Module: Legion::LLM

Extended by:
Legion::Logging::Helper
Defined in:
lib/legion/llm.rb,
lib/legion/llm/api.rb,
lib/legion/llm/call.rb,
lib/legion/llm/audit.rb,
lib/legion/llm/cache.rb,
lib/legion/llm/fleet.rb,
lib/legion/llm/hooks.rb,
lib/legion/llm/tools.rb,
lib/legion/llm/types.rb,
lib/legion/llm/compat.rb,
lib/legion/llm/config.rb,
lib/legion/llm/errors.rb,
lib/legion/llm/helper.rb,
lib/legion/llm/router.rb,
lib/legion/llm/skills.rb,
lib/legion/llm/context.rb,
lib/legion/llm/quality.rb,
lib/legion/llm/version.rb,
lib/legion/llm/api/auth.rb,
lib/legion/llm/metering.rb,
lib/legion/llm/settings.rb,
lib/legion/llm/discovery.rb,
lib/legion/llm/inference.rb,
lib/legion/llm/inventory.rb,
lib/legion/llm/transport.rb,
lib/legion/llm/fleet/lane.rb,
lib/legion/llm/scheduling.rb,
lib/legion/llm/router/rule.rb,
lib/legion/llm/skills/base.rb,
lib/legion/llm/types/chunk.rb,
lib/legion/llm/call/dispatch.rb,
lib/legion/llm/call/registry.rb,
lib/legion/llm/fleet/handler.rb,
lib/legion/llm/skills/errors.rb,
lib/legion/llm/tools/special.rb,
lib/legion/llm/types/message.rb,
lib/legion/llm/cache/response.rb,
lib/legion/llm/call/providers.rb,
lib/legion/llm/hooks/metering.rb,
lib/legion/llm/metering/usage.rb,
lib/legion/llm/api/native/chat.rb,
lib/legion/llm/call/embeddings.rb,
lib/legion/llm/caller_identity.rb,
lib/legion/llm/context/curator.rb,
lib/legion/llm/hooks/rag_guard.rb,
lib/legion/llm/inference/steps.rb,
lib/legion/llm/metering/tokens.rb,
lib/legion/llm/quality/checker.rb,
lib/legion/llm/skills/registry.rb,
lib/legion/llm/types/tool_call.rb,
lib/legion/llm/api/native/tiers.rb,
lib/legion/llm/discovery/system.rb,
lib/legion/llm/fleet/dispatcher.rb,
lib/legion/llm/hooks/reflection.rb,
lib/legion/llm/inference/prompt.rb,
lib/legion/llm/metering/tracker.rb,
lib/legion/llm/router/arbitrage.rb,
lib/legion/llm/scheduling/batch.rb,
lib/legion/llm/token_estimation.rb,
lib/legion/llm/tools/confidence.rb,
lib/legion/llm/tools/dispatcher.rb,
lib/legion/llm/api/native/models.rb,
lib/legion/llm/api/openai/models.rb,
lib/legion/llm/hooks/reciprocity.rb,
lib/legion/llm/inference/profile.rb,
lib/legion/llm/inference/request.rb,
lib/legion/llm/inference/tracing.rb,
lib/legion/llm/router/resolution.rb,
lib/legion/llm/tools/interceptor.rb,
lib/legion/llm/transport/message.rb,
lib/legion/llm/api/native/helpers.rb,
lib/legion/llm/api/native/routing.rb,
lib/legion/llm/api/shared_helpers.rb,
lib/legion/llm/call/daemon_client.rb,
lib/legion/llm/context/compressor.rb,
lib/legion/llm/fleet/token_issuer.rb,
lib/legion/llm/hooks/budget_guard.rb,
lib/legion/llm/inference/executor.rb,
lib/legion/llm/inference/response.rb,
lib/legion/llm/inference/timeline.rb,
lib/legion/llm/metering/estimator.rb,
lib/legion/llm/publisher_identity.rb,
lib/legion/llm/skills/disk_loader.rb,
lib/legion/llm/skills/step_result.rb,
lib/legion/llm/hooks/cost_tracking.rb,
lib/legion/llm/quality/shadow_eval.rb,
lib/legion/llm/scheduling/off_peak.rb,
lib/legion/llm/types/content_block.rb,
lib/legion/llm/api/native/inference.rb,
lib/legion/llm/api/native/instances.rb,
lib/legion/llm/api/native/offerings.rb,
lib/legion/llm/api/native/providers.rb,
lib/legion/llm/api/openai/responses.rb,
lib/legion/llm/call/lex_llm_adapter.rb,
lib/legion/llm/hooks/response_guard.rb,
lib/legion/llm/inference/steps/rbac.rb,
lib/legion/llm/vector_store/storage.rb,
lib/legion/llm/api/openai/embeddings.rb,
lib/legion/llm/discovery/memory_gate.rb,
lib/legion/llm/fleet/token_validator.rb,
lib/legion/llm/inference/gaia_caller.rb,
lib/legion/llm/router/health_tracker.rb,
lib/legion/llm/types/tool_definition.rb,
lib/legion/llm/api/anthropic/messages.rb,
lib/legion/llm/api/namespaces/helpers.rb,
lib/legion/llm/call/structured_output.rb,
lib/legion/llm/fleet/reply_dispatcher.rb,
lib/legion/llm/fleet/worker_execution.rb,
lib/legion/llm/inference/conversation.rb,
lib/legion/llm/inference/steps/debate.rb,
lib/legion/llm/inference/steps/billing.rb,
lib/legion/llm/inference/steps/logging.rb,
lib/legion/llm/router/escalation/chain.rb,
lib/legion/llm/skills/skill_run_result.rb,
lib/legion/llm/discovery/rule_generator.rb,
lib/legion/llm/fleet/provider_responder.rb,
lib/legion/llm/inference/route_attempts.rb,
