digitalkin 1.0.6.dev4__tar.gz → 1.0.6.dev6__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (286) hide show
  1. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/PKG-INFO +1 -1
  2. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/pyproject.toml +1 -1
  3. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/__version__.py +1 -1
  4. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/__init__.py +2 -1
  5. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/compaction.py +73 -27
  6. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/safe_litellm.py +39 -27
  7. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/litellm/cost.py +33 -6
  8. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/litellm/registry.py +63 -3
  9. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/litellm/settings.py +5 -0
  10. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/m2m_call_registry.py +4 -4
  11. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/module_server.py +4 -4
  12. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/server/channel.py +18 -0
  13. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin.egg-info/PKG-INFO +1 -1
  14. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/LICENSE +0 -0
  15. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/README.md +0 -0
  16. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/base_server/__init__.py +0 -0
  17. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/base_server/mock/__init__.py +0 -0
  18. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/base_server/mock/mock_pb2.py +0 -0
  19. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/base_server/mock/mock_pb2_grpc.py +0 -0
  20. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/base_server/server_async_insecure.py +0 -0
  21. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/base_server/server_async_secure.py +0 -0
  22. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/base_server/server_sync_insecure.py +0 -0
  23. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/base_server/server_sync_secure.py +0 -0
  24. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/bench_module/__init__.py +0 -0
  25. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/bench_module/echo_module.py +0 -0
  26. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/bench_module/models/__init__.py +0 -0
  27. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/bench_module/models/input.py +0 -0
  28. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/bench_module/models/output.py +0 -0
  29. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/bench_module/models/secret.py +0 -0
  30. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/bench_module/models/setup.py +0 -0
  31. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/bench_module/server.py +0 -0
  32. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/bench_module/triggers/__init__.py +0 -0
  33. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/bench_module/triggers/message_trigger.py +0 -0
  34. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/modules/__init__.py +0 -0
  35. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/modules/archetype_with_tools_module.py +0 -0
  36. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/modules/cpu_intensive_module.py +0 -0
  37. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/modules/dynamic_setup_module.py +0 -0
  38. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/modules/minimal_llm_module.py +0 -0
  39. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/modules/text_transform_module.py +0 -0
  40. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/monitoring/digitalkin_observability/__init__.py +0 -0
  41. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/monitoring/digitalkin_observability/http_server.py +0 -0
  42. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/monitoring/digitalkin_observability/interceptors.py +0 -0
  43. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/monitoring/digitalkin_observability/metrics.py +0 -0
  44. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/monitoring/digitalkin_observability/prometheus.py +0 -0
  45. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/monitoring/tests/test_metrics.py +0 -0
  46. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/redis_demo/client.py +0 -0
  47. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/redis_demo/echo_module.py +0 -0
  48. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/redis_demo/models/__init__.py +0 -0
  49. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/redis_demo/models/input.py +0 -0
  50. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/redis_demo/models/output.py +0 -0
  51. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/redis_demo/models/secret.py +0 -0
  52. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/redis_demo/models/setup.py +0 -0
  53. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/redis_demo/server.py +0 -0
  54. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/redis_demo/triggers/__init__.py +0 -0
  55. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/redis_demo/triggers/message_trigger.py +0 -0
  56. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/services/filesystem_module.py +0 -0
  57. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/examples/services/storage_module.py +0 -0
  58. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/setup.cfg +0 -0
  59. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/__init__.py +0 -0
  60. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/__init__.py +0 -0
  61. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/agno_adapter.py +0 -0
  62. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/agui_content.py +0 -0
  63. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/agui_tools.py +0 -0
  64. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/cancel_guard.py +0 -0
  65. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/db.py +0 -0
  66. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/hitl.py +0 -0
  67. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/models.py +0 -0
  68. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/module_toolkit.py +0 -0
  69. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/__init__.py +0 -0
  70. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/base.py +0 -0
  71. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/chat_history.py +0 -0
  72. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/defaults.py +0 -0
  73. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/media_viewer.py +0 -0
  74. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/module_toolkits.py +0 -0
  75. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/__init__.py +0 -0
  76. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/action.py +0 -0
  77. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/base.py +0 -0
  78. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/kins/__init__.py +0 -0
  79. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/kins/action.py +0 -0
  80. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/kins/kit.py +0 -0
  81. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/loader/__init__.py +0 -0
  82. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/loader/action.py +0 -0
  83. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/loader/kit.py +0 -0
  84. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/services/__init__.py +0 -0
  85. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/services/action.py +0 -0
  86. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/services/kit.py +0 -0
  87. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/tools/__init__.py +0 -0
  88. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/tools/action.py +0 -0
  89. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/registry/tools/kit.py +0 -0
  90. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/agno/toolkits/user_profile.py +0 -0
  91. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/langfuse/__init__.py +0 -0
  92. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/langfuse/settings.py +0 -0
  93. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/langfuse/tracing.py +0 -0
  94. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/litellm/__init__.py +0 -0
  95. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/community/litellm/model_specs.py +0 -0
  96. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/__init__.py +0 -0
  97. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/common/__init__.py +0 -0
  98. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/common/factories.py +0 -0
  99. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/exceptions.py +0 -0
  100. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/job_manager/__init__.py +0 -0
  101. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/job_manager/base_job_manager.py +0 -0
  102. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/job_manager/single_job_manager.py +0 -0
  103. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/profiling/__init__.py +0 -0
  104. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/profiling/step_timer.py +0 -0
  105. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/profiling/task_profiler.py +0 -0
  106. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/resilience/__init__.py +0 -0
  107. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/resilience/bulkhead.py +0 -0
  108. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/resilience/task_supervisor.py +0 -0
  109. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/task_manager/__init__.py +0 -0
  110. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/task_manager/base_task_manager.py +0 -0
  111. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/task_manager/local_task_manager.py +0 -0
  112. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/task_manager/module_runner.py +0 -0
  113. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/task_manager/redis/__init__.py +0 -0
  114. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/task_manager/redis/proto_streams.py +0 -0
  115. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/task_manager/redis/redis_client.py +0 -0
  116. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/task_manager/redis/redis_idempotency.py +0 -0
  117. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/task_manager/redis/redis_signal.py +0 -0
  118. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/task_manager/remote_task_manager.py +0 -0
  119. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/task_manager/task_executor.py +0 -0
  120. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/core/task_manager/task_session.py +0 -0
  121. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/exceptions.py +0 -0
  122. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/__init__.py +0 -0
  123. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/_base_server.py +0 -0
  124. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/exceptions.py +0 -0
  125. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/gateway_servicer.py +0 -0
  126. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/interceptors/__init__.py +0 -0
  127. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/interceptors/permission.py +0 -0
  128. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/interceptors/request_ids.py +0 -0
  129. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/interceptors/validation.py +0 -0
  130. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/module_servicer.py +0 -0
  131. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/stream_registry.py +0 -0
  132. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/stream_session.py +0 -0
  133. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/utils/__init__.py +0 -0
  134. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/utils/circuit_breaker.py +0 -0
  135. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/utils/grpc_client_wrapper.py +0 -0
  136. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/utils/grpc_error_handler.py +0 -0
  137. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/utils/utility_schema_extender.py +0 -0
  138. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/grpc_servers/utils/validators.py +0 -0
  139. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/logger.py +0 -0
  140. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/mixins/__init__.py +0 -0
  141. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/mixins/agui_mixin.py +0 -0
  142. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/mixins/base_mixin.py +0 -0
  143. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/mixins/cost_mixin.py +0 -0
  144. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/mixins/file_history_mixin.py +0 -0
  145. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/mixins/filesystem_mixin.py +0 -0
