synth-ai 0.2.9.dev4__py3-none-any.whl → 0.2.9.dev6__py3-none-any.whl

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.

Potentially problematic release.


This version of synth-ai might be problematic. Click here for more details.

Files changed (353) hide show
  1. examples/__init__.py +16 -0
  2. examples/crafter_debug_render.py +23 -17
  3. examples/qwen_coder/README.md +102 -0
  4. examples/qwen_coder/_shared.py +113 -0
  5. examples/qwen_coder/configs/coder_lora_30b.toml +61 -0
  6. examples/qwen_coder/configs/coder_lora_4b.toml +57 -0
  7. examples/qwen_coder/configs/coder_lora_small.toml +58 -0
  8. examples/qwen_coder/generate_dataset.py +98 -0
  9. examples/qwen_coder/infer_ft_smoke.py +64 -0
  10. examples/qwen_coder/infer_prod_proxy.py +73 -0
  11. examples/qwen_coder/infer_via_synth.py +87 -0
  12. examples/qwen_coder/scripts/infer_coder.sh +18 -0
  13. examples/qwen_coder/scripts/train_coder_30b.sh +21 -0
  14. examples/qwen_coder/sft_full_17b.py +103 -0
  15. examples/qwen_coder/sft_lora_30b.py +110 -0
  16. examples/qwen_coder/subset_jsonl.py +38 -0
  17. examples/qwen_coder/validate_jsonl.py +59 -0
  18. examples/rl/configs/eval_base_qwen.toml +1 -1
  19. examples/rl/configs/rl_from_base_qwen17.toml +1 -1
  20. examples/rl/download_dataset.py +26 -10
  21. examples/rl/run_eval.py +53 -52
  22. examples/rl/run_rl_and_save.py +29 -12
  23. examples/rl/task_app/math_single_step.py +180 -41
  24. examples/rl/task_app/math_task_app.py +14 -6
  25. examples/sft/README.md +139 -0
  26. examples/sft/configs/crafter_fft_qwen0p6b.toml +44 -0
  27. examples/sft/configs/crafter_lora_qwen0p6b.toml +45 -0
  28. examples/sft/evaluate.py +117 -0
  29. examples/sft/export_dataset.py +117 -0
  30. examples/sft/generate_traces.py +162 -0
  31. examples/swe/__init__.py +12 -0
  32. examples/swe/task_app/README.md +105 -0
  33. examples/swe/task_app/__init__.py +2 -0
  34. examples/swe/task_app/grpo_swe_mini.py +571 -0
  35. examples/swe/task_app/grpo_swe_mini_task_app.py +136 -0
  36. examples/swe/task_app/hosted/README.md +173 -0
  37. examples/swe/task_app/hosted/__init__.py +5 -0
  38. examples/swe/task_app/hosted/branching.py +143 -0
  39. examples/swe/task_app/hosted/environment_routes.py +1289 -0
  40. examples/swe/task_app/hosted/envs/__init__.py +1 -0
  41. examples/swe/task_app/hosted/envs/crafter/__init__.py +6 -0
  42. examples/swe/task_app/hosted/envs/crafter/app.py +1 -0
  43. examples/swe/task_app/hosted/envs/crafter/environment.py +522 -0
  44. examples/swe/task_app/hosted/envs/crafter/policy.py +478 -0
  45. examples/swe/task_app/hosted/envs/crafter/react_agent.py +108 -0
  46. examples/swe/task_app/hosted/envs/crafter/shared.py +305 -0
  47. examples/swe/task_app/hosted/envs/crafter/tools.py +47 -0
  48. examples/swe/task_app/hosted/envs/mini_swe/__init__.py +8 -0
  49. examples/swe/task_app/hosted/envs/mini_swe/environment.py +1164 -0
  50. examples/swe/task_app/hosted/envs/mini_swe/policy.py +355 -0
  51. examples/swe/task_app/hosted/envs/mini_swe/shared.py +83 -0
  52. examples/swe/task_app/hosted/envs/mini_swe/tools.py +96 -0
  53. examples/swe/task_app/hosted/hosted_app.py +204 -0
  54. examples/swe/task_app/hosted/inference/__init__.py +5 -0
  55. examples/swe/task_app/hosted/inference/openai_client.py +618 -0
  56. examples/swe/task_app/hosted/main.py +100 -0
  57. examples/swe/task_app/hosted/policy_routes.py +1079 -0
  58. examples/swe/task_app/hosted/registry.py +195 -0
  59. examples/swe/task_app/hosted/rollout.py +1869 -0
  60. examples/swe/task_app/hosted/storage/__init__.py +5 -0
  61. examples/swe/task_app/hosted/storage/volume.py +211 -0
  62. examples/swe/task_app/hosted/test_agents.py +161 -0
  63. examples/swe/task_app/hosted/test_service.py +137 -0
  64. examples/swe/task_app/hosted/utils.py +62 -0
  65. examples/vlm/README.md +68 -0
  66. examples/vlm/configs/crafter_vlm_gpt4o.toml +44 -0
  67. examples/vlm/crafter_image_only_agent.py +207 -0
  68. examples/vlm/crafter_openai_vlm_agent.py +277 -0
  69. examples/vlm/filter_image_rows.py +63 -0
  70. examples/vlm/run_crafter_vlm_benchmark.py +316 -0
  71. examples/warming_up_to_rl/analyze_trace_db.py +12 -10
  72. examples/warming_up_to_rl/configs/rl_from_base_qwen4b.toml +11 -1
  73. examples/warming_up_to_rl/export_trace_sft.py +218 -36
  74. examples/warming_up_to_rl/groq_test.py +15 -8
  75. examples/warming_up_to_rl/manage_secrets.py +29 -25
  76. examples/warming_up_to_rl/readme.md +9 -2
  77. examples/warming_up_to_rl/run_eval.py +137 -61
  78. examples/warming_up_to_rl/run_fft_and_save.py +131 -60
  79. examples/warming_up_to_rl/run_local_rollout.py +88 -39
  80. examples/warming_up_to_rl/run_local_rollout_modal.py +114 -28
  81. examples/warming_up_to_rl/run_local_rollout_parallel.py +81 -20
  82. examples/warming_up_to_rl/run_local_rollout_traced.py +126 -23
  83. examples/warming_up_to_rl/run_rl_and_save.py +35 -12
  84. examples/warming_up_to_rl/run_rollout_remote.py +44 -19
  85. examples/warming_up_to_rl/task_app/README.md +6 -2
  86. examples/warming_up_to_rl/task_app/grpo_crafter.py +319 -57
  87. examples/warming_up_to_rl/task_app/grpo_crafter_task_app.py +11 -30
  88. examples/warming_up_to_rl/task_app/synth_envs_hosted/__init__.py +1 -1
  89. examples/warming_up_to_rl/task_app/synth_envs_hosted/branching.py +9 -11
  90. examples/warming_up_to_rl/task_app/synth_envs_hosted/environment_routes.py +137 -182
  91. examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/__init__.py +1 -1
  92. examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/crafter/__init__.py +1 -1
  93. examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/crafter/app.py +1 -1
  94. examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/crafter/environment.py +150 -57
  95. examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/crafter/policy.py +105 -69
