Module: Legion::Extensions::Apollo::Runners::Gas

Extended by:
Helpers::Lex
Includes:
Helpers::Lex
Defined in:
lib/legion/extensions/apollo/runners/gas.rb

Constant Summary collapse

RELATION_TYPES = 
%w[
  similar_to contradicts depends_on causes
  part_of supersedes supports_by extends
].freeze
RELATE_CONFIDENCE_GATE = 
0.7
SYNTHESIS_CONFIDENCE_CAP = 
0.7
MAX_ANTICIPATIONS = 
3

Class Method Summary collapse

Class Method Details

.build_synthesis_entry(item, facts) ⇒ Object 



311
312
313
314
315
316
317
318
319
320
321
322
323
324
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 311

def build_synthesis_entry(item, facts)
  source_indices = item[:source_indices] || item['source_indices'] || []
  source_confs = source_indices.filter_map { |i| facts[i]&.dig(:confidence) }
  fb = fallback_confidence
  geo_mean = source_confs.empty? ? fb : geometric_mean(source_confs)

  {
    content:        item[:content] || item['content'],
    content_type:   (item[:content_type] || item['content_type'] || 'inference').to_sym,
    status:         :candidate,
    confidence:     [geo_mean, synthesis_confidence_cap].min,
    source_indices: source_indices
  }
end

.classify_relation(fact, entry) ⇒ Object 



199
200
201
202
203
204
205
206
207
208
209
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 199

def classify_relation(fact, entry)
  fb_conf = fallback_confidence
  if llm_available?
    llm_classify_relation(fact, entry)
  else
    { from_content: fact[:content], to_id: entry[:id], relation_type: 'similar_to', confidence: fb_conf }
  end
rescue StandardError => e
  log.warn("GAS classify_relation failed: #{e.message}")
  { from_content: fact[:content], to_id: entry[:id], relation_type: 'similar_to', confidence: fallback_confidence }
end

.fallback_confidence ⇒ Object 



30
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 30

def fallback_confidence     = Helpers::Confidence.apollo_setting(:gas, :fallback_confidence, default: 0.5)

.fallback_relation(fact, entry) ⇒ Object 



261
262
263
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 261

def fallback_relation(fact, entry)
  { from_content: fact[:content], to_id: entry[:id], relation_type: 'similar_to', confidence: fallback_confidence }
end

.fetch_similar_entries(facts) ⇒ Object 



183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 183

def fetch_similar_entries(facts)
  lim = similar_entries_limit
  min_conf = Helpers::GraphQuery.default_query_min_confidence
  entries = []
  facts.each do |fact|
    result = Runners::Knowledge.retrieve_relevant(query: fact[:content], limit: lim, min_confidence: min_conf)
    entries.concat(result[:entries]) if result[:success] && result[:entries]&.any?
  rescue StandardError => e
    log.warn("GAS fetch_similar_entries failed for fact: #{e.message}")
    next
  end
  unique = entries.uniq { |e| e[:id] }
  log.debug("GAS fetch_similar_entries facts=#{facts.size} entries=#{unique.size}")
  unique
end

.geometric_mean(values) ⇒ Object 



326
327
328
329
330
331
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 326

def geometric_mean(values)
  return 0.0 if values.empty?

  product = values.reduce(1.0) { |acc, v| acc * v }
  product**(1.0 / values.length)
end

.json_load(str) ⇒ Object 



22
23
24
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 22

def json_load(str)
  ::JSON.parse(str, symbolize_names: true)
end

.llm_anticipate(facts) ⇒ Object 



333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 333

def llm_anticipate(facts)
  facts_text = facts.map { |f| "(#{f[:content_type]}) #{f[:content]}" }.join("\n")

  prompt = <<~PROMPT
    Given these knowledge entries, generate 1-3 likely follow-up questions a user might ask.