  146. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/mixins/logger_mixin.py +0 -0
  147. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/mixins/storage_mixin.py +0 -0
  148. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/__init__.py +0 -0
  149. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/core/__init__.py +0 -0
  150. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/core/job_manager_models.py +0 -0
  151. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/core/redis.py +0 -0
  152. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/core/task_monitor.py +0 -0
  153. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/events/__init__.py +0 -0
  154. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/events/agent_events.py +0 -0
  155. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/grpc_servers/__init__.py +0 -0
  156. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/grpc_servers/circuit_breaker.py +0 -0
  157. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/grpc_servers/m2m.py +0 -0
  158. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/grpc_servers/models.py +0 -0
  159. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/grpc_servers/stream_error_codes.py +0 -0
  160. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/grpc_servers/types.py +0 -0
  161. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/module/__init__.py +0 -0
  162. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/module/ag_ui.py +0 -0
  163. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/module/base_types.py +0 -0
  164. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/module/knowledge_files.py +0 -0
  165. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/module/loaded_tools.py +0 -0
  166. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/module/module.py +0 -0
  167. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/module/module_context.py +0 -0
  168. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/module/module_types.py +0 -0
  169. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/module/request_metadata.py +0 -0
  170. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/module/select_schema.py +0 -0
  171. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/module/setup_types.py +0 -0
  172. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/module/tool_cache.py +0 -0
  173. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/module/tool_reference.py +0 -0
  174. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/module/utility.py +0 -0
  175. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/services/__init__.py +0 -0
  176. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/services/cost.py +0 -0
  177. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/services/filesystem.py +0 -0
  178. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/services/registry.py +0 -0
  179. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/services/services.py +0 -0
  180. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/services/storage.py +0 -0
  181. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/__init__.py +0 -0
  182. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/agui.py +0 -0
  183. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/certificate.py +0 -0
  184. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/client/__init__.py +0 -0
  185. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/client/channel.py +0 -0
  186. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/client/circuit_breaker.py +0 -0
  187. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/client/client.py +0 -0
  188. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/client/grpc.py +0 -0
  189. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/gateway.py +0 -0
  190. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/log.py +0 -0
  191. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/module.py +0 -0
  192. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/profiling.py +0 -0
  193. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/queue.py +0 -0
  194. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/redis.py +0 -0
  195. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/registry.py +0 -0
  196. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/resilience.py +0 -0
  197. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/server/__init__.py +0 -0
  198. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/server/grpc.py +0 -0
  199. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/server/server.py +0 -0
  200. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/server/servicer.py +0 -0
  201. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/task_manager.py +0 -0
  202. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/utils/__init__.py +0 -0
  203. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/settings/utils/channel.py +0 -0
  204. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/utils/__init__.py +0 -0
  205. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/models/utils/dynamic_schema.py +0 -0
  206. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/modules/__init__.py +0 -0
  207. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/modules/_base_module.py +0 -0
  208. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/modules/archetype_module.py +0 -0
  209. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/modules/tool_module.py +0 -0
  210. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/modules/trigger_handler.py +0 -0
  211. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/modules/triggers/__init__.py +0 -0
  212. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/modules/triggers/healthcheck_ping_trigger.py +0 -0
  213. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/modules/triggers/healthcheck_services_trigger.py +0 -0
  214. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/modules/triggers/healthcheck_status_trigger.py +0 -0
  215. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/py.typed +0 -0
  216. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/__init__.py +0 -0
  217. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/base_strategy.py +0 -0
  218. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/communication/__init__.py +0 -0
  219. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/communication/communication_strategy.py +0 -0
  220. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/communication/default_communication.py +0 -0
  221. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/communication/exceptions.py +0 -0
  222. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/communication/grpc_communication.py +0 -0
  223. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/cost/__init__.py +0 -0
  224. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/cost/cost_strategy.py +0 -0
  225. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/cost/default_cost.py +0 -0
  226. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/cost/exceptions.py +0 -0
  227. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/cost/grpc_cost.py +0 -0
  228. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/filesystem/__init__.py +0 -0
  229. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/filesystem/default_filesystem.py +0 -0
  230. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/filesystem/exceptions.py +0 -0
  231. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/filesystem/filesystem_strategy.py +0 -0
  232. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/filesystem/grpc_filesystem.py +0 -0
  233. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/identity/__init__.py +0 -0
  234. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/identity/default_identity.py +0 -0
  235. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/identity/identity_strategy.py +0 -0
  236. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/registry/__init__.py +0 -0
  237. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/registry/default_registry.py +0 -0
  238. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/registry/exceptions.py +0 -0
  239. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/registry/grpc_registry.py +0 -0
  240. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/registry/registry_strategy.py +0 -0
  241. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/secret/__init__.py +0 -0
  242. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/secret/default_secret.py +0 -0
  243. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/secret/exceptions.py +0 -0
  244. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/secret/grpc_secret.py +0 -0
  245. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/secret/secret_strategy.py +0 -0
  246. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/services_config.py +0 -0
  247. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/services_models.py +0 -0
  248. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/setup/__init__.py +0 -0
  249. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/setup/default_setup.py +0 -0
  250. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/setup/exceptions.py +0 -0
  251. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/setup/grpc_setup.py +0 -0
  252. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/setup/setup_strategy.py +0 -0
  253. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/storage/__init__.py +0 -0
  254. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/storage/default_storage.py +0 -0
  255. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/storage/exceptions.py +0 -0
  256. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/storage/grpc_storage.py +0 -0
  257. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/storage/storage_strategy.py +0 -0
  258. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/task_manager/__init__.py +0 -0
  259. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/task_manager/default_task_manager.py +0 -0
  260. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/task_manager/exceptions.py +0 -0
  261. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/task_manager/redis_task_manager.py +0 -0
  262. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/task_manager/task_manager_strategy.py +0 -0
  263. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/user_profile/__init__.py +0 -0
  264. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/user_profile/default_user_profile.py +0 -0
  265. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/user_profile/exceptions.py +0 -0
  266. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/user_profile/grpc_user_profile.py +0 -0
  267. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/services/user_profile/user_profile_strategy.py +0 -0
  268. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/utils/__init__.py +0 -0
  269. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/utils/arg_parser.py +0 -0
  270. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/utils/conditional_schema.py +0 -0
  271. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/utils/development_mode_action.py +0 -0
  272. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/utils/dynamic_schema.py +0 -0
  273. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/utils/env_manager.py +0 -0
  274. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/utils/env_template.py +0 -0
  275. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/utils/exceptions.py +0 -0
  276. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/utils/json_structure.py +0 -0
  277. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/utils/llm_ready_schema.py +0 -0
  278. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/utils/package_discover.py +0 -0
  279. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/utils/proto_utils.py +0 -0
  280. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/utils/schema_splitter.py +0 -0
  281. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin/utils/setup_content_validator.py +0 -0
  282. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin.egg-info/SOURCES.txt +0 -0
  283. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin.egg-info/dependency_links.txt +0 -0
  284. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin.egg-info/requires.txt +0 -0
  285. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/src/digitalkin.egg-info/top_level.txt +0 -0
  286. {digitalkin-1.0.6.dev4 → digitalkin-1.0.6.dev6}/tests/test_exceptions.py +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: digitalkin
3
- Version: 1.0.6.dev4
3
+ Version: 1.0.6.dev6
4
4
  Summary: SDK to build kin used in DigitalKin
5
5
  Author-email: "DigitalKin.ai" <contact@digitalkin.ai>
6
6
  License: Attribution-NonCommercial-ShareAlike 4.0 International
@@ -39,7 +39,7 @@
39
39
  "pydantic>=2.12.4",
40
40
  "redis[hiredis]>=7.4.0,<9",
41
41
  ]
42
- version = "1.0.6.dev4"
42
+ version = "1.0.6.dev6"
43
43
 