  96. examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/crafter/react_agent.py +19 -7
  97. examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/crafter/shared.py +45 -42
  98. examples/warming_up_to_rl/task_app/synth_envs_hosted/envs/crafter/tools.py +1 -1
  99. examples/warming_up_to_rl/task_app/synth_envs_hosted/hosted_app.py +47 -45
  100. examples/warming_up_to_rl/task_app/synth_envs_hosted/inference/__init__.py +1 -1
  101. examples/warming_up_to_rl/task_app/synth_envs_hosted/inference/openai_client.py +198 -92
  102. examples/warming_up_to_rl/task_app/synth_envs_hosted/main.py +0 -2
  103. examples/warming_up_to_rl/task_app/synth_envs_hosted/policy_routes.py +361 -263
  104. examples/warming_up_to_rl/task_app/synth_envs_hosted/registry.py +21 -23
  105. examples/warming_up_to_rl/task_app/synth_envs_hosted/rollout.py +394 -274
  106. examples/warming_up_to_rl/task_app/synth_envs_hosted/storage/__init__.py +1 -1
  107. examples/warming_up_to_rl/task_app/synth_envs_hosted/storage/volume.py +56 -62
  108. examples/warming_up_to_rl/task_app/synth_envs_hosted/test_agents.py +1 -0
  109. examples/warming_up_to_rl/task_app/synth_envs_hosted/test_service.py +6 -15
  110. examples/warming_up_to_rl/task_app/synth_envs_hosted/utils.py +4 -3
  111. synth/__init__.py +14 -0
  112. synth_ai/__init__.py +20 -4
  113. synth_ai/api/models/supported.py +376 -0
  114. synth_ai/api/train/builders.py +157 -26
  115. synth_ai/api/train/cli.py +213 -57
  116. synth_ai/api/train/config_finder.py +65 -5
  117. synth_ai/api/train/env_resolver.py +33 -15
  118. synth_ai/api/train/pollers.py +13 -4
  119. synth_ai/api/train/supported_algos.py +139 -0
  120. synth_ai/api/train/task_app.py +5 -3
  121. synth_ai/api/train/utils.py +33 -48
  122. synth_ai/cli/__init__.py +19 -4
  123. synth_ai/cli/_modal_wrapper.py +28 -0
  124. synth_ai/cli/_typer_patch.py +49 -0
  125. synth_ai/cli/balance.py +2 -3
  126. synth_ai/cli/calc.py +1 -1
  127. synth_ai/cli/demo.py +21 -6
  128. synth_ai/cli/recent.py +2 -2
  129. synth_ai/cli/rl_demo.py +77 -17
  130. synth_ai/cli/root.py +116 -39
  131. synth_ai/cli/status.py +2 -2
  132. synth_ai/cli/task_apps.py +1709 -243
  133. synth_ai/cli/traces.py +7 -4
  134. synth_ai/cli/turso.py +73 -0
  135. synth_ai/cli/watch.py +12 -18
  136. synth_ai/core/experiment.py +0 -2
  137. synth_ai/demo_registry.py +68 -31
  138. synth_ai/demos/core/cli.py +516 -194
  139. synth_ai/demos/demo_task_apps/__init__.py +3 -3
  140. synth_ai/demos/demo_task_apps/core.py +64 -28
  141. synth_ai/demos/demo_task_apps/crafter/configs/crafter_fft_4b.toml +2 -3
  142. synth_ai/demos/demo_task_apps/crafter/grpo_crafter_task_app.py +37 -30
  143. synth_ai/demos/demo_task_apps/math/_common.py +1 -2
  144. synth_ai/demos/demo_task_apps/math/app.py +2 -1
  145. synth_ai/demos/demo_task_apps/math/deploy_modal.py +3 -6
  146. synth_ai/demos/demo_task_apps/math/modal_task_app.py +183 -82
  147. synth_ai/demos/demo_task_apps/math/task_app_entry.py +0 -2
  148. synth_ai/environments/examples/bandit/engine.py +12 -4
  149. synth_ai/environments/examples/bandit/taskset.py +4 -4
  150. synth_ai/environments/examples/crafter_classic/environment.py +76 -1
  151. synth_ai/environments/reproducibility/tree.py +5 -6
  152. synth_ai/environments/service/app.py +11 -12
  153. synth_ai/environments/service/core_routes.py +10 -9
  154. synth_ai/environments/stateful/engine.py +1 -1
  155. synth_ai/environments/tasks/core.py +1 -0
  156. synth_ai/environments/tasks/filters.py +5 -6
  157. synth_ai/environments/tasks/utils.py +4 -5
  158. synth_ai/evals/base.py +0 -2
  159. synth_ai/handshake.py +11 -9
  160. synth_ai/http.py +1 -1
  161. synth_ai/http_client.py +43 -11
  162. synth_ai/inference/__init__.py +0 -2
  163. synth_ai/inference/client.py +20 -6
  164. synth_ai/jobs/client.py +103 -78
  165. synth_ai/learning/__init__.py +41 -6
  166. synth_ai/learning/algorithms.py +14 -0
  167. synth_ai/learning/client.py +121 -29
  168. synth_ai/learning/config.py +2 -40
  169. synth_ai/learning/constants.py +0 -2
  170. synth_ai/learning/ft_client.py +4 -56
  171. synth_ai/learning/health.py +13 -7
  172. synth_ai/learning/jobs.py +43 -47
  173. synth_ai/{rl → learning/rl}/__init__.py +14 -5
  174. synth_ai/learning/rl/client.py +267 -0
  175. synth_ai/learning/rl/config.py +31 -0
  176. synth_ai/{rl → learning/rl}/contracts.py +5 -10
  177. synth_ai/{rl → learning/rl}/env_keys.py +45 -16
  178. synth_ai/learning/rl/secrets.py +13 -0
  179. synth_ai/learning/rl_client.py +2 -253
  180. synth_ai/learning/sft/__init__.py +29 -0
  181. synth_ai/learning/sft/client.py +68 -0
  182. synth_ai/learning/sft/config.py +270 -0
  183. synth_ai/learning/sft/data.py +295 -0
  184. synth_ai/learning/sse.py +25 -26
  185. synth_ai/learning/validators.py +25 -24
  186. synth_ai/lm/__init__.py +21 -47
  187. synth_ai/task/__init__.py +26 -27
  188. synth_ai/task/apps/__init__.py +18 -19
  189. synth_ai/task/auth.py +35 -23
  190. synth_ai/task/client.py +15 -13
  191. synth_ai/task/contracts.py +37 -35
  192. synth_ai/task/datasets.py +9 -6
  193. synth_ai/task/errors.py +11 -10
  194. synth_ai/task/health.py +17 -11
  195. synth_ai/task/json.py +58 -24
  196. synth_ai/task/proxy.py +15 -14
  197. synth_ai/task/rubrics.py +22 -15
  198. synth_ai/task/server.py +43 -17
  199. synth_ai/task/tracing_utils.py +12 -7
  200. synth_ai/task/validators.py +0 -1
  201. synth_ai/task/vendors.py +5 -7
  202. synth_ai/tracing_v3/__init__.py +2 -0
  203. synth_ai/tracing_v3/abstractions.py +21 -4
  204. synth_ai/tracing_v3/db_config.py +26 -1
  205. synth_ai/tracing_v3/decorators.py +18 -15
  206. synth_ai/tracing_v3/examples/basic_usage.py +3 -2
  207. synth_ai/tracing_v3/hooks.py +6 -4
  208. synth_ai/tracing_v3/llm_call_record_helpers.py +6 -6
  209. synth_ai/tracing_v3/replica_sync.py +1 -0