    Knowledge:
    #{facts_text}

    Return JSON with a "questions" array of question strings.
  PROMPT

  result = Legion::LLM::Pipeline::GaiaCaller.structured(
    message: prompt.strip,
    schema:  {
      type:       :object,
      properties: {
        questions: { type: :array, items: { type: :string } }
      },
      required:   ['questions']
    },
    phase:   'gas_anticipate'
  )

  content = result.respond_to?(:message) ? result.message[:content] : result.to_s
  parsed = json_load(content)
  questions = parsed.is_a?(Hash) ? (parsed[:questions] || parsed['questions'] || []) : []
  questions = questions.first(max_anticipations)

  questions.map do |q|
    promote_to_pattern_store(question: q, facts: facts)
    { question: q }
  end
rescue StandardError => e
  log.warn("GAS llm_anticipate failed: #{e.message}")
  []
end

.llm_available? ⇒ Boolean

Returns:

  • (Boolean) 


384
385
386
387
388
389
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 384

def llm_available?
  defined?(Legion::LLM::Pipeline::GaiaCaller)
rescue StandardError => e
  log.warn("GAS llm_available? check failed: #{e.message}")
  false
end

.llm_classify_relation(fact, entry) ⇒ Object 



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
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 211

def llm_classify_relation(fact, entry)
  prompt = <<~PROMPT
    Classify the relationship between these two knowledge entries.
    Valid types: #{RELATION_TYPES.join(', ')}

    Entry A (new): #{fact[:content]}
    Entry B (existing): #{entry[:content]}

    Return JSON with relation_type and confidence (0.0-1.0).
  PROMPT

  result = Legion::LLM::Pipeline::GaiaCaller.structured(
    message: prompt.strip,
    schema:  {
      type:       :object,
      properties: {
        relations: {
          type:  :array,
          items: {
            type:       :object,
            properties: {
              relation_type: { type: :string },
              confidence:    { type: :number }
            },
            required:   %w[relation_type confidence]
          }
        }
      },
      required:   ['relations']
    },
    phase:   'gas_relate'
  )

  content = result.respond_to?(:message) ? result.message[:content] : result.to_s
  parsed = json_load(content)
  rels = parsed.is_a?(Hash) ? (parsed[:relations] || parsed['relations'] || []) : []
  best = rels.max_by { |r| r[:confidence] || r['confidence'] || 0 }

  return fallback_relation(fact, entry) unless best

  conf = best[:confidence] || best['confidence'] || 0
  rtype = best[:relation_type] || best['relation_type']
  return fallback_relation(fact, entry) if conf < relate_confidence_gate || !RELATION_TYPES.include?(rtype)

  { from_content: fact[:content], to_id: entry[:id], relation_type: rtype, confidence: conf }
rescue StandardError => e
  log.warn("GAS llm_classify_relation failed: #{e.message}")
  fallback_relation(fact, entry)
end

.llm_comprehend(messages, response) ⇒ Object 



395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 395

def llm_comprehend(messages, response)
  prompt = <<~PROMPT
    Extract distinct facts from this exchange. Return JSON array of {content:, content_type:} where content_type is one of: fact, concept, procedure, association.

    User: #{messages.last&.dig(:content)}
    Assistant: #{response}
  PROMPT

  result = Legion::LLM::Pipeline::GaiaCaller.structured(
    message: prompt.strip,
    schema:  {
      type:       :object,
      properties: {
        facts: {
          type:  :array,
          items: {
            type:       :object,
            properties: {
              content:      { type: :string },
              content_type: { type: :string }
            },
            required:   %w[content content_type]
          }
        }
      },
      required:   ['facts']
    },
    phase:   'gas_comprehend'
  )

  content = result.respond_to?(:message) ? result.message[:content] : result.to_s
  parsed = json_load(content)
  facts_array = parsed.is_a?(Hash) ? (parsed[:facts] || parsed['facts'] || []) : Array(parsed)
  facts_array.map do |f|
    {
      content:      f[:content] || f['content'],
      content_type: (f[:content_type] || f['content_type'] || 'fact').to_sym
    }
  end
rescue StandardError => e
  log.warn("GAS llm_comprehend failed: #{e.message}")
  mechanical_comprehend(messages, response)
end

.llm_synthesize(facts) ⇒ Object 



265
266
267
268
269
270
271
272
273
274
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
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 265

def llm_synthesize(facts)
  facts_text = facts.each_with_index.map { |f, i| "[#{i}] (#{f[:content_type]}) #{f[:content]}" }.join("\n")

  prompt = <<~PROMPT
    Given these knowledge entries, generate derivative insights (inferences, implications, or connections).
    Each synthesis should combine information from multiple sources.