44
44
  [project.optional-dependencies]
45
45
  agno = [ "agno>=3.1.8,<4" ]
@@ -5,4 +5,4 @@ from importlib.metadata import PackageNotFoundError, version
5
5
  try:
6
6
  __version__ = version("digitalkin")
7
7
  except PackageNotFoundError:
8
- __version__ = "1.0.6.dev4"
8
+ __version__ = "1.0.6.dev6"
@@ -30,6 +30,7 @@ therefore NOT exported here; import them directly::
30
30
  StorageDb,
31
31
  )
32
32
  from digitalkin.community.agno.compaction import (
33
+ ContextSize,
33
34
  DkCompaction,
34
35
  HistoryCompactor,
35
36
  TokenCalibration,
@@ -41,7 +42,7 @@ therefore NOT exported here; import them directly::
41
42
  - ``CancelGuard`` — tool hook refusing tool calls once the task is cancelled.
42
43
  - ``StorageDb`` (+ ``AgnoSession``, ``AgnoCompaction``, ``AGNO_STORAGE_CONFIG``) — Agno
43
44
  session storage over the module's StorageStrategy.
44
- - ``HistoryCompactor``, ``DkCompaction``, ``TokenCalibration`` — chat history compaction.
45
+ - ``HistoryCompactor``, ``DkCompaction``, ``TokenCalibration``, ``ContextSize`` — chat history compaction.
45
46
  - ``SafeLiteLLM`` — Agno LiteLLM model routed through the LiteLLM proxy.
46
47
 
