Module: Legion::LLM::Settings

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

Constant Summary collapse

CLIENT_TOOL_PASSTHROUGH_BLACKLIST_DEFAULT = 
[
  'sudo', 'visudo', 'su', 'legion', 'legionio', 'legionio do', 'legionio/legion',
  'computer_use_session', 'computer_use_control', 'computer_use_session_info',
  'computer_use_session_message', 'plugin__aithena__recall', 'plugin__aithena__remember',
  'plugin__aithena__skill_search', 'plugin__aithena__skill_feedback', 'plugin__aithena__memory_stats',
  'plugin__cron__create', 'plugin__cron__list', 'plugin__cron__get', 'plugin__cron__update',
  'plugin__cron__delete', 'plugin__cron__get_history', 'plugin__cron__run_now', 'plugin__cron__stop'
].freeze
CLIENT_TOOL_PASSTHROUGH_WHITELIST_DEFAULT = 
[].freeze

Class Method Summary collapse

Class Method Details

.api_defaults ⇒ Object 



410
411
412
413
414
415
416
417
418
419
 
# File 'lib/legion/llm/settings.rb', line 410

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

.arbitrage_defaults ⇒ Object 



249
250
251
252
253
254
255
256
257
 
# File 'lib/legion/llm/settings.rb', line 249

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

.batch_defaults ⇒ Object 



259
260
261
262
263
264
265
266
 
# File 'lib/legion/llm/settings.rb', line 259

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

.budget_defaults ⇒ Object 



240
241
242
243
244
245
246
247
 
# File 'lib/legion/llm/settings.rb', line 240

def self.budget_defaults
  {
    session_max_tokens:  nil,
    session_warn_tokens: nil,
    daily_max_tokens:    nil,
    session_usd:         0.0
  }
end

.claude_cli_defaults ⇒ Object 



87
88
89
90
91
92
 
# File 'lib/legion/llm/settings.rb', line 87

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

.compliance_defaults ⇒ Object 



447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
 
# File 'lib/legion/llm/settings.rb', line 447

def self.compliance_defaults
  {
    classification_scan:   false,
    encrypt_audit:         false,
    encrypt_metering:      false,
    phi_block_cloud:       false,
    cloud_providers:       %w[bedrock anthropic openai gemini azure],
    redact_pii:            false,
    redaction_placeholder: '[REDACTED]',
    strict_hipaa:          false,
    standalone_email_pii:  false,
    default_level:         :public,
    audit_max_messages:    20
  }
end

.confidence_defaults ⇒ Object 



122
123
124
125
126
127
128
129
130
131
 
# File 'lib/legion/llm/settings.rb', line 122

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

.context_curation_defaults ⇒ Object 



349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
 
# File 'lib/legion/llm/settings.rb', line 349

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,
    archive_dropped_turns:   true,
    archive_preserve_recent: 10
  }
end

.conversation_defaults ⇒ Object 



367
368
369
370
371
372
373
374
 
# File 'lib/legion/llm/settings.rb', line 367

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

.daemon_defaults ⇒ Object 



133
134
135
136
137
138
 
# File 'lib/legion/llm/settings.rb', line 133

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

.debate_defaults ⇒ Object 



397
398
399
400
401
402
403
404
405
406
407
408
 
# File 'lib/legion/llm/settings.rb', line 397

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 



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
 
# File 'lib/legion/llm/settings.rb', line 21

def self.default
  model_override = ENV.fetch('ANTHROPIC_MODEL', nil)
  {
    enabled:                        true,
    connected:                      false,
    pipeline_enabled:               true,
    pipeline_async_post_steps:      true,
    context_window:                 262_144,
    max_output_tokens:              16_384,
    max_tool_rounds:                200,
    max_tool_calls_per_turn:        50,
    tool_result_max_dispatch_chars: 4000,
    default_model:                  model_override,
    default_temperature:            1.0,
    default_provider:               nil,
    system_baseline:                system_baseline_default,
    fleet:                          fleet_defaults,
    routing:                        routing_defaults,
    budget:                         budget_defaults,
    confidence:                     confidence_defaults,
    discovery:                      discovery_defaults,
    daemon:                         daemon_defaults,
    prompt_caching:                 prompt_caching_defaults,
    arbitrage:                      arbitrage_defaults,
    batch:                          batch_defaults,
    scheduling:                     scheduling_defaults,
    rag:                            rag_defaults,
    rag_guard:                      rag_guard_defaults,
    gaia:                           gaia_defaults,
    embedding:                      embedding_defaults,
    conversation:                   conversation_defaults,
    telemetry:                      telemetry_defaults,
    metering:                       metering_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,
    fallback:                       fallback_defaults,
    structured_output:              structured_output_defaults
  }
end

.discovery_defaults ⇒ Object 



158
159
160
161
162
163
164
165
 
# File 'lib/legion/llm/settings.rb', line 158

def self.discovery_defaults
  {
    enabled:                true,
    refresh_seconds:        60,
    memory_floor_mb:        2048,
    memory_overhead_factor: 1.4
  }
end