    Entries:
    #{facts_text}

    Return JSON with a "synthesis" array where each item has: content (string), content_type (inference/implication/connection), source_indices (array of entry indices used).
  PROMPT

  result = Legion::LLM::Pipeline::GaiaCaller.structured(
    message: prompt.strip,
    schema:  {
      type:       :object,
      properties: {
        synthesis: {
          type:  :array,
          items: {
            type:       :object,
            properties: {
              content:        { type: :string },
              content_type:   { type: :string },
              source_indices: { type: :array, items: { type: :integer } }
            },
            required:   %w[content content_type source_indices]
          }
        }
      },
      required:   ['synthesis']
    },
    phase:   'gas_synthesize'
  )

  content = result.respond_to?(:message) ? result.message[:content] : result.to_s
  parsed = json_load(content)
  items = parsed.is_a?(Hash) ? (parsed[:synthesis] || parsed['synthesis'] || []) : []

  items.map { |item| build_synthesis_entry(item, facts) }
rescue StandardError => e
  log.warn("GAS llm_synthesize failed: #{e.message}")
  []
end

.max_anticipations ⇒ Object 



28
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 28

def max_anticipations       = Helpers::Confidence.apollo_setting(:gas, :max_anticipations, default: MAX_ANTICIPATIONS)

.mechanical_comprehend(_messages, response) ⇒ Object 



391
392
393
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 391

def mechanical_comprehend(_messages, response)
  [{ content: response, content_type: :observation }]
end

.phase_anticipate(facts, _synthesis) ⇒ Object

Phase 6: Anticipate - pre-cache likely follow-up questions



165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 165

def phase_anticipate(facts, _synthesis)
  if facts.empty?
    log.debug('GAS phase_anticipate skipped reason=no_facts')
    return []
  end
  unless llm_available?
    log.debug('GAS phase_anticipate skipped reason=llm_unavailable')
    return []
  end

  anticipations = llm_anticipate(facts)
  log.debug("GAS phase_anticipate anticipations=#{anticipations.size}")
  anticipations
rescue StandardError => e
  log.warn("GAS phase_anticipate failed: #{e.message}")
  []
end

.phase_comprehend(audit_event) ⇒ Object

Phase 1: Comprehend - extract typed facts from the exchange



68
69
70
71
72
73
74
75
76
77
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 68

def phase_comprehend(audit_event)
  messages = audit_event[:messages]
  response = audit_event[:response_content]

  mode = llm_available? ? :llm : :mechanical
  log.debug("GAS phase_comprehend mode=#{mode} messages=#{Array(messages).size} response_length=#{response.to_s.length}")
  facts = mode == :llm ? llm_comprehend(messages, response) : mechanical_comprehend(messages, response)
  log.debug("GAS phase_comprehend facts=#{facts.size}")
  facts
end

.phase_deposit(facts, _entities, _relations, _synthesis, audit_event) ⇒ Object

Phase 5: Deposit - atomic write to Apollo



139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 139

def phase_deposit(facts, _entities, _relations, _synthesis, audit_event)
  unless defined?(Runners::Knowledge)
    log.debug('GAS phase_deposit skipped reason=knowledge_runner_unavailable')
    return { deposited: 0 }
  end

  deposited = 0
  facts.each do |fact|
    Runners::Knowledge.handle_ingest(
      content:          fact[:content],
      content_type:     fact[:content_type].to_s,
      tags:             %w[gas auto_extracted],
      source_agent:     'gas_pipeline',
      source_provider:  audit_event.dig(:routing, :provider)&.to_s,
      knowledge_domain: 'general',
      context:          { source_request_id: audit_event[:request_id] }
    )
    deposited += 1
  rescue StandardError => e
    log.warn("GAS deposit error: #{e.message}")
  end
  log.info("GAS phase_deposit deposited=#{deposited} facts=#{facts.size}")
  { deposited: deposited }
end

.phase_extract(audit_event, _facts) ⇒ Object

Phase 2: Extract - entity extraction (delegates to existing EntityExtractor)



80
81
82
83
84
85
86
87
88
89
90
91
92
93
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 80

def phase_extract(audit_event, _facts)
  unless defined?(Runners::EntityExtractor)
    log.debug('GAS phase_extract skipped reason=entity_extractor_unavailable')
    return []
  end

  result = Runners::EntityExtractor.extract_entities(text: audit_event[:response_content])
  entities = result[:success] ? (result[:entities] || []) : []
  log.debug("GAS phase_extract success=#{result[:success]} entities=#{entities.size}")
  entities
rescue StandardError => e
  log.warn("GAS phase_extract failed: #{e.message}")
  []
end