47
48
  Default agent toolkits (``ChatHistoryTools``, ``UserProfileTools``, the registry
@@ -21,6 +21,7 @@ from __future__ import annotations
21
21
 
22
22
  import asyncio
23
23
  from dataclasses import dataclass, field
24
+ from enum import Enum
24
25
  from functools import lru_cache
25
26
  from typing import TYPE_CHECKING, Any, ClassVar
26
27
 
@@ -45,20 +46,25 @@ if TYPE_CHECKING:
45
46
  from digitalkin.community.litellm import ModelSpec
46
47
 
47
48
 
49
+ class ContextSize(Enum):
50
+ """Share of the model's input budget the replayed history may fill before it is folded."""
51
+
52
+ LARGE = 0.9
53
+ MEDIUM = 0.7
54
+ SMALL = 0.5
55
+
56
+
48
57
  class CompactionSettings(BaseSettings):
49
58
  """When a head model folds its replayed history into a summary, and which model writes it."""
50
59
 
51
60
  model_config = SettingsConfigDict(env_prefix="DIGITALKIN_COMPACTION_", case_sensitive=False)
52
61
 
53
62
  enabled: bool = Field(default=True, description="Compact the replayed history at the token threshold.")
54
- max_tokens: int = Field(
55
- default=600_000, gt=0, description="Context size (tokens) that triggers compaction on large-window models."
56
- )
57
63
  window_ratio: float = Field(
58
- default=0.9,
64
+ default=ContextSize.LARGE.value,
59
65
  gt=0,
60
66
  le=1,
61
- description="Share of the model's input budget the threshold is capped at on smaller models.",
67
+ description="Share of the model's input budget that triggers compaction, unless the setup picks a ContextSize.",
62
68
  )
63
69
  uncompacted_runs: int = Field(default=5, ge=0, description="Recent runs kept verbatim after a compaction.")
64
70
  summary_model: str = Field(
@@ -247,7 +253,8 @@ class HistoryCompactor:
247
253
 
248
254
  Built per task with its model, so the cached record and the "already measured" set live as
249
255
  long as the task's requests. The threshold is measured once per distinct history, i.e. once
250
- per run: within a run the replayed history does not change, only the current run grows.
256
+ per run: within a run the replayed history does not change, only the current run grows. That
257
+ growth is caught on the provider's count of each response, which folds the history mid-run.
251
258
 
252
259
  Attributes:
253
260
  compaction: The fold policy, with its summarizer model.
@@ -288,27 +295,26 @@ class HistoryCompactor:
288
295
  self._calibrated = False
289
296
  self._loaded = False
290
297
  self._measured: set[tuple[Any, ...]] = set()
298
+ # The provider's count of the last response, and the history last folded because of it.
299
+ self._last_provider_tokens: int | None = None
300
+ self._outgrown: tuple[Any, ...] | None = None
291
301
  # One model can serve concurrent calls; two of them must not both fold the same history.
292
302
  self._lock = asyncio.Lock()
293
303
 
294
304
  @staticmethod
295
- def threshold(context_window: int, max_tokens: int, window_ratio: float) -> int:
305
+ def threshold(context_window: int, window_ratio: float) -> int:
296
306
  """Context size, in tokens, at which the replayed history is folded.
297
307
 
298
- ``max_tokens`` on large-window models, ``window_ratio`` of the window on smaller ones, so a
299
- fold always comes before the provider's limit.
300
-
301
308
  Args:
302
- context_window: The model's input window; 0 when unknown.
303
- max_tokens: The threshold cap.
304
- window_ratio: Share of the window the threshold may reach.
309
+ context_window: The model's input budget; 0 when unknown.
310
+ window_ratio: Share of the budget the context may fill.
305
311
 
306
312
  Returns:
307
- The threshold in tokens.
313
+ The threshold in tokens; 200k for an unknown window.
308
314
  """
309
315
  if context_window <= 0:
310
- return max_tokens
311
- return max(1, min(max_tokens, int(context_window * window_ratio)))
316
+ return 200_000
317
+ return max(1, int(context_window * window_ratio))
312
318
 
313
319
  @staticmethod
314
320
  def split_request(messages: list[Message]) -> tuple[list[Message], list[Message], list[Message]]:
@@ -343,7 +349,6 @@ class HistoryCompactor:
343
349
  db: StorageDb,
344
350
  session_id: str,
345
351
  model_id: str,
346
- max_tokens: int,
347
352
  window_ratio: float,
348
353
  uncompacted_runs: int,
349
354
  max_output: int = 0,
@@ -360,8 +365,7 @@ class HistoryCompactor:
360
365
  db: Storage of the session's compaction record.
361
366
  session_id: The session whose history is folded.
362
367
  model_id: Id of the head model.
363
- max_tokens: Threshold cap (see ``threshold``).
364
- window_ratio: Share of a smaller window the threshold may reach.
368
+ window_ratio: Share of the input budget that triggers a fold (see ``threshold``).
365
369
  uncompacted_runs: Recent runs kept verbatim after a fold.
366
370
  max_output: Output the head model may request, reserved in its window: the provider
367
371
  rejects a request whose input plus ``max_tokens`` exceeds it.
@@ -369,7 +373,7 @@ class HistoryCompactor:
369
373
  Returns:
370
374
  The compactor.
371
375
  """
372
- threshold_tokens = cls.threshold(context_window - max_output, max_tokens, window_ratio)
376
+ threshold_tokens = cls.threshold(context_window - max_output, window_ratio)
373
377
  compaction = DkCompaction(
374
378
  model=summarizer,
375
379
  compact_at_tokens=threshold_tokens,
@@ -456,6 +460,7 @@ class HistoryCompactor:
456
460
  mission_id: str,
457
461
  extra_headers: dict[str, str],
458
462
  cancel_event: asyncio.Event | None = None,
463
+ context_size: ContextSize | None = None,
459
464
  ) -> HistoryCompactor | None:
460
465
  """The compactor of a head model, under ``CompactionSettings``.
461
466
 
@@ -467,6 +472,7 @@ class HistoryCompactor:
467
472
  mission_id: The mission the summaries' cost is booked to.
468
473
  extra_headers: The head model's proxy headers.
469
474
  cancel_event: Stops the summarizer with the run.
475
+ context_size: Fold point picked by the module's setup; ``window_ratio`` when None.
470
476
 