lib/legion/llm/inference/steps/metering.rb,
lib/legion/llm/quality/confidence/score.rb,
lib/legion/llm/inference/audit_publisher.rb,
lib/legion/llm/inference/steps/rag_guard.rb,
lib/legion/llm/inference/tool_dispatcher.rb,
lib/legion/llm/quality/confidence/scorer.rb,
lib/legion/llm/router/escalation/history.rb,
lib/legion/llm/router/escalation/tracker.rb,
lib/legion/llm/skills/external_discovery.rb,
lib/legion/llm/transport/exchanges/audit.rb,
lib/legion/llm/transport/exchanges/fleet.rb,
lib/legion/llm/api/namespaces/native/chat.rb,
lib/legion/llm/inference/native_tool_loop.rb,
lib/legion/llm/inference/steps/tool_calls.rb,
lib/legion/llm/api/namespaces/native/tiers.rb,
lib/legion/llm/api/namespaces/openai/files.rb,
lib/legion/llm/api/namespaces/registration.rb,
lib/legion/llm/api/openai/chat_completions.rb,
lib/legion/llm/inference/steps/rag_context.rb,
lib/legion/llm/api/namespaces/native/models.rb,
lib/legion/llm/api/namespaces/openai/images.rb,
lib/legion/llm/api/namespaces/openai/models.rb,
lib/legion/llm/inference/steps/prompt_cache.rb,
lib/legion/llm/inference/steps/token_budget.rb,
lib/legion/llm/inference/steps/tool_history.rb,
lib/legion/llm/transport/exchanges/metering.rb,
lib/legion/llm/api/namespaces/native/routing.rb,
lib/legion/llm/api/namespaces/openai/batches.rb,
lib/legion/llm/api/namespaces/openai/uploads.rb,
lib/legion/llm/inference/enrichment_injector.rb,
lib/legion/llm/inference/steps/gaia_advisory.rb,
lib/legion/llm/inference/steps/mcp_discovery.rb,
lib/legion/llm/inference/steps/post_response.rb,
lib/legion/llm/inference/steps/tier_assigner.rb,
lib/legion/llm/inference/steps/trigger_match.rb,
lib/legion/llm/transport/messages/tool_event.rb,
lib/legion/llm/api/namespaces/anthropic/files.rb,
lib/legion/llm/api/translators/openai_request.rb,
lib/legion/llm/inference/steps/classification.rb,
lib/legion/llm/inference/steps/skill_injector.rb,
lib/legion/llm/inference/steps/span_annotator.rb,
lib/legion/llm/inference/steps/sticky_helpers.rb,
lib/legion/llm/inference/steps/sticky_persist.rb,
lib/legion/llm/inference/steps/sticky_runners.rb,
lib/legion/llm/inference/steps/tool_discovery.rb,
lib/legion/llm/tools/interceptors/python_venv.rb,
lib/legion/llm/transport/exchanges/escalation.rb,
lib/legion/llm/transport/messages/audit_event.rb,
lib/legion/llm/transport/messages/fleet_error.rb,
lib/legion/llm/transport/messages/skill_event.rb,
lib/legion/llm/api/namespaces/anthropic/models.rb,
lib/legion/llm/api/namespaces/native/inference.rb,
lib/legion/llm/api/namespaces/native/instances.rb,
lib/legion/llm/api/namespaces/native/offerings.rb,
lib/legion/llm/api/namespaces/native/providers.rb,
lib/legion/llm/api/namespaces/openai/responses.rb,
lib/legion/llm/api/translators/openai_response.rb,
lib/legion/llm/transport/messages/prompt_event.rb,
lib/legion/llm/api/namespaces/openai/embeddings.rb,
lib/legion/llm/transport/messages/fleet_request.rb,
lib/legion/llm/api/namespaces/anthropic/messages.rb,
lib/legion/llm/api/namespaces/openai/completions.rb,
lib/legion/llm/api/namespaces/openai/moderations.rb,
lib/legion/llm/api/translators/anthropic_request.rb,
lib/legion/llm/inference/steps/knowledge_capture.rb,
lib/legion/llm/transport/messages/fleet_response.rb,
lib/legion/llm/transport/messages/metering_event.rb,
lib/legion/llm/api/namespaces/openai/audio/speech.rb,
lib/legion/llm/api/translators/anthropic_response.rb,
lib/legion/llm/inference/steps/confidence_scoring.rb,
lib/legion/llm/api/namespaces/openai/chat/messages.rb,
lib/legion/llm/api/namespaces/openai/conversations.rb,
lib/legion/llm/api/namespaces/openai/uploads/parts.rb,
lib/legion/llm/api/namespaces/openai/vector_stores.rb,
lib/legion/llm/transport/messages/escalation_event.rb,
lib/legion/llm/api/namespaces/openai/chat/completions.rb,
lib/legion/llm/api/namespaces/openai/audio/translations.rb,
lib/legion/llm/api/namespaces/anthropic/messages/batches.rb,
lib/legion/llm/api/namespaces/openai/conversations/items.rb,
lib/legion/llm/api/namespaces/openai/vector_stores/files.rb,
lib/legion/llm/api/namespaces/openai/audio/transcriptions.rb,
lib/legion/llm/api/namespaces/anthropic/messages/count_tokens.rb,
lib/legion/llm/api/namespaces/openai/vector_stores/file_batches.rb