  210. synth_ai/tracing_v3/session_tracer.py +63 -16
  211. synth_ai/tracing_v3/storage/base.py +89 -1
  212. synth_ai/tracing_v3/storage/config.py +21 -8
  213. synth_ai/tracing_v3/storage/factory.py +10 -8
  214. synth_ai/tracing_v3/storage/utils.py +4 -2
  215. synth_ai/tracing_v3/turso/daemon.py +7 -2
  216. synth_ai/tracing_v3/turso/models.py +5 -2
  217. synth_ai/tracing_v3/turso/native_manager.py +1173 -0
  218. synth_ai/tracing_v3/utils.py +4 -3
  219. synth_ai/v0/api/__init__.py +8 -0
  220. synth_ai/v0/api/models/__init__.py +8 -0
  221. synth_ai/v0/api/models/supported.py +8 -0
  222. synth_ai/v0/config/__init__.py +15 -0
  223. synth_ai/v0/config/base_url.py +12 -0
  224. synth_ai/v0/lm/__init__.py +51 -0
  225. synth_ai/{lm → v0/lm}/caching/ephemeral.py +3 -5
  226. synth_ai/{lm → v0/lm}/caching/handler.py +4 -4
  227. synth_ai/{lm → v0/lm}/caching/initialize.py +1 -1
  228. synth_ai/{lm → v0/lm}/caching/persistent.py +1 -1
  229. synth_ai/{lm → v0/lm}/config.py +6 -1
  230. synth_ai/{lm → v0/lm}/core/all.py +9 -9
  231. synth_ai/{lm → v0/lm}/core/exceptions.py +0 -2
  232. synth_ai/{lm → v0/lm}/core/main.py +19 -7
  233. synth_ai/{lm → v0/lm}/core/main_v3.py +10 -10
  234. synth_ai/{lm → v0/lm}/core/synth_models.py +2 -15
  235. synth_ai/{lm → v0/lm}/core/vendor_clients.py +6 -4
  236. synth_ai/{lm → v0/lm}/overrides.py +4 -4
  237. synth_ai/{lm → v0/lm}/provider_support/anthropic.py +4 -4
  238. synth_ai/{lm → v0/lm}/provider_support/openai.py +5 -5
  239. synth_ai/{lm → v0/lm}/structured_outputs/handler.py +5 -5
  240. synth_ai/{lm → v0/lm}/structured_outputs/rehabilitate.py +1 -1
  241. synth_ai/{lm → v0/lm}/vendors/core/anthropic_api.py +16 -16
  242. synth_ai/{lm → v0/lm}/vendors/core/gemini_api.py +5 -5
  243. synth_ai/{lm → v0/lm}/vendors/core/mistral_api.py +5 -5
  244. synth_ai/{lm → v0/lm}/vendors/core/openai_api.py +12 -10
  245. synth_ai/{lm → v0/lm}/vendors/openai_standard.py +11 -9
  246. synth_ai/{lm → v0/lm}/vendors/openai_standard_responses.py +8 -5
  247. synth_ai/{lm → v0/lm}/vendors/supported/custom_endpoint.py +4 -6
  248. synth_ai/{lm → v0/lm}/vendors/supported/deepseek.py +2 -2
  249. synth_ai/{lm → v0/lm}/vendors/supported/grok.py +2 -2
  250. synth_ai/{lm → v0/lm}/vendors/supported/groq.py +1 -1
  251. synth_ai/{lm → v0/lm}/vendors/supported/ollama.py +1 -1
  252. synth_ai/{lm → v0/lm}/vendors/supported/openrouter.py +3 -3
  253. synth_ai/{lm → v0/lm}/vendors/supported/together.py +1 -1
  254. synth_ai/{lm → v0/lm}/vendors/synth_client.py +38 -11
  255. synth_ai/v0/tracing/upload.py +32 -135
  256. synth_ai/v0/tracing_v3/__init__.py +10 -0
  257. synth_ai/v0/tracing_v3/abstractions.py +3 -0
  258. synth_ai/v0/tracing_v3/decorators.py +3 -0
  259. synth_ai/v0/tracing_v3/llm_call_record_helpers.py +3 -0
  260. synth_ai/v0/tracing_v3/session_tracer.py +3 -0
  261. synth_ai-0.2.9.dev6.dist-info/METADATA +191 -0
  262. {synth_ai-0.2.9.dev4.dist-info → synth_ai-0.2.9.dev6.dist-info}/RECORD +291 -264
  263. {synth_ai-0.2.9.dev4.dist-info → synth_ai-0.2.9.dev6.dist-info}/top_level.txt +1 -0
  264. examples/common_old/backend.py +0 -21
  265. examples/evals_old/README.md +0 -98
  266. examples/evals_old/__init__.py +0 -6
  267. examples/evals_old/compare_models.py +0 -1037
  268. examples/evals_old/example_log.md +0 -145
  269. examples/evals_old/run_demo.sh +0 -126
  270. examples/evals_old/trace_analysis.py +0 -270
  271. examples/finetuning_old/_backup_synth_qwen/config.toml +0 -29
  272. examples/finetuning_old/_backup_synth_qwen/example_log.md +0 -324
  273. examples/finetuning_old/_backup_synth_qwen/filter_traces.py +0 -60
  274. examples/finetuning_old/_backup_synth_qwen/filter_traces_achievements.py +0 -239
  275. examples/finetuning_old/_backup_synth_qwen/purge_v3_traces.py +0 -109
  276. examples/finetuning_old/_backup_synth_qwen/react_agent_lm.py +0 -1924
  277. examples/finetuning_old/_backup_synth_qwen/readme.md +0 -49
  278. examples/finetuning_old/_backup_synth_qwen/run_crafter_qwen4b.py +0 -114
  279. examples/finetuning_old/_backup_synth_qwen/run_demo.sh +0 -195
  280. examples/finetuning_old/_backup_synth_qwen/sft_kickoff.py +0 -118
  281. examples/finetuning_old/synth_qwen_v1/README.md +0 -68
  282. examples/finetuning_old/synth_qwen_v1/filter_traces.py +0 -60
  283. examples/finetuning_old/synth_qwen_v1/filter_traces_achievements.py +0 -239
  284. examples/finetuning_old/synth_qwen_v1/finetune.py +0 -46
  285. examples/finetuning_old/synth_qwen_v1/hello_ft_model.py +0 -71
  286. examples/finetuning_old/synth_qwen_v1/infer.py +0 -37
  287. examples/finetuning_old/synth_qwen_v1/poll.py +0 -44
  288. examples/finetuning_old/synth_qwen_v1/prepare_data.py +0 -35
  289. examples/finetuning_old/synth_qwen_v1/purge_v3_traces.py +0 -109
  290. examples/finetuning_old/synth_qwen_v1/react_agent_lm.py +0 -1932
  291. examples/finetuning_old/synth_qwen_v1/run_crafter_sft_job.py +0 -207
  292. examples/finetuning_old/synth_qwen_v1/run_ft_job.py +0 -232
  293. examples/finetuning_old/synth_qwen_v1/upload_data.py +0 -34
  294. examples/finetuning_old/synth_qwen_v1/util.py +0 -147
  295. examples/rl_old/task_app.py +0 -962
  296. examples/warming_up_to_rl/old/event_rewards.md +0 -234
  297. examples/warming_up_to_rl/old/notes.md +0 -73
  298. examples/warming_up_to_rl/task_app/synth_envs_hosted/test_stepwise_rewards.py +0 -58
  299. synth_ai/environments/examples/crafter_classic/agent_demos/crafter_modal_ft/filter_traces_sft_turso.py +0 -738
  300. synth_ai/environments/examples/crafter_classic/agent_demos/crafter_openai_ft/filter_traces_sft_turso.py +0 -580
  301. synth_ai/environments/examples/sokoban/units/astar_common.py +0 -95
  302. synth_ai/experimental/synth_oss.py +0 -446
  303. synth_ai/install_sqld.sh +0 -40
  304. synth_ai/learning/filtering.py +0 -0
  305. synth_ai/learning/offline/dpo.py +0 -0