471
477
  Returns:
472
478
  The compactor, or None when ``DIGITALKIN_COMPACTION_ENABLED`` is off.
@@ -488,8 +494,7 @@ class HistoryCompactor:
488
494
  db=db,
489
495
  session_id=session_id,
490
496
  model_id=model_spec.name,
491
- max_tokens=settings.max_tokens,
492
- window_ratio=settings.window_ratio,
497
+ window_ratio=context_size.value if context_size is not None else settings.window_ratio,
493
498
  uncompacted_runs=settings.uncompacted_runs,
494
499
  max_output=model_spec.capacity.context_window - model_spec.capacity.input_budget,
495
500
  )
@@ -556,6 +561,27 @@ class HistoryCompactor:
556
561
  )
557
562
  if self.compaction.should_compact(in_context, context_tokens=tokens):
558
563
  record = await self._fold(prefix, history, in_context, tools, tokens, run_response) or record
564
+ elif (
565
+ self._last_provider_tokens is not None
566
+ and self._last_provider_tokens >= self.threshold_tokens
567
+ and signature != self._outgrown
568
+ ):
569
+ # Folded at most once per history: a run whose own messages fill the context
570
+ # would otherwise be summarized on every call.
571
+ self._outgrown = signature
572
+ # TODO(validate): COMPACTION-MID-RUN-FOLD a tool loop past the threshold folds the history mid-run
573
+ logger.info(
574
+ "[VALIDATE COMPACTION-MID-RUN-FOLD] Run outgrew the threshold, folding mid-run: "
575
+ "session=%s model=%s provider=%d threshold=%d",
576
+ self.session_id,
577
+ self.model_id,
578
+ self._last_provider_tokens,
579
+ self.threshold_tokens,
580
+ )
581
+ record = (
582
+ await self._fold(prefix, history, self._apply(history, record), tools, None, run_response)
583
+ or record
584
+ )
559
585
  except Exception:
560
586
  logger.exception(
561
587
  "History compaction failed, sending the history uncompacted: session=%s model=%s",
@@ -595,18 +621,36 @@ class HistoryCompactor:
595
621
  return None
596
622
  return [*prefix, *self._apply(history, new_record), *current]
597
623
 
598
- async def observe(self, sent: list[Message], tools: list[Any] | None, provider_tokens: int) -> None:
599
- """Calibrate the estimate on the provider's count of a request that was just sent.
624
+ async def observe(
625
+ self, sent: list[Message], tools: list[Any] | None, provider_tokens: int, *, calibrate: bool = True
626
+ ) -> None:
627
+ """Record the provider's count of a request that was just sent, and calibrate on it.
600
628
 
601
- Once per compactor, i.e. once per run: the first response is enough, and estimating every
602
- request of a long tool loop would cost a full tokenization each time. Never raises.
629
+ Every count is kept for the mid-run fold. The calibration runs once per compactor, i.e. once
630
+ per run: the first response is enough, and estimating every request of a long tool loop would
631
+ cost a full tokenization each time. Never raises.
603
632
 
604
633
  Args:
605
634
  sent: The messages of the request, as sent.
606
635
  tools: The request's tool definitions.
607
636
  provider_tokens: The prompt size the provider reported for it.
637
+ calibrate: False for a fallback model's response: its tokenizer is not the head model's,
638
+ and the samples are stored with the session.
608
639
  """
609
- if self._calibrated or provider_tokens <= 0:
640
+ if provider_tokens <= 0:
641
+ return
642
+ self._last_provider_tokens = provider_tokens
643
+ if self._calibrated:
644
+ return
645
+ if not calibrate:
646
+ # TODO(validate): CALIBRATION-SKIP-FALLBACK a fallback model's count never calibrates the session
647
+ logger.info(
648
+ "[VALIDATE CALIBRATION-SKIP-FALLBACK] Fallback response not used to calibrate: "
649
+ "session=%s model=%s provider=%d",
650
+ self.session_id,
651
+ self.model_id,
652
+ provider_tokens,
653
+ )
610
654
  return
611
655
  self._calibrated = True
612
656
  try: # ruff: ignore[too-many-statements-in-try-clause]
@@ -722,6 +766,8 @@ class HistoryCompactor:
722
766
  if new_record is None:
723
767
  return None
724
768
  self._record = new_record
769
+ # That count was for the unfolded request.
770
+ self._last_provider_tokens = None
725
771
  stored = await self.db.upsert_compaction_record(self.session_id, self._stored())
726
772
  logger.info(
727
773
  "History compacted: session=%s model=%s messages=%d tokens %s -> %s stored=%s",
@@ -24,7 +24,7 @@ from __future__ import annotations
24
24
 
25
25
  import re
26
26
  from contextvars import ContextVar
27
- from dataclasses import dataclass
27
+ from dataclasses import dataclass, field
28
28
  from typing import TYPE_CHECKING, Any, ClassVar, cast
29
29
  from urllib.parse import urlsplit
30
30
 
@@ -40,6 +40,7 @@ from pydantic import BaseModel
40
40
 
41
41
  from digitalkin.community.litellm.cost import CostTrackingLogger
42
42
  from digitalkin.community.litellm.registry import LiteLLMRegistry
43
+ from digitalkin.community.litellm.settings import get_litellm_settings
43
44
  from digitalkin.logger import logger
44
45
  from digitalkin.models.events import ModelInfo
45
46
 
@@ -70,6 +71,7 @@ class SafeLiteLLM(LiteLLM):
70
71
  drop_image_urls: Latched when the provider cannot fetch an image URL: later requests of this
71
72
  instance keep inline images and drop only the ones sent by URL.
72
73
  compactor: Folds the replayed history at its threshold. Head models only.
74
+ cache_control: The Anthropic cache marker, with ``ttl`` when ``LITELLM_PROMPT_CACHE_TTL`` is set.
73
75
  input_tokens: Input budget: the context window minus the output reserved by ``max_tokens``.
74
76
  The last-resort trim fits the request into it; None falls back to litellm's own
75
77
  window for the model id, unknown for most proxy model groups.
@@ -85,6 +87,7 @@ class SafeLiteLLM(LiteLLM):
85
87
  drop_images: bool = False
86
88
  drop_image_urls: bool = False
87
89
  compactor: HistoryCompactor | None = None
90
+ cache_control: dict[str, str] = field(default_factory=lambda: {"type": "ephemeral"})
88
91
  input_tokens: int | None = None
89
92
 