Defined Under Namespace

Modules: API, Audit, Cache, Call, CallerIdentity, CompatWarning, Config, Context, Discovery, EscalationHistory, EscalationTracker, Fleet, Helper, Hooks, Inference, Inventory, Metering, PublisherIdentity, Quality, Router, Scheduling, Settings, Skills, TokenEstimation, Tools, Transport, Types, VectorStore Classes: AuthError, ContextOverflow, DaemonDeniedError, DaemonRateLimitedError, DaemonUnavailableError, EmbeddingUnavailableError, EscalationExhausted, LLMError, PipelineError, PrivacyModeError, ProviderDown, ProviderError, RateLimitError, TokenBudgetExceeded, UnsupportedCapability, Usage

Constant Summary collapse

Routes = 
API
VERSION = 
'0.12.14'

Class Method Summary collapse

Class Method Details

.agent(agent_class) ⇒ Object 



186
 
# File 'lib/legion/llm.rb', line 186

def agent(agent_class, **) = agent_class.new(**)

.ask ⇒ Object 



136
 
# File 'lib/legion/llm.rb', line 136

def ask(...) = Inference.ask(...)

.can_embed? ⇒ Boolean

These methods check Discovery first, then fall back to instance ivars set directly on LLM (ivar fallback preserves backwards compat for specs that do Legion::LLM.instance_variable_set)

Returns:

  • (Boolean) 


166
167
168
 
# File 'lib/legion/llm.rb', line 166

def can_embed?
  Discovery.can_embed? || @can_embed == true
end

.chat ⇒ Object 



135
 
# File 'lib/legion/llm.rb', line 135

def chat(...) = Inference.chat(...)

.chat_direct ⇒ Object 



137
 
# File 'lib/legion/llm.rb', line 137

def chat_direct(...) = Inference.chat_direct(...)