  306. synth_ai/learning/offline/providers.py +0 -7
  307. synth_ai/learning/offline/sft.py +0 -0
  308. synth_ai/learning/offline/shared.py +0 -0
  309. synth_ai/learning/online/grpo.py +0 -0
  310. synth_ai/learning/online/irft.py +0 -0
  311. synth_ai/learning/prompts/banking77_injection_eval.py +0 -168
  312. synth_ai/learning/prompts/gepa.py +0 -0
  313. synth_ai/learning/prompts/hello_world_in_context_injection_ex.py +0 -213
  314. synth_ai/learning/prompts/mipro.py +0 -289
  315. synth_ai/learning/prompts/random_search.py +0 -246
  316. synth_ai/learning/prompts/run_mipro_banking77.py +0 -172
  317. synth_ai/learning/prompts/run_random_search_banking77.py +0 -324
  318. synth_ai/rl/secrets.py +0 -19
  319. synth_ai/scripts/verify_rewards.py +0 -100
  320. synth_ai/tracing/__init__.py +0 -30
  321. synth_ai/tracing_v1/__init__.py +0 -33
  322. synth_ai/tracing_v3/turso/__init__.py +0 -25
  323. synth_ai/tracing_v3/turso/manager.py +0 -774
  324. synth_ai/zyk/__init__.py +0 -30
  325. synth_ai-0.2.9.dev4.dist-info/METADATA +0 -131
  326. /synth_ai/{lm → v0/lm}/caching/__init__.py +0 -0
  327. /synth_ai/{lm → v0/lm}/caching/constants.py +0 -0
  328. /synth_ai/{lm → v0/lm}/caching/dbs.py +0 -0
  329. /synth_ai/{lm → v0/lm}/constants.py +0 -0
  330. /synth_ai/{lm → v0/lm}/core/__init__.py +0 -0
  331. /synth_ai/{lm → v0/lm}/cost/__init__.py +0 -0
  332. /synth_ai/{lm → v0/lm}/cost/monitor.py +0 -0
  333. /synth_ai/{lm → v0/lm}/cost/statefulness.py +0 -0
  334. /synth_ai/{lm → v0/lm}/injection.py +0 -0
  335. /synth_ai/{lm → v0/lm}/provider_support/__init__.py +0 -0
  336. /synth_ai/{lm → v0/lm}/provider_support/suppress_logging.py +0 -0
  337. /synth_ai/{lm → v0/lm}/structured_outputs/__init__.py +0 -0
  338. /synth_ai/{lm → v0/lm}/structured_outputs/inject.py +0 -0
  339. /synth_ai/{lm → v0/lm}/tools/__init__.py +0 -0
  340. /synth_ai/{lm → v0/lm}/tools/base.py +0 -0
  341. /synth_ai/{lm → v0/lm}/unified_interface.py +0 -0
  342. /synth_ai/{lm → v0/lm}/vendors/__init__.py +0 -0
  343. /synth_ai/{lm → v0/lm}/vendors/base.py +0 -0
  344. /synth_ai/{lm → v0/lm}/vendors/core/__init__.py +0 -0
  345. /synth_ai/{lm → v0/lm}/vendors/core/synth_dev_api.py +0 -0
  346. /synth_ai/{lm → v0/lm}/vendors/local/__init__.py +0 -0
  347. /synth_ai/{lm → v0/lm}/vendors/local/ollama.py +0 -0
  348. /synth_ai/{lm → v0/lm}/vendors/retries.py +0 -0
  349. /synth_ai/{lm → v0/lm}/vendors/supported/__init__.py +0 -0
  350. /synth_ai/{lm → v0/lm}/warmup.py +0 -0
  351. {synth_ai-0.2.9.dev4.dist-info → synth_ai-0.2.9.dev6.dist-info}/WHEEL +0 -0
  352. {synth_ai-0.2.9.dev4.dist-info → synth_ai-0.2.9.dev6.dist-info}/entry_points.txt +0 -0
  353. {synth_ai-0.2.9.dev4.dist-info → synth_ai-0.2.9.dev6.dist-info}/licenses/LICENSE +0 -0
@@ -0,0 +1,1173 @@
1
+ """LibSQL-native trace manager prototype.
2
+
3
+ This module provides the Turso/libsql-backed trace storage implementation. It
4
+ mirrors the public surface area of the historical SQLAlchemy manager while
5
+ executing all operations directly via libsql.
6
+ """
7
+
8
+ from __future__ import annotations
9
+
10
+ import asyncio
11
+ import json
12
+ import logging
13
+ import re
14
+ from dataclasses import asdict, dataclass
15
+ from datetime import UTC, datetime
16
+ from typing import Any
17
+
18
+ import libsql
19
+ from sqlalchemy.engine import make_url
20
+
21
+ try: # pragma: no cover - exercised only when pandas present
22
+ import pandas as pd # type: ignore
23
+ except Exception: # pragma: no cover
24
+ pd = None # type: ignore[assignment]
25
+
26
+ from ..abstractions import (
27
+ EnvironmentEvent,
28
+ LMCAISEvent,
29
+ RuntimeEvent,
30
+ SessionMessageContent,
31
+ SessionTrace,
32
+ )
33
+ from ..config import CONFIG
34
+ from ..storage.base import TraceStorage
35
+ from .models import analytics_views
36
+
37
+ logger = logging.getLogger(__name__)
38
+
39
+
40
+ @dataclass(slots=True)
41
+ class _ConnectionTarget:
42
+ """Resolved connection target for libsql."""
43
+
44
+ database: str
45
+ sync_url: str | None = None
46
+ auth_token: str | None = None
47
+
48
+
49
+ def _resolve_connection_target(db_url: str | None, auth_token: str | None) -> _ConnectionTarget:
50
+ """Normalise the configured database URL."""
51
+ url = db_url or CONFIG.db_url
52
+
53
+ # Fast-path local SQLite URLs (`sqlite+aiosqlite:///path/to/db`)
54
+ if url.startswith("sqlite+aiosqlite:///"):
55
+ return _ConnectionTarget(database=url.replace("sqlite+aiosqlite:///", ""), auth_token=auth_token)
56
+
57
+ # SQLAlchemy-compatible libsql scheme (`sqlite+libsql://<endpoint or path>`)
58
+ if url.startswith("sqlite+libsql://"):
59
+ target = url.replace("sqlite+libsql://", "", 1)
60
+ return _ConnectionTarget(database=target, sync_url=target if target.startswith("libsql://") else None, auth_token=auth_token)
61
+
62
+ # Native libsql URLs (`libsql://...`).
63
+ if url.startswith("libsql://"):
64
+ return _ConnectionTarget(database=url, sync_url=url, auth_token=auth_token)
65
+
66
+ # Fallback to SQLAlchemy URL parsing for anything else we missed.
67
+ try:
68
+ parsed = make_url(url)
69
+ if parsed.drivername.startswith("sqlite"):
70
+ if parsed.database:
71
+ return _ConnectionTarget(database=parsed.database, auth_token=auth_token)
72
+ if parsed.drivername.startswith("libsql"):
73
+ database = parsed.render_as_string(hide_password=False)
74
+ return _ConnectionTarget(database=database, sync_url=database, auth_token=auth_token)
75
+ except Exception: # pragma: no cover - defensive guardrail
76
+ logger.debug("Unable to parse db_url via SQLAlchemy", exc_info=True)
77
+
78
+ # As a last resort use the raw value (libsql.connect can handle absolute paths).
79
+ return _ConnectionTarget(database=url, auth_token=auth_token)
80
+
81
+
82
+ def _json_dumps(value: Any) -> str | None:
83
+ """Serialise Python objects as JSON compatible with the existing schema."""
