
  keys.reduce(current_settings) do |current, key|
    return default unless current.respond_to?(:key?)

    value = config_value(current, key, missing)
    return default if value.equal?(missing)

    value
  end
rescue StandardError => e
  handle_exception(e, level: :warn, operation: 'llm.settings.value')
  default
end