90
93
  # Image parts of the request the current asyncio task is sending. One model can serve two calls
@@ -150,6 +153,11 @@ class SafeLiteLLM(LiteLLM):
150
153
  )
151
154
  config = LiteLLMRegistry.shared().config
152
155
  capacity = model_spec.capacity
156
+ cache_ttl = get_litellm_settings().prompt_cache_ttl
157
+ headers = dict(extra_headers or {})
158
+ if session_id:
159
+ # The proxy's session affinity reads the header only, never the body field.
160
+ headers["x-litellm-session-id"] = session_id
153
161
  return cls(
154
162
  id=model_spec.name,
155
163
  api_base=config.base_url,
@@ -161,8 +169,9 @@ class SafeLiteLLM(LiteLLM):
161
169
  enable_prompt_caching=enable_prompt_caching,
162
170
  cancel_event=cancel_event,
163
171
  compactor=compactor,
172
+ cache_control={"type": "ephemeral", "ttl": cache_ttl} if cache_ttl else {"type": "ephemeral"},
164
173
  input_tokens=capacity.input_budget or None,
165
- extra_headers=extra_headers or {},
174
+ extra_headers=headers,
166
175
  extra_body={"litellm_session_id": session_id} if session_id else None,
167
176
  request_params=request_params,
168
177
  )
@@ -273,10 +282,9 @@ class SafeLiteLLM(LiteLLM):
273
282
  """Read which model group of the LiteLLM proxy answered, from its response headers.
274
283
 
275
284
  The proxy stamps ``x-litellm-model-group`` with the group that served the call (the
276
- fallback group after a fallback) and ``x-litellm-attempted-fallbacks`` with the fallback
277
- depth; litellm keeps them, ``llm_provider-`` prefixed, in ``_hidden_params``. The body's
285
+ fallback group after a fallback), read by :meth:`LiteLLMRegistry.served_route`. The body's
278
286
  ``model`` cannot be trusted: on a stream the proxy restamps every chunk with the requested
279
- model, fallback or not.
287
+ model, fallback or not. A fallback is a served group other than the requested one (``self.id``).
280
288
 
281
289
  Args:
282
290
  response: A litellm completion or streaming chunk.
@@ -284,19 +292,10 @@ class SafeLiteLLM(LiteLLM):
284
292
  Returns:
285
293
  The served model, or None when the response carries no proxy headers.
286
294
  """
287
- hidden = getattr(response, "_hidden_params", None)
288
- if hidden is not None and not isinstance(hidden, dict):
289
- hidden = hidden.model_dump() if hasattr(hidden, "model_dump") else None
290
- headers = (hidden or {}).get("additional_headers") or {}
291
- group = headers.get("llm_provider-x-litellm-model-group")
295
+ group, _ = LiteLLMRegistry.shared().served_route(response)
292
296
  if not group:
293
297
  return None
294
- try:
295
- fallbacks = int(headers.get("llm_provider-x-litellm-attempted-fallbacks") or 0)
296
- except (TypeError, ValueError):
297
- fallbacks = 0
298
- replaced = fallbacks > 0 and group != self.id
299
- return ModelInfo(name=group, requested_name=self.id if replaced else None)
298
+ return ModelInfo(name=group, requested_name=self.id if group != self.id else None)
300
299
 
301
300
  def _report_served_model(self, model_response: ModelResponse, response: Any, *, streamed: bool) -> None:
302
301
  """Put the served model in ``provider_data``, once per LLM call.
@@ -527,7 +526,7 @@ class SafeLiteLLM(LiteLLM):
527
526
  raise
528
527
  sent = refit
529
528
  response = await self._ainvoke_once(sent, assistant_message, kwargs)
530
- await self._observe(sent, kwargs, response.response_usage)
529
+ await self._observe(sent, kwargs, response.response_usage, response.provider_data)
531
530
  return response
532
531
 
533
532
  async def _ainvoke_stream_fitting(
@@ -549,11 +548,13 @@ class SafeLiteLLM(LiteLLM):
549
548
  """
550
549
  sent = await self._compacted(messages, kwargs)
551
550
  usage: MessageMetrics | None = None
551
+ provider_data: dict[str, Any] | None = None
552
552
  streamed = False
553
553
  try:
554
554
  async for chunk in self._ainvoke_stream_once(sent, assistant_message, kwargs):
555
555
  streamed = True
556
556
  usage = chunk.response_usage or usage
557
+ provider_data = chunk.provider_data or provider_data
557
558
  yield chunk
558
559
  except BadRequestError as exc:
559
560
  refit = None if streamed else await self._refit_after_overflow(exc, messages, kwargs)
@@ -562,19 +563,31 @@ class SafeLiteLLM(LiteLLM):
562
563
  sent = refit
563
564
  async for chunk in self._ainvoke_stream_once(sent, assistant_message, kwargs):
564
565
  usage = chunk.response_usage or usage
566
+ provider_data = chunk.provider_data or provider_data
565
567
  yield chunk
566
- await self._observe(sent, kwargs, usage)
568
+ await self._observe(sent, kwargs, usage, provider_data)
567
569
 
568
- async def _observe(self, sent: list[Message], kwargs: dict[str, Any], usage: MessageMetrics | None) -> None:
569
- """Hand the provider's prompt size of a request to the compactor, which calibrates on it.
570
+ async def _observe(
571
+ self,
572
+ sent: list[Message],
573
+ kwargs: dict[str, Any],
574
+ usage: MessageMetrics | None,
575
+ provider_data: dict[str, Any] | None,
576
+ ) -> None:
577
+ """Hand the provider's prompt size of a request to the compactor.
570
578
 
571
579
  Args:
572
580
  sent: The messages as sent.
573
581
  kwargs: The remaining ``ainvoke`` arguments.
574
582
  usage: The response usage; LiteLLM's ``prompt_tokens`` already counts cached tokens.
583
+ provider_data: The response's provider data, naming the served model; a fallback's count
584
+ is not calibrated on.
575
585
  """
576
- if self.compactor is not None and usage is not None and usage.input_tokens:
577
- await self.compactor.observe(sent, kwargs.get("tools"), usage.input_tokens)
586
+ if self.compactor is None or usage is None or not usage.input_tokens:
587
+ return
588
+ served = (provider_data or {}).get(ModelInfo.PROVIDER_DATA_KEY)
589
+ fallback = isinstance(served, dict) and served.get("requested_name") is not None
590
+ await self.compactor.observe(sent, kwargs.get("tools"), usage.input_tokens, calibrate=not fallback)
578
591
 