84
+
85
+ def _default(obj: Any):
86
+ if isinstance(obj, datetime):
87
+ return obj.isoformat()
88
+ return str(obj)
89
+
90
+ if value is None:
91
+ return None
92
+ return json.dumps(value, separators=(",", ":"), default=_default)
93
+
94
+
95
+ def _maybe_datetime(value: Any) -> Any:
96
+ if value is None or isinstance(value, datetime):
97
+ return value
98
+ if isinstance(value, str):
99
+ try:
100
+ return datetime.fromisoformat(value)
101
+ except ValueError:
102
+ pass
103
+ return value
104
+
105
+
106
+ def _load_json(value: Any) -> Any:
107
+ if value is None or isinstance(value, (dict, list)):
108
+ return value or {}
109
+ if isinstance(value, str):
110
+ try:
111
+ return json.loads(value)
112
+ except (TypeError, ValueError):
113
+ return {}
114
+ return value
115
+
116
+
117
+ _TABLE_DEFINITIONS: tuple[str, ...] = (
118
+ """
119
+ CREATE TABLE IF NOT EXISTS experiments (
120
+ experiment_id VARCHAR PRIMARY KEY,
121
+ name VARCHAR NOT NULL,
122
+ description TEXT,
123
+ created_at DATETIME DEFAULT CURRENT_TIMESTAMP,
124
+ updated_at DATETIME DEFAULT CURRENT_TIMESTAMP,
125
+ configuration TEXT,
126
+ metadata TEXT
127
+ )
128
+ """,
129
+ """
130
+ CREATE TABLE IF NOT EXISTS systems (
131
+ system_id VARCHAR PRIMARY KEY,
132
+ name VARCHAR NOT NULL,
133
+ system_type VARCHAR,
134
+ description TEXT,
135
+ created_at DATETIME DEFAULT CURRENT_TIMESTAMP,
136
+ metadata TEXT
137
+ )
138
+ """,
139
+ """
140
+ CREATE TABLE IF NOT EXISTS system_versions (
141
+ version_id VARCHAR PRIMARY KEY,
142
+ system_id VARCHAR NOT NULL,
143
+ version_number VARCHAR NOT NULL,
144
+ commit_hash VARCHAR,
145
+ created_at DATETIME DEFAULT CURRENT_TIMESTAMP,
146
+ configuration TEXT,
147
+ metadata TEXT,
148
+ FOREIGN KEY(system_id) REFERENCES systems(system_id),
149
+ UNIQUE(system_id, version_number)
150
+ )
151
+ """,
152
+ """
153
+ CREATE TABLE IF NOT EXISTS experimental_systems (
154
+ id INTEGER PRIMARY KEY AUTOINCREMENT,
155
+ experiment_id VARCHAR NOT NULL,
156
+ system_id VARCHAR NOT NULL,
157
+ version_id VARCHAR NOT NULL,
158
+ FOREIGN KEY(experiment_id) REFERENCES experiments(experiment_id),
159
+ FOREIGN KEY(system_id) REFERENCES systems(system_id),
160
+ FOREIGN KEY(version_id) REFERENCES system_versions(version_id)
161
+ )
162
+ """,
163
+ """
164
+ CREATE TABLE IF NOT EXISTS session_traces (
165
+ session_id VARCHAR PRIMARY KEY,
166
+ created_at DATETIME NOT NULL,
167
+ num_timesteps INTEGER NOT NULL,
168
+ num_events INTEGER NOT NULL,
169
+ num_messages INTEGER NOT NULL,
170
+ metadata TEXT,
171
+ experiment_id VARCHAR,
172
+ embedding VECTOR,
173
+ FOREIGN KEY(experiment_id) REFERENCES experiments(experiment_id)
174
+ )
175
+ """,
176
+ """
177
+ CREATE TABLE IF NOT EXISTS session_timesteps (
178
+ id INTEGER PRIMARY KEY AUTOINCREMENT,
179
+ session_id VARCHAR NOT NULL,
180
+ step_id VARCHAR NOT NULL,
181
+ step_index INTEGER NOT NULL,
182
+ turn_number INTEGER,
183
+ started_at DATETIME,
184
+ completed_at DATETIME,
185
+ num_events INTEGER,
186
+ num_messages INTEGER,
187
+ step_metadata TEXT,
188
+ UNIQUE(session_id, step_id),
189
+ FOREIGN KEY(session_id) REFERENCES session_traces(session_id)
190
+ )
191
+ """,
192
+ """
193
+ CREATE TABLE IF NOT EXISTS events (
194
+ id INTEGER PRIMARY KEY AUTOINCREMENT,
195
+ session_id VARCHAR NOT NULL,
196
+ timestep_id INTEGER,
197
+ event_type VARCHAR NOT NULL,
198
+ system_instance_id VARCHAR,
199
+ event_time FLOAT,
200
+ message_time INTEGER,
201
+ created_at DATETIME,
202
+ model_name VARCHAR,
203
+ provider VARCHAR,
204
+ input_tokens INTEGER,
205
+ output_tokens INTEGER,
206
+ total_tokens INTEGER,
207
+ cost_usd INTEGER,
208
+ latency_ms INTEGER,
209
+ span_id VARCHAR,
210
+ trace_id VARCHAR,
211
+ call_records TEXT,
212
+ reward FLOAT,
213
+ terminated BOOLEAN,
214
+ truncated BOOLEAN,
215
+ system_state_before TEXT,
216
+ system_state_after TEXT,
217
+ metadata TEXT,
218
+ event_metadata TEXT,
219
+ embedding VECTOR,
220
+ CHECK (event_type IN ('cais', 'environment', 'runtime')),
221
+ FOREIGN KEY(session_id) REFERENCES session_traces(session_id),
222
+ FOREIGN KEY(timestep_id) REFERENCES session_timesteps(id)
223
+ )
224
+ """,
225
+ """
226
+ CREATE TABLE IF NOT EXISTS messages (
227
+ id INTEGER PRIMARY KEY AUTOINCREMENT,
228
+ session_id VARCHAR NOT NULL,
229
+ timestep_id INTEGER,
230
+ message_type VARCHAR NOT NULL,
231
+ content TEXT NOT NULL,
232
+ timestamp DATETIME,
233
+ event_time FLOAT,
234
+ message_time INTEGER,
235
+ metadata TEXT,
236
+ embedding VECTOR,
237
+ CHECK (message_type IN ('user', 'assistant', 'system', 'tool_use', 'tool_result')),
238
+ FOREIGN KEY(session_id) REFERENCES session_traces(session_id),
239
+ FOREIGN KEY(timestep_id) REFERENCES session_timesteps(id)
240
+ )
241
+ """,
242
+ """
243
+ CREATE TABLE IF NOT EXISTS outcome_rewards (
244
+ id INTEGER PRIMARY KEY AUTOINCREMENT,
245
+ session_id VARCHAR NOT NULL,
246
+ total_reward INTEGER NOT NULL,
247
+ achievements_count INTEGER NOT NULL,
248
+ total_steps INTEGER NOT NULL,
249
+ created_at DATETIME NOT NULL,
250
+ reward_metadata TEXT,
251
+ FOREIGN KEY(session_id) REFERENCES session_traces(session_id)
252
+ )
253
+ """,
254
+ """
255
+ CREATE TABLE IF NOT EXISTS event_rewards (
256
+ id INTEGER PRIMARY KEY AUTOINCREMENT,
257
+ event_id INTEGER NOT NULL,
258
+ session_id VARCHAR NOT NULL,
259
+ message_id INTEGER,
260
+ turn_number INTEGER,
261
+ reward_value FLOAT NOT NULL,
262
+ reward_type VARCHAR,
263
+ "key" VARCHAR,
264
+ annotation TEXT,
265
+ source VARCHAR,
266
+ created_at DATETIME NOT NULL,
267
+ FOREIGN KEY(event_id) REFERENCES events(id),
268
+ FOREIGN KEY(session_id) REFERENCES session_traces(session_id),
269
+ FOREIGN KEY(message_id) REFERENCES messages(id)
270
+ )
271
+ """
272
+ )
273
+
274
+
275
+ _INDEX_DEFINITIONS: tuple[str, ...] = (
276
+ "CREATE INDEX IF NOT EXISTS idx_session_created ON session_traces (created_at)",
277
+ "CREATE INDEX IF NOT EXISTS idx_session_experiment ON session_traces (experiment_id)",
278
+ "CREATE INDEX IF NOT EXISTS idx_timestep_session_step ON session_timesteps (session_id, step_id)",
279
+ "CREATE INDEX IF NOT EXISTS idx_timestep_turn ON session_timesteps (turn_number)",
280
+ "CREATE INDEX IF NOT EXISTS idx_event_session_step ON events (session_id, timestep_id)",
281
+ "CREATE INDEX IF NOT EXISTS idx_event_type ON events (event_type)",
282
+ "CREATE INDEX IF NOT EXISTS idx_event_created ON events (created_at)",
283
+ "CREATE INDEX IF NOT EXISTS idx_event_model ON events (model_name)",
284
+ "CREATE INDEX IF NOT EXISTS idx_event_trace ON events (trace_id)",
285
+ "CREATE INDEX IF NOT EXISTS idx_message_session_step ON messages (session_id, timestep_id)",
286
+ "CREATE INDEX IF NOT EXISTS idx_message_type ON messages (message_type)",
287
+ "CREATE INDEX IF NOT EXISTS idx_message_timestamp ON messages (timestamp)",
288
+ "CREATE INDEX IF NOT EXISTS idx_experiment_created ON experiments (created_at)",
289
+ "CREATE INDEX IF NOT EXISTS idx_experiment_name ON experiments (name)",
290
+ "CREATE INDEX IF NOT EXISTS idx_system_name ON systems (name)",
291
+ "CREATE INDEX IF NOT EXISTS idx_system_type ON systems (system_type)",
292
+ "CREATE UNIQUE INDEX IF NOT EXISTS uq_system_version ON system_versions (system_id, version_number)",
293
+ "CREATE INDEX IF NOT EXISTS idx_version_system ON system_versions (system_id)",
294
+ "CREATE INDEX IF NOT EXISTS idx_version_created ON system_versions (created_at)",
295
+ "CREATE UNIQUE INDEX IF NOT EXISTS uq_experiment_system ON experimental_systems (experiment_id, system_id)",
296
+ "CREATE INDEX IF NOT EXISTS idx_experimental_system ON experimental_systems (experiment_id, system_id)",
297
+ "CREATE INDEX IF NOT EXISTS idx_outcome_rewards_session ON outcome_rewards (session_id)",
298
+ "CREATE INDEX IF NOT EXISTS idx_outcome_rewards_total ON outcome_rewards (total_reward)",
299
+ "CREATE INDEX IF NOT EXISTS idx_event_rewards_session ON event_rewards (session_id)",
300
+ "CREATE INDEX IF NOT EXISTS idx_event_rewards_event ON event_rewards (event_id)",
301
+ "CREATE INDEX IF NOT EXISTS idx_event_rewards_type ON event_rewards (reward_type)",
302
+ 'CREATE INDEX IF NOT EXISTS idx_event_rewards_key ON event_rewards ("key")',
303
+ )
304
+
305
+
306
+ class NativeLibsqlTraceManager(TraceStorage):
307
+ """Libsql-backed trace manager."""