.const_missing(name) ⇒ Object 



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
103
104
105
 
# File 'lib/legion/llm/compat.rb', line 25

def self.const_missing(name)
  case name
  when :Pipeline
    CompatWarning.warn_once('Legion::LLM::Pipeline', 'Legion::LLM::Inference')
    Inference
  when :ConversationStore
    CompatWarning.warn_once('Legion::LLM::ConversationStore', 'Legion::LLM::Inference::Conversation')
    Inference::Conversation
  when :NativeDispatch
    CompatWarning.warn_once('Legion::LLM::NativeDispatch', 'Legion::LLM::Call::Dispatch')
    Call::Dispatch
  when :NativeResponseAdapter
    CompatWarning.warn_once('Legion::LLM::NativeResponseAdapter', 'Legion::LLM::Call::NativeResponseAdapter')
    Call::NativeResponseAdapter
  when :ProviderRegistry
    CompatWarning.warn_once('Legion::LLM::ProviderRegistry', 'Legion::LLM::Call::Registry')
    Call::Registry
  when :CostEstimator
    CompatWarning.warn_once('Legion::LLM::CostEstimator', 'Legion::LLM::Metering::Pricing')
    Metering::Pricing
  when :CostTracker
    CompatWarning.warn_once('Legion::LLM::CostTracker', 'Legion::LLM::Metering::Recorder')
    Metering::Recorder
  when :TokenTracker
    CompatWarning.warn_once('Legion::LLM::TokenTracker', 'Legion::LLM::Metering::Tokens')
    Metering::Tokens
  when :QualityChecker
    CompatWarning.warn_once('Legion::LLM::QualityChecker', 'Legion::LLM::Quality::Checker')
    Quality::Checker
  when :ConfidenceScorer
    CompatWarning.warn_once('Legion::LLM::ConfidenceScorer', 'Legion::LLM::Quality::Confidence::Scorer')
    Quality::Confidence::Scorer
  when :ConfidenceScore
    CompatWarning.warn_once('Legion::LLM::ConfidenceScore', 'Legion::LLM::Quality::Confidence::Score')
    Quality::Confidence::Score
  when :OverrideConfidence
    CompatWarning.warn_once('Legion::LLM::OverrideConfidence', 'Legion::LLM::Tools::Confidence')
    Tools::Confidence
  when :ResponseCache
    CompatWarning.warn_once('Legion::LLM::ResponseCache', 'Legion::LLM::Cache::Response')
    Cache::Response
  when :Compressor
    CompatWarning.warn_once('Legion::LLM::Compressor', 'Legion::LLM::Context::Compressor')
    Context::Compressor
  when :DaemonClient
    CompatWarning.warn_once('Legion::LLM::DaemonClient', 'Legion::LLM::Call::DaemonClient')
    Call::DaemonClient
  when :Providers
    CompatWarning.warn_once('Legion::LLM::Providers', 'Legion::LLM::Call::Providers')
    Call::Providers
  when :Prompt
    CompatWarning.warn_once('Legion::LLM::Prompt', 'Legion::LLM::Inference::Prompt')
    Inference::Prompt
  when :ShadowEval
    CompatWarning.warn_once('Legion::LLM::ShadowEval', 'Legion::LLM::Quality::ShadowEval')
    Quality::ShadowEval
  when :Arbitrage
    CompatWarning.warn_once('Legion::LLM::Arbitrage', 'Legion::LLM::Router::Arbitrage')
    Router::Arbitrage
  when :Batch
    CompatWarning.warn_once('Legion::LLM::Batch', 'Legion::LLM::Scheduling::Batch')
    Scheduling::Batch
  when :ContextCurator
    CompatWarning.warn_once('Legion::LLM::ContextCurator', 'Legion::LLM::Context::Curator')
    Context::Curator
  when :Embeddings
    CompatWarning.warn_once('Legion::LLM::Embeddings', 'Legion::LLM::Call::Embeddings')
    Call::Embeddings
  when :OffPeak
    CompatWarning.warn_once('Legion::LLM::OffPeak', 'Legion::LLM::Scheduling::OffPeak')
    Scheduling::OffPeak
  when :InferenceError
    CompatWarning.warn_once('Legion::LLM::InferenceError', 'Legion::LLM::PipelineError')
    PipelineError
  when :Routes, :API
    require_relative '../llm/api'
    const_get(name)
  else
    super
  end
end