.phase_relate(facts, _entities) ⇒ Object

Phase 3: Relate - classify relationships between new and existing entries



96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 96

def phase_relate(facts, _entities)
  unless defined?(Runners::Knowledge)
    log.debug('GAS phase_relate skipped reason=knowledge_runner_unavailable')
    return []
  end

  existing = fetch_similar_entries(facts)
  if existing.empty?
    log.debug("GAS phase_relate skipped reason=no_existing_entries facts=#{facts.size}")
    return []
  end

  relations = []
  facts.each do |fact|
    existing.each do |entry|
      relation = classify_relation(fact, entry)
      relations << relation if relation
    end
  end
  log.debug("GAS phase_relate facts=#{facts.size} existing=#{existing.size} relations=#{relations.size}")
  relations
end

.phase_synthesize(facts, _relations) ⇒ Object

Phase 4: Synthesize - generate derivative knowledge



120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 120

def phase_synthesize(facts, _relations)
  if facts.length < 2
    log.debug("GAS phase_synthesize skipped reason=insufficient_facts facts=#{facts.length}")
    return []
  end
  unless llm_available?
    log.debug('GAS phase_synthesize skipped reason=llm_unavailable')
    return []
  end

  synthesis = llm_synthesize(facts)
  log.debug("GAS phase_synthesize synthesis=#{synthesis.size}")
  synthesis
rescue StandardError => e
  log.warn("GAS phase_synthesize failed: #{e.message}")
  []
end

.process(audit_event) ⇒ Object 



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
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 32

def process(audit_event)
  unless processable?(audit_event)
    log.debug('GAS process skipped reason=no_content')
    return { phases_completed: 0, reason: 'no content' }
  end

  log.debug("GAS process start request_id=#{audit_event[:request_id] || 'nil'} messages=#{Array(audit_event[:messages]).size} response_length=#{audit_event[:response_content].to_s.length}") # rubocop:disable Layout/LineLength

  facts = phase_comprehend(audit_event)
  entities = phase_extract(audit_event, facts)
  relations = phase_relate(facts, entities)
  synthesis = phase_synthesize(facts, relations)
  deposit_result = phase_deposit(facts, entities, relations, synthesis, audit_event)
  anticipations = phase_anticipate(facts, synthesis)

  result = {
    phases_completed: 6,
    facts:            facts.length,
    entities:         entities.length,
    relations:        relations.length,
    synthesis:        synthesis.length,
    deposited:        deposit_result,
    anticipations:    anticipations.length
  }
  log.info("GAS process complete facts=#{result[:facts]} entities=#{result[:entities]} relations=#{result[:relations]} synthesis=#{result[:synthesis]} anticipations=#{result[:anticipations]}") # rubocop:disable Layout/LineLength
  result
rescue StandardError => e
  log.error("GAS pipeline error: #{e.message}")
  { phases_completed: 0, error: e.message }
end

.processable?(event) ⇒ Boolean

Returns:

  • (Boolean) 


63
64
65
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 63

def processable?(event)
  event[:messages]&.any? == true && !event[:response_content].nil?
end

.promote_to_pattern_store(question:, facts:) ⇒ Object 



371
372
373
374
375
376
377
378
379
380
381
382
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 371

def promote_to_pattern_store(question:, facts:)
  return unless defined?(Legion::Extensions::Agentic::TBI::PatternStore)

  Legion::Extensions::Agentic::TBI::PatternStore.promote_candidate(
    intent:     question,
    resolution: { source: 'gas_anticipate', facts: facts.map { |f| f[:content] } },
    confidence: fallback_confidence
  )
rescue StandardError => e
  log.warn("GAS promote_to_pattern_store failed: #{e.message}")
  nil
end

.relate_confidence_gate ⇒ Object 



26
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 26

def relate_confidence_gate = Helpers::Confidence.apollo_setting(:gas, :relate_confidence_gate, default: RELATE_CONFIDENCE_GATE)

.similar_entries_limit ⇒ Object 



29
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 29

def similar_entries_limit   = Helpers::Confidence.apollo_setting(:gas, :similar_entries_limit, default: 3)

.synthesis_confidence_cap ⇒ Object 



27
 
# File 'lib/legion/extensions/apollo/runners/gas.rb', line 27

def synthesis_confidence_cap = Helpers::Confidence.apollo_setting(:gas, :synthesis_confidence_cap, default: SYNTHESIS_CONFIDENCE_CAP)