308
+
309
+ def __init__(
310
+ self,
311
+ db_url: str | None = None,
312
+ *,
313
+ auth_token: str | None = None,
314
+ ):
315
+ self._config_auth_token = auth_token
316
+ self._target = _resolve_connection_target(db_url, auth_token)
317
+ self._conn: libsql.Connection | None = None
318
+ self._conn_lock = asyncio.Lock()
319
+ self._op_lock = asyncio.Lock()
320
+ self._initialized = False
321
+
322
+ def _open_connection(self) -> libsql.Connection:
323
+ """Open a libsql connection for the resolved target."""
324
+ kwargs: dict[str, Any] = {}
325
+ if self._target.sync_url and self._target.sync_url.startswith("libsql://"):
326
+ kwargs["sync_url"] = self._target.sync_url
327
+ if self._target.auth_token:
328
+ kwargs["auth_token"] = self._target.auth_token
329
+ # Disable automatic background sync; ReplicaSync drives this explicitly.
330
+ kwargs.setdefault("sync_interval", 0)
331
+ logger.debug("Opening libsql connection to %s", self._target.database)
332
+ return libsql.connect(self._target.database, **kwargs)
333
+
334
+ async def initialize(self):
335
+ """Initialise the backend."""
336
+ async with self._conn_lock:
337
+ if self._initialized:
338
+ return
339
+
340
+ # Establish a libsql connection for future native operations.
341
+ self._conn = self._open_connection()
342
+ self._ensure_schema()
343
+ self._initialized = True
344
+
345
+ async def close(self):
346
+ """Close the libsql connection."""
347
+ async with self._conn_lock:
348
+ if self._conn:
349
+ logger.debug("Closing libsql connection to %s", self._target.database)
350
+ self._conn.close()
351
+ self._conn = None
352
+ self._initialized = False
353
+
354
+ # ------------------------------------------------------------------
355
+ # Delegated operations (to be swapped with native libsql versions).
356
+ # ------------------------------------------------------------------
357
+
358
+ async def insert_session_trace(self, trace: SessionTrace) -> str:
359
+ await self.initialize()
360
+
361
+ if await self._session_exists(trace.session_id):
362
+ async with self._op_lock:
363
+ conn = self._conn
364
+ assert conn is not None
365
+ conn.execute(
366
+ "UPDATE session_traces SET metadata = ? WHERE session_id = ?",
367
+ (_json_dumps(trace.metadata or {}), trace.session_id),
368
+ )
369
+ conn.commit()
370
+ return trace.session_id
371
+
372
+ created_at = trace.created_at or datetime.now(UTC)
373
+
374
+ async with self._op_lock:
375
+ conn = self._conn
376
+ assert conn is not None
377
+ conn.execute(
378
+ """
379
+ INSERT INTO session_traces (
380
+ session_id,
381
+ created_at,
382
+ num_timesteps,
383
+ num_events,
384
+ num_messages,
385
+ metadata
386
+ )
387
+ VALUES (?, ?, 0, 0, 0, ?)
388
+ """,
389
+ (
390
+ trace.session_id,
391
+ created_at.isoformat(),
392
+ _json_dumps(trace.metadata or {}),
393
+ ),
394
+ )
395
+ conn.commit()
396
+
397
+ step_id_map: dict[str, int] = {}
398
+
399
+ for step in trace.session_time_steps:
400
+ step_db_id = await self.ensure_timestep(
401
+ trace.session_id,
402
+ step_id=step.step_id,
403
+ step_index=step.step_index,
404
+ turn_number=step.turn_number,
405
+ started_at=step.timestamp,
406
+ completed_at=step.completed_at,
407
+ metadata=step.step_metadata or {},
408
+ )
409
+ step_id_map[step.step_id] = step_db_id
410
+
411
+ for event in trace.event_history:
412
+ step_ref = None
413
+ metadata = event.metadata or {}
414
+ if isinstance(metadata, dict):
415
+ step_ref = metadata.get("step_id")
416
+ timestep_db_id = step_id_map.get(step_ref) if step_ref else None
417
+ await self.insert_event_row(
418
+ trace.session_id,
419
+ timestep_db_id=timestep_db_id,
420
+ event=event,
421
+ metadata_override=event.metadata or {},
422
+ )
423
+
424
+ for msg in trace.markov_blanket_message_history:
425
+ metadata = dict(getattr(msg, "metadata", {}) or {})
426
+ step_ref = metadata.get("step_id")
427
+ content_value = msg.content
428
+ if isinstance(msg.content, SessionMessageContent):
429
+ if msg.content.json_payload:
430
+ metadata.setdefault("json_payload", msg.content.json_payload)
431
+ content_value = msg.content.json_payload
432
+ else:
433
+ content_value = msg.content.as_text()
434
+ if msg.content.text:
435
+ metadata.setdefault("text", msg.content.text)
436
+ elif not isinstance(content_value, str):
437
+ try:
438
+ content_value = json.dumps(content_value, ensure_ascii=False)
439
+ except (TypeError, ValueError):
440
+ content_value = str(content_value)
441
+
442
+ await self.insert_message_row(
443
+ trace.session_id,
444
+ timestep_db_id=step_id_map.get(step_ref) if step_ref else None,
445
+ message_type=msg.message_type,
446
+ content=content_value,
447
+ event_time=msg.time_record.event_time,
448
+ message_time=msg.time_record.message_time,
449
+ metadata=metadata,
450
+ )
451
+
452
+ async with self._op_lock:
453
+ conn = self._conn
454
+ assert conn is not None
455
+ conn.execute(
456
+ "UPDATE session_traces SET num_timesteps = ?, num_events = ?, num_messages = ?, metadata = ? WHERE session_id = ?",
457
+ (
458
+ len(trace.session_time_steps),
459
+ len(trace.event_history),
460
+ len(trace.markov_blanket_message_history),
461
+ _json_dumps(trace.metadata or {}),
462
+ trace.session_id,
463
+ ),
464
+ )
465
+ conn.commit()
466
+
467
+ return trace.session_id
468
+
469
+ async def get_session_trace(self, session_id: str) -> dict[str, Any] | None:
470
+ await self.initialize()
471
+
472
+ async with self._op_lock:
473
+ conn = self._conn
474
+ assert conn is not None
475
+
476
+ session_cursor = conn.execute(
477
+ """
478
+ SELECT session_id,
479
+ created_at,
480
+ num_timesteps,
481
+ num_events,
482
+ num_messages,
483
+ metadata
484
+ FROM session_traces
485
+ WHERE session_id = ?