.embed(text) ⇒ Object 



139
140
141
142
143
144
145
146
147
 
# File 'lib/legion/llm.rb', line 139

def embed(text, **)
  if defined?(Legion::Telemetry::OpenInference)
    Legion::Telemetry::OpenInference.embedding_span(
      model: (Legion::Settings[:llm][:default_model] || 'unknown').to_s
    ) { |_span| Call::Embeddings.generate(text: text, **) }
  else
    Call::Embeddings.generate(text: text, **)
  end
end

.embed_batch(texts) ⇒ Object 



150
 
# File 'lib/legion/llm.rb', line 150

def embed_batch(texts, **) = Call::Embeddings.generate_batch(texts: texts, **)

.embed_direct(text) ⇒ Object 



149
 
# File 'lib/legion/llm.rb', line 149

def embed_direct(text, **) = Call::Embeddings.generate(text: text, **)

.embedding_fallback_chain ⇒ Object 



182
183
184
 
# File 'lib/legion/llm.rb', line 182

def embedding_fallback_chain
  Discovery.embedding_fallback_chain || @embedding_fallback_chain
end

.embedding_instance ⇒ Object 



178
179
180
 
# File 'lib/legion/llm.rb', line 178

def embedding_instance
  Discovery.embedding_instance || @embedding_instance
end

.embedding_model ⇒ Object 



174
175
176
 
# File 'lib/legion/llm.rb', line 174

def embedding_model
  Discovery.embedding_model || @embedding_model
end

.embedding_provider ⇒ Object 



170
171
172
 
# File 'lib/legion/llm.rb', line 170

def embedding_provider
  Discovery.embedding_provider || @embedding_provider
end

.settings ⇒ Object 



131
132
133
 
# File 'lib/legion/llm.rb', line 131

def settings
  Legion::Settings[:llm]
end

.shutdown ⇒ Object 



107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
 
# File 'lib/legion/llm.rb', line 107

def shutdown
  log.debug '[llm] shutdown.enter'
  Legion::Settings[:llm][:connected] = false
  @started = false
  Discovery.reset!
  Call::Registry.reset!
  # Clear LLM-level embedding ivars that may have been set via instance_variable_set for testing
  @can_embed = nil
  @embedding_provider = nil
  @embedding_model = nil
  @embedding_instance = nil
  @embedding_fallback_chain = nil
  # Gracefully shut down the async thread pool (curation, reflection, knowledge capture)
  if (pool = Inference::Executor::ASYNC_THREAD_POOL).running?
    pool.shutdown
    pool.wait_for_termination(5)
  end
  log.info '[llm] shut down'
end

.start ⇒ Object 



81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
 
# File 'lib/legion/llm.rb', line 81

def start
  log.debug '[llm] start.enter'
  Call::Providers.setup
  Discovery.run
  Router.populate_auto_rules(Discovery.discovered_instances) if Router.respond_to?(:populate_auto_rules)
  Discovery.detect_embedding_capability
  Config.set_defaults
  Hooks.install_defaults
  Tools::Interceptor.load_defaults

  Legion::LLM::Skills.start if defined?(Legion::LLM::Skills) && Legion::Settings[:llm][:skills][:enabled] != false

  LLM::Transport.load_all
  LLM::Fleet.load_transport
  LLM::Audit.load_transport
  LLM::Metering.load_transport

  @started = true
  Legion::Settings[:llm][:connected] = true
  log.info '[llm] started'
  API.register_routes if defined?(API)
rescue StandardError => e
  handle_exception(e, level: :error, operation: 'llm.start')
  raise
end

.started? ⇒ Boolean

Returns:

  • (Boolean) 


127
128
129
 
# File 'lib/legion/llm.rb', line 127

def started?
  @started == true
end

.structured(messages:, schema:) ⇒ Object 



152
153
154
155
156
157
158
159
160
 
# File 'lib/legion/llm.rb', line 152

def structured(messages:, schema:, **)
  if defined?(Legion::Telemetry::OpenInference)
    Legion::Telemetry::OpenInference.llm_span(
      model: (Legion::Settings[:llm][:default_model] || 'unknown').to_s, input: messages.to_s
    ) { |_span| Call::StructuredOutput.generate(messages: messages, schema: schema, **) }
  else
    Call::StructuredOutput.generate(messages: messages, schema: schema, **)
  end
end

.structured_direct(messages:, schema:) ⇒ Object 



162
 
# File 'lib/legion/llm.rb', line 162

def structured_direct(messages:, schema:, **) = Call::StructuredOutput.generate(messages: messages, schema: schema, **)