579
592
  async def _ainvoke_once(
580
593
  self, messages: list[Message], assistant_message: Message, kwargs: dict[str, Any]
@@ -745,7 +758,7 @@ class SafeLiteLLM(LiteLLM):
745
758
  if not self.send_temperature:
746
759
  params.pop("temperature", None)
747
760
  if self.enable_prompt_caching and tools:
748
- params["tools"] = [*tools[:-1], {**tools[-1], "cache_control": {"type": "ephemeral"}}]
761
+ params["tools"] = [*tools[:-1], {**tools[-1], "cache_control": self.cache_control}]
749
762
  return params
750
763
 
751
764
  def _format_messages(
@@ -944,13 +957,12 @@ class SafeLiteLLM(LiteLLM):
944
957
  if marked:
945
958
  last = marked[-1]
946
959
  if last.get("role") == "tool":
947
- marked[-1] = {**last, "cache_control": {"type": "ephemeral"}}
960
+ marked[-1] = {**last, "cache_control": self.cache_control}
948
961
  elif last.get("role") == "user":
949
962
  marked[-1] = {**last, "content": self._as_cached_blocks(last.get("content"))}
950
963
  return marked
951
964
 
952
- @staticmethod
953
- def _as_cached_blocks(content: Any) -> list[dict[str, Any]]:
965
+ def _as_cached_blocks(self, content: Any) -> list[dict[str, Any]]:
954
966
  """Turn message content into blocks whose last text block carries the cache marker.
955
967
 
956
968
  Args:
@@ -964,6 +976,6 @@ class SafeLiteLLM(LiteLLM):
964
976
  )
965
977
  for block in reversed(blocks):
966
978
  if block.get("type") == "text":
967
- block["cache_control"] = {"type": "ephemeral"}
979
+ block["cache_control"] = self.cache_control
968
980
  break
969
981
  return blocks
@@ -192,12 +192,29 @@ class CostTrackingLogger(CustomLogger):
192
192
  logger.debug("No usage on response; cost not tracked", extra={"mission_id": mission_id})
193
193
  return
194
194
  model = kwargs.get("model")
195
- model_name = model.removeprefix("litellm_proxy/") if isinstance(model, str) else None
196
- spec = LiteLLMRegistry.shared().try_get(model_name) if model_name else None
195
+ requested = model.removeprefix("litellm_proxy/") if isinstance(model, str) else None
196
+ registry = LiteLLMRegistry.shared()
197
+ served, deployment = registry.served_route(response_obj)
198
+ spec = registry.try_get(served) if served else None
197
199
  if spec is None:
198
- logger.warning("No pricing for model %s; cost not tracked", model_name, extra={"mission_id": mission_id})
200
+ spec = registry.try_get(requested) if requested else None
201
+ elif served != requested:
202
+ # TODO(validate): COST-SERVED-MODEL a fallback call is billed at the served model's price
203
+ logger.warning(
204
+ "[VALIDATE COST-SERVED-MODEL] Billing on the served model %s instead of the requested %s",
205
+ served,
206
+ requested,
207
+ extra={"mission_id": mission_id},
208
+ )
209
+ if spec is None:
210
+ logger.warning(
211
+ "No pricing for model %s (served %s); cost not tracked",
212
+ requested,
213
+ served,
214
+ extra={"mission_id": mission_id},
215
+ )
199
216
  return
200
- await self._book(usage, spec, entry[0], entry[1], mission_id)
217
+ await self._book(usage, spec, entry[0], entry[1], mission_id, deployment)
201
218
 
202
219
  @staticmethod
203
220
  def _as_dict(details: Any) -> dict[str, Any]:
@@ -218,7 +235,14 @@ class CostTrackingLogger(CustomLogger):
218
235
  return details.model_dump() if isinstance(details, BaseModel) else {}
219
236
 
220
237
  @staticmethod
221
- async def _book(usage: Any, spec: ModelSpec, cost: CostStrategy, multiplier: Decimal, mission_id: str) -> None:
238
+ async def _book(
239
+ usage: Any,
240
+ spec: ModelSpec,
241
+ cost: CostStrategy,
242
+ multiplier: Decimal,
243
+ mission_id: str,
244
+ deployment: str | None = None,
245
+ ) -> None:
222
246
  """Split usage into cost categories and add each to the cost service.
223
247
 
224
248
  ``prompt_tokens`` includes cache reads (not cache writes) and ``completion_tokens`` includes
@@ -230,6 +254,7 @@ class CostTrackingLogger(CustomLogger):
230
254
  cost: The mission's cost service.
231
255
  multiplier: Price multiplier.
232
256
  mission_id: Mission id, for logging.
257
+ deployment: Proxy deployment that answered (``x-litellm-model-id``), for logging.
233
258
  """
234
259
  prompt_details = CostTrackingLogger._as_dict(usage.prompt_tokens_details)
235
260
  completion_details = CostTrackingLogger._as_dict(usage.completion_tokens_details)
@@ -268,8 +293,10 @@ class CostTrackingLogger(CustomLogger):
268
293
  credit(cache_creation, costs.cache_creation_input_token_cost or costs.input_cost_per_token, multiplier),
269
294
  )
270
295
  logger.info(
271
- "LLM cost tracked: model=%s input=%d output=%d reasoning=%d cache_read=%d cache_write=%d multiplier=%s",
296
+ "LLM cost tracked: model=%s deployment=%s input=%d output=%d reasoning=%d cache_read=%d cache_write=%d "
297
+ "multiplier=%s",
272
298
  spec.name,
299
+ deployment,
273
300
  non_cached_input,
274
301
  non_reasoning_output,
275
302
  reasoning,
@@ -8,6 +8,7 @@ from decimal import ROUND_HALF_EVEN, Decimal, InvalidOperation
8
8
  from typing import Any, ClassVar
9
9
 
10
10
  import httpx
11
+ from litellm.types.utils import ModelResponseBase
11
12
  from pydantic import BaseModel, Field
12
13
 
13
14
  from digitalkin.community.litellm.model_specs import (
@@ -327,6 +328,10 @@ class LiteLLMRegistry:
327
328
  self._config = config or LiteLLMConfig()
328
329
  self._client = LiteLLMClient(self._config)
329
330
  self._models: dict[str, ModelSpec] = {}
331
+ # Deployment id (``model_info.id``) -> model group, for every row of every ``/model/info``
332
+ # read: a streamed response names its deployment only (``x-litellm-model-id``), the group
333
+ # header being absent from proxy streams before litellm 1.88.
334
+ self._deployment_groups: dict[str, str] = {}
330
335
  self._lock = asyncio.Lock()
331
336
  self._initialized = False
332
337
 