486
+ """,
487
+ (session_id,),
488
+ )
489
+ session_row = session_cursor.fetchone()
490
+ session_cursor.close()
491
+
492
+ if not session_row:
493
+ return None
494
+
495
+ session_columns = ["session_id", "created_at", "num_timesteps", "num_events", "num_messages", "metadata"]
496
+ session_data = dict(zip(session_columns, session_row))
497
+
498
+ timestep_cursor = conn.execute(
499
+ """
500
+ SELECT step_id,
501
+ step_index,
502
+ turn_number,
503
+ started_at,
504
+ completed_at,
505
+ step_metadata
506
+ FROM session_timesteps
507
+ WHERE session_id = ?
508
+ ORDER BY step_index ASC
509
+ """,
510
+ (session_id,),
511
+ )
512
+ timestep_rows = timestep_cursor.fetchall()
513
+ timestep_cursor.close()
514
+
515
+ return {
516
+ "session_id": session_data["session_id"],
517
+ "created_at": _maybe_datetime(session_data["created_at"]),
518
+ "num_timesteps": session_data["num_timesteps"],
519
+ "num_events": session_data["num_events"],
520
+ "num_messages": session_data["num_messages"],
521
+ "metadata": _load_json(session_data["metadata"]),
522
+ "timesteps": [
523
+ {
524
+ "step_id": row[0],
525
+ "step_index": row[1],
526
+ "turn_number": row[2],
527
+ "started_at": _maybe_datetime(row[3]),
528
+ "completed_at": _maybe_datetime(row[4]),
529
+ "metadata": _load_json(row[5]),
530
+ }
531
+ for row in timestep_rows
532
+ ],
533
+ }
534
+
535
+ async def _session_exists(self, session_id: str) -> bool:
536
+ await self.initialize()
537
+ async with self._op_lock:
538
+ conn = self._conn
539
+ assert conn is not None
540
+ cursor = conn.execute(
541
+ "SELECT 1 FROM session_traces WHERE session_id = ?", (session_id,)
542
+ )
543
+ row = cursor.fetchone()
544
+ cursor.close()
545
+ return row is not None
546
+
547
+ @staticmethod
548
+ def _normalise_params(params: dict[str, Any] | None) -> dict[str, Any]:
549
+ if not params:
550
+ return {}
551
+ normalised: dict[str, Any] = {}
552
+ for key, value in params.items():
553
+ if isinstance(value, datetime):
554
+ normalised[key] = value.isoformat()
555
+ else:
556
+ normalised[key] = value
557
+ return normalised
558
+
559
+ @staticmethod
560
+ def _prepare_query_params(query: str, params: dict[str, Any] | list[Any] | tuple[Any, ...]) -> tuple[str, tuple[Any, ...]]:
561
+ if isinstance(params, dict):
562
+ keys: list[str] = []
563
+
564
+ def _replace(match: re.Match[str]) -> str:
565
+ key = match.group(1)
566
+ keys.append(key)
567
+ return "?"
568
+
569
+ new_query = re.sub(r":([a-zA-Z_][a-zA-Z0-9_]*)", _replace, query)
570
+ if not keys:
571
+ raise ValueError("No named parameters found in query for provided mapping")
572
+ values = tuple(params[key] for key in keys)
573
+ return new_query, values
574
+ if isinstance(params, (list, tuple)):
575
+ return query, tuple(params)
576
+ raise TypeError("Unsupported parameter type for query execution")
577
+
578
+ def _ensure_schema(self) -> None:
579
+ if not self._conn:
580
+ raise RuntimeError("Connection not initialised")
581
+
582
+ for ddl in _TABLE_DEFINITIONS:
583
+ self._conn.execute(ddl)
584
+ for ddl in _INDEX_DEFINITIONS:
585
+ self._conn.execute(ddl)
586
+ for view_sql in analytics_views.values():
587
+ self._conn.execute(view_sql)
588
+ self._conn.commit()
589
+
590
+ async def query_traces(self, query: str, params: dict[str, Any] | None = None) -> Any:
591
+ await self.initialize()
592
+
593
+ async with self._op_lock:
594
+ conn = self._conn
595
+ assert conn is not None
596
+ normalised = self._normalise_params(params)
597
+ if normalised:
598
+ prepared_query, prepared_params = self._prepare_query_params(query, normalised)
599
+ cursor = conn.execute(prepared_query, prepared_params)
600
+ else:
601
+ cursor = conn.execute(query)
602
+ try:
603
+ description = cursor.description or []
604
+ columns = [col[0] for col in description]
605
+ rows = cursor.fetchall()
606
+ finally:
607
+ cursor.close()
608
+
609
+ if not rows:
610
+ if pd is not None:
611
+ return pd.DataFrame(columns=[col for col in columns])
612
+ return []
613
+
614
+ records = [dict(zip(columns, row)) for row in rows]
615
+ if pd is not None:
616
+ return pd.DataFrame(records)
617
+ return records
618
+
619
+ async def get_model_usage(
620
+ self,
621
+ start_date=None,
622
+ end_date=None,
623
+ model_name=None,
624
+ ) -> Any:
625
+ query = """
626
+ SELECT * FROM model_usage_stats
627
+ WHERE 1=1
628
+ """
629
+ params: dict[str, Any] = {}
630
+ if start_date:
631
+ params["start_date"] = start_date
632
+ query += " AND last_used >= :start_date"
633
+ if end_date:
634
+ params["end_date"] = end_date
635
+ query += " AND first_used <= :end_date"
636
+ if model_name:
637
+ params["model_name"] = model_name
638
+ query += " AND model_name = :model_name"
639
+ query += " ORDER BY usage_count DESC"
640
+ return await self.query_traces(query, params)
641
+
642
+ async def delete_session(self, session_id: str) -> bool:
643
+ await self.initialize()
644
+
645
+ async with self._op_lock:
646
+ conn = self._conn
647
+ assert conn is not None
648
+
649
+ cursor = conn.execute(
650
+ "SELECT 1 FROM session_traces WHERE session_id = ?", (session_id,)
651
+ )
652
+ exists = cursor.fetchone() is not None
653
+ cursor.close()
654
+ if not exists:
655
+ return False
656
+
657
+ conn.execute("DELETE FROM event_rewards WHERE session_id = ?", (session_id,))
658
+ conn.execute("DELETE FROM outcome_rewards WHERE session_id = ?", (session_id,))
659
+ conn.execute("DELETE FROM messages WHERE session_id = ?", (session_id,))
660
+ conn.execute("DELETE FROM events WHERE session_id = ?", (session_id,))
661
+ conn.execute("DELETE FROM session_timesteps WHERE session_id = ?", (session_id,))
662
+ conn.execute("DELETE FROM session_traces WHERE session_id = ?", (session_id,))
663
+ conn.commit()
664
+ return True
665
+
666
+ # Experiment helpers -------------------------------------------------
667
+ async def create_experiment(
668
+ self,
669
+ experiment_id: str,
670
+ name: str,
671
+ description: str | None = None,
672
+ configuration: dict[str, Any] | None = None,
673
+ ) -> str:
674
+ await self.initialize()
675
+
676
+ async with self._op_lock:
677
+ conn = self._conn
678
+ assert conn is not None
679
+ conn.execute(
680
+ """
681
+ INSERT INTO experiments (experiment_id, name, description, configuration)
682
+ VALUES (?, ?, ?, ?)