.embedding_defaults ⇒ Object 



306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
 
# File 'lib/legion/llm/settings.rb', line 306

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

.fallback_defaults ⇒ Object 



434
435
436
437
438
 
# File 'lib/legion/llm/settings.rb', line 434

def self.fallback_defaults
  {
    allow_local: true
  }
end

.fleet_defaults ⇒ Object 



167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
 
# File 'lib/legion/llm/settings.rb', line 167

def self.fleet_defaults
  {
    dispatch:  {
      enabled:                true,
      exchange:               'llm.fleet',
      routing_style:          :shared_lane,
      mandatory:              true,
      publisher_confirm:      true,
      spool:                  false,
      timeout_seconds:        30,
      timeouts:               { chat: 30, stream: 30, embed: 10, image: 60, default: 30 },
      require_auth:           nil,
      token_ttl_seconds:      180,
      reply_queue_expires_ms: 60_000,
      reply_queue_prefix:     'llm.fleet.reply',
      request_ttl_ms:         120_000
    },
    auth:      {
      require_signed_token:   true,
      issuer:                 'legion-llm',
      audience:               'lex-llm-fleet-worker',
      algorithm:              'HS256',
      accepted_issuers:       ['legion-llm'],
      max_clock_skew_seconds: 30
    },
    responder: {
      enabled:                    true,
      require_auth:               nil,
      require_policy:             false,
      require_idempotency:        true,
      idempotency_ttl_seconds:    600,
      accepted_protocol_version:  2,
      mandatory:                  false,
      publisher_confirm:          false,
      publish_confirm_timeout_ms: 500,
      spool:                      false
    }
  }
end

.gaia_defaults ⇒ Object 



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

def self.gaia_defaults
  {
    advisory_enabled: true
  }
end

.metering_defaults ⇒ Object 



340
341
342
343
344
345
346
347
 
# File 'lib/legion/llm/settings.rb', line 340

def self.metering_defaults
  {
    spool: {
      max_events:        10_000,
      flush_batch_sleep: 0.0
    }
  }
end

.prompt_caching_defaults ⇒ Object 



140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
 
# File 'lib/legion/llm/settings.rb', line 140

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 



376
377
378
379
380
381
382
383
384
 
# File 'lib/legion/llm/settings.rb', line 376

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

.rag_defaults ⇒ Object 



277
278
279
280
281
282
283
284
285
286
287
288
289
290
 
# File 'lib/legion/llm/settings.rb', line 277

def self.rag_defaults
  {
    enabled:                       true,
    full_limit:                    5,
    compact_limit:                 3,
    min_confidence:                0.92,
    utilization_compact_threshold: 0.7,
    utilization_skip_threshold:    0.9,
    conversation_history_enabled:  false,
    trivial_max_chars:             20,
    trivial_patterns:              %w[hello hi hey ping pong test ok okay yes no thanks thank],
    exclude_source_agents:         %w[teams-api-ingest unknown teams-entity-extractor legion-interlink]
  }
end

.rag_guard_defaults ⇒ Object 



292
293
294
295
296
297
298
 
# File 'lib/legion/llm/settings.rb', line 292

def self.rag_guard_defaults
  {
    threshold:        0.7,
    block_on_failure: true,
    evaluators:       %i[faithfulness rag_relevancy]
  }
end

.register_defaults! ⇒ Object 



67
68
69
70
 
# File 'lib/legion/llm/settings.rb', line 67

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

.routing_defaults ⇒ Object 



207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
 
# File 'lib/legion/llm/settings.rb', line 207

def self.routing_defaults
  {
    enabled:        true,
    tier_priority:  %w[local direct fleet 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 }
      },
      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 



268
269
270
271
272
273
274
275
 
# File 'lib/legion/llm/settings.rb', line 268

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

.skills_defaults ⇒ Object 



421
422
423
424
425
426
427
428
429
430
431
432
 
# File 'lib/legion/llm/settings.rb', line 421

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

.structured_output_defaults ⇒ Object 



440
441
442
443
444
445
 
# File 'lib/legion/llm/settings.rb', line 440

def self.structured_output_defaults
  {
    retry_on_parse_failure: true,
    max_retries:            2
  }
end

.system_baseline_default ⇒ Object 



94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
 
# File 'lib/legion/llm/settings.rb', line 94

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.

    Tool use:
    - There is no tool call limit per turn. You may make as many tool calls as needed to complete the task.
    - Do not stop mid-task claiming you hit a limit. Continue until the work is done or you need user input.
  PROMPT
end

.telemetry_defaults ⇒ Object 



334
335
336
337
338
 
# File 'lib/legion/llm/settings.rb', line 334

def self.telemetry_defaults
  {
    pipeline_spans: true
  }
end

.tool_trigger_defaults ⇒ Object 



386
387
388
389
390
391
392
393
394
395
 
# File 'lib/legion/llm/settings.rb', line 386

def self.tool_trigger_defaults
  {
    scan_depth:                        10,
    tool_limit:                        25,
    local_tool_limit:                  100,
    client_tool_passthrough:           true,
    client_tool_passthrough_whitelist: CLIENT_TOOL_PASSTHROUGH_WHITELIST_DEFAULT.dup,
    client_tool_passthrough_blacklist: CLIENT_TOOL_PASSTHROUGH_BLACKLIST_DEFAULT.dup
  }
end

.validate!(settings) ⇒ Object 



72
73
74
75
76
77
78
79
80
81
82
83
84
85
 
# File 'lib/legion/llm/settings.rb', line 72

def self.validate!(settings)
  if settings.is_a?(Hash) && (settings.key?(:gateway) || settings.key?('gateway'))
    raise ArgumentError,
          'llm.gateway has been removed; configure provider instances instead'
  end

  routing = settings.is_a?(Hash) ? (settings[:routing] || settings['routing'] || {}) : {}
  if routing.is_a?(Hash) && (routing.key?(:use_fleet) || routing.key?('use_fleet'))
    raise ArgumentError,
          'routing.use_fleet has been removed; configure fleet.dispatch.enabled instead'
  end

  settings
end