@@ -374,12 +379,18 @@ class LiteLLMRegistry:
374
379
  return
375
380
  if not model_name and self._initialized:
376
381
  return
382
+ rows = await self._client.fetch_raw_models()
383
+ self._remember_deployments(rows)
377
384
  if model_name:
378
385
  logger.info("Loading model '%s'", model_name)
379
- self._parse_model(await self._client.fetch_model(model_name), strict=True)
386
+ row = next((r for r in rows if r.get("model_name") == model_name), None)
387
+ if row is None:
388
+ msg = f"Model '{model_name}' not found in LiteLLM"
389
+ raise ValueError(msg)
390
+ self._parse_model(row, strict=True)
380
391
  else:
381
392
  logger.info("Loading all models")
382
- for model_data in await self._client.fetch_raw_models():
393
+ for model_data in rows:
383
394
  self._parse_model(model_data, strict=False)
384
395
  self._initialized = True
385
396
  logger.info("LiteLLM registry now holds %d models", len(self._models))
@@ -398,7 +409,9 @@ class LiteLLMRegistry:
398
409
  if not wanted:
399
410
  return
400
411
  logger.info("Loading %d models from LiteLLM: %s", len(wanted), sorted(wanted))
401
- for model_data in await self._client.fetch_raw_models():
412
+ rows = await self._client.fetch_raw_models()
413
+ self._remember_deployments(rows)
414
+ for model_data in rows:
402
415
  row_name = model_data.get("model_name")
403
416
  if isinstance(row_name, str) and row_name in wanted:
404
417
  self._parse_model(model_data, strict=True)
@@ -447,6 +460,53 @@ class LiteLLMRegistry:
447
460
  raise ValueError(msg)
448
461
  return model
449
462
 
463
+ def _remember_deployments(self, rows: list[dict[str, Any]]) -> None:
464
+ """Record the model group of every deployment of a ``/model/info`` read.
465
+
466
+ Args:
467
+ rows: Raw ``/model/info`` entries.
468
+ """
469
+ for row in rows:
470
+ info = row.get("model_info")
471
+ deployment = info.get("id") if isinstance(info, dict) else None
472
+ group = row.get("model_name")
473
+ if isinstance(deployment, str) and deployment and isinstance(group, str) and group:
474
+ self._deployment_groups[deployment] = group
475
+
476
+ def deployment_group(self, deployment_id: str) -> str | None:
477
+ """Model group a proxy deployment belongs to.
478
+
479
+ Args:
480
+ deployment_id: The ``model_info.id`` the proxy returns as ``x-litellm-model-id``.
481
+
482
+ Returns:
483
+ The group name, or None for a deployment no ``/model/info`` read has listed.
484
+ """
485
+ return self._deployment_groups.get(deployment_id)
486
+
487
+ def served_route(self, response: Any) -> tuple[str | None, str | None]:
488
+ """The proxy model group and deployment that answered a call, from its response headers.
489
+
490
+ litellm keeps the proxy headers, ``llm_provider-`` prefixed, in ``_hidden_params``. A stream
491
+ from a proxy before litellm 1.88 names only its deployment; the group is then resolved here.
492
+
493
+ Args:
494
+ response: A litellm completion or streaming chunk.
495
+
496
+ Returns:
497
+ ``(group, deployment_id)``; None for what the response does not carry.
498
+ """
499
+ # ``_hidden_params``, not the ``hidden_params`` property: litellm before 1.10x has no property.
500
+ hidden = response._hidden_params if isinstance(response, ModelResponseBase) else None # ruff: ignore[private-member-access]
501
+ headers = hidden.get("additional_headers") if isinstance(hidden, dict) else None
502
+ if not isinstance(headers, dict):
503
+ return None, None
504
+ deployment = headers.get("llm_provider-x-litellm-model-id") or None
505
+ group = headers.get("llm_provider-x-litellm-model-group") or (
506
+ self._deployment_groups.get(deployment) if deployment else None
507
+ )
508
+ return group, deployment
509
+
450
510
  def try_get(self, name: str) -> ModelSpec | None:
451
511
  """Return a loaded model, or None.
452
512
 
@@ -1,6 +1,7 @@
1
1
  """LiteLLM proxy settings."""
2
2
 
3
3
  from functools import lru_cache
4
+ from typing import Literal
4
5
 
5
6
  from pydantic import Field, SecretStr
6
7
  from pydantic_settings import BaseSettings, SettingsConfigDict
@@ -23,6 +24,10 @@ class LiteLLMSettings(BaseSettings):
23
24
  description="Bearer token for the LiteLLM proxy.",
24
25
  json_schema_extra={"env_example": "sk-xxxx", "env_required": True},
25
26
  )
27
+ prompt_cache_ttl: Literal["", "5m", "1h"] = Field(
28
+ default="",
29
+ description="Anthropic prompt-cache TTL on the cache markers (1h); empty keeps the provider's 5 minutes.",
30
+ )
26
31
 
27
32
 
28
33
  @lru_cache(maxsize=1)
@@ -114,11 +114,11 @@ class M2MCallRegistry:
114
114
  """``host:port`` the local gateway advertises as its dial-back target.
115
115
 
116
116
  Returns:
117
- ``host:port`` string from channel settings (``advertise_host`` falls back to ``host``).
117
+ ``host:port`` string from channel settings (``advertise_host`` / ``advertise_port`` fall back to
118
+ ``host`` / ``port``).
118
119
  """
119
- ch = get_server_channel_settings()
120
- host = ch.advertise_host or ch.host
121
- return f"{host}:{ch.port}"
120
+ host, port = get_server_channel_settings().advertise_address
121
+ return f"{host}:{port}"
122
122
 
123
123
  async def start(self) -> None:
124
124
  """Spawn the TTL sweeper task."""