683
+ ON CONFLICT(experiment_id) DO UPDATE SET
684
+ name = excluded.name,
685
+ description = excluded.description,
686
+ configuration = excluded.configuration
687
+ """,
688
+ (
689
+ experiment_id,
690
+ name,
691
+ description,
692
+ _json_dumps(configuration or {}),
693
+ ),
694
+ )
695
+ conn.commit()
696
+ return experiment_id
697
+
698
+ async def link_session_to_experiment(self, session_id: str, experiment_id: str):
699
+ await self.initialize()
700
+
701
+ async with self._op_lock:
702
+ conn = self._conn
703
+ assert conn is not None
704
+ conn.execute(
705
+ "UPDATE session_traces SET experiment_id = ? WHERE session_id = ?",
706
+ (experiment_id, session_id),
707
+ )
708
+ conn.commit()
709
+
710
+ async def get_sessions_by_experiment(
711
+ self, experiment_id: str, limit: int | None = None
712
+ ) -> list[dict[str, Any]]:
713
+ await self.initialize()
714
+
715
+ sql = """
716
+ SELECT session_id,
717
+ created_at,
718
+ num_timesteps,
719
+ num_events,
720
+ num_messages,
721
+ metadata
722
+ FROM session_traces
723
+ WHERE experiment_id = ?
724
+ ORDER BY created_at DESC
725
+ """
726
+ params: list[Any] = [experiment_id]
727
+ if limit is not None:
728
+ sql += " LIMIT ?"
729
+ params.append(limit)
730
+
731
+ async with self._op_lock:
732
+ conn = self._conn
733
+ assert conn is not None
734
+ cursor = conn.execute(sql, params)
735
+ rows = cursor.fetchall()
736
+ cursor.close()
737
+
738
+ return [
739
+ {
740
+ "session_id": row[0],
741
+ "created_at": _maybe_datetime(row[1]),
742
+ "num_timesteps": row[2],
743
+ "num_events": row[3],
744
+ "num_messages": row[4],
745
+ "metadata": _load_json(row[5]),
746
+ }
747
+ for row in rows
748
+ ]
749
+
750
+ async def batch_insert_sessions(
751
+ self, traces: list[SessionTrace], batch_size: int | None = None
752
+ ) -> list[str]:
753
+ batch_size = batch_size or CONFIG.batch_size
754
+ inserted: list[str] = []
755
+
756
+ for i in range(0, len(traces), batch_size):
757
+ chunk = traces[i : i + batch_size]
758
+ for trace in chunk:
759
+ session_id = await self.insert_session_trace(trace)
760
+ inserted.append(session_id)
761
+ return inserted
762
+
763
+ # Incremental helpers -----------------------------------------------
764
+ async def ensure_session(
765
+ self,
766
+ session_id: str,
767
+ *,
768
+ created_at=None,
769
+ metadata=None,
770
+ ) -> None:
771
+ await self.initialize()
772
+
773
+ created_at_val = (created_at or datetime.now(UTC)).isoformat()
774
+ metadata_json = _json_dumps(metadata or {})
775
+
776
+ async with self._op_lock:
777
+ conn = self._conn
778
+
779
+ assert conn is not None
780
+ conn.execute(
781
+ """
782
+ INSERT INTO session_traces (
783
+ session_id, created_at, num_timesteps, num_events, num_messages, metadata
784
+ )
785
+ VALUES (?, ?, 0, 0, 0, ?)
786
+ ON CONFLICT(session_id) DO NOTHING
787
+ """,
788
+ (session_id, created_at_val, metadata_json),
789
+ )
790
+ conn.commit()
791
+
792
+ async def ensure_timestep(
793
+ self,
794
+ session_id: str,
795
+ *,
796
+ step_id: str,
797
+ step_index: int,
798
+ turn_number: int | None = None,
799
+ started_at=None,
800
+ completed_at=None,
801
+ metadata=None,
802
+ ) -> int:
803
+ await self.initialize()
804
+
805
+ started_at_val = (started_at or datetime.now(UTC)).isoformat()
806
+ completed_at_val = completed_at.isoformat() if completed_at else None
807
+ metadata_json = _json_dumps(metadata or {})
808
+
809
+ async with self._op_lock:
810
+ conn = self._conn
811
+
812
+ assert conn is not None
813
+ cur = conn.execute(
814
+ """
815
+ SELECT id FROM session_timesteps
816
+ WHERE session_id = ? AND step_id = ?
817
+ """,
818
+ (session_id, step_id),
819
+ )
820
+ row = cur.fetchone()
821
+ if row:
822
+ return int(row[0])
823
+
824
+ cur = conn.execute(
825
+ """
826
+ INSERT INTO session_timesteps (
827
+ session_id,
828
+ step_id,
829
+ step_index,
830
+ turn_number,
831
+ started_at,
832
+ completed_at,
833
+ num_events,
834
+ num_messages,
835
+ step_metadata
836
+ )
837
+ VALUES (?, ?, ?, ?, ?, ?, 0, 0, ?)
838
+ """,
839
+ (
840
+ session_id,
841
+ step_id,
842
+ step_index,
843
+ turn_number,
844
+ started_at_val,
845
+ completed_at_val,
846
+ metadata_json,
847
+ ),
848
+ )
849
+ timestep_id = int(cur.lastrowid)
850
+ conn.execute(
851
+ """
852
+ UPDATE session_traces
853
+ SET num_timesteps = num_timesteps + 1
854
+ WHERE session_id = ?
855
+ """,
856
+ (session_id,),
857
+ )
858
+ conn.commit()
859
+ return timestep_id
860
+
861
+ async def insert_event_row(
862
+ self,
863
+ session_id: str,
864
+ *,
865
+ timestep_db_id: int | None,
866
+ event: Any,
867
+ metadata_override: dict[str, Any] | None = None,
868
+ ) -> int:
869
+ await self.initialize()
870
+
871
+ if not isinstance(event, (EnvironmentEvent, LMCAISEvent, RuntimeEvent)):
872
+ raise TypeError(f"Unsupported event type for native manager: {type(event)!r}")
873
+
874
+ metadata_json = metadata_override or event.metadata or {}
875
+ event_extra_metadata = getattr(event, "event_metadata", None)
876
+ system_state_before = getattr(event, "system_state_before", None)
877
+ system_state_after = getattr(event, "system_state_after", None)
878
+
879
+ payload: dict[str, Any] = {
880
+ "session_id": session_id,
881
+ "timestep_id": timestep_db_id,
882
+ "system_instance_id": event.system_instance_id,
883
+ "event_time": event.time_record.event_time,
884
+ "message_time": event.time_record.message_time,
885
+ "metadata": metadata_json,
886
+ "event_metadata": event_extra_metadata,
887
+ "system_state_before": system_state_before,
888
+ "system_state_after": system_state_after,
889
+ }
890
+
891
+ if isinstance(event, LMCAISEvent):
892
+ call_records = None
893
+ if getattr(event, "call_records", None):
894
+ call_records = [asdict(record) for record in event.call_records]
895
+ payload.update(
896
+ {
897
+ "event_type": "cais",
898
+ "model_name": event.model_name,
899
+ "provider": event.provider,
900
+ "input_tokens": event.input_tokens,
901
+ "output_tokens": event.output_tokens,
902
+ "total_tokens": event.total_tokens,
903
+ "cost_usd": int(event.cost_usd * 100) if event.cost_usd is not None else None,
904
+ "latency_ms": event.latency_ms,
905
+ "span_id": event.span_id,
906
+ "trace_id": event.trace_id,
907
+ "call_records": call_records,
908
+ }
909
+ )
910
+ elif isinstance(event, EnvironmentEvent):
911
+ payload.update(
912
+ {
913
+ "event_type": "environment",
914
+ "reward": event.reward,
915
+ "terminated": event.terminated,
916
+ "truncated": event.truncated,
917
+ }
918
+ )
919
+ elif isinstance(event, RuntimeEvent):
920
+ payload.update(
921
+ {
922
+ "event_type": "runtime",
923
+ "metadata": {**(event.metadata or {}), "actions": event.actions},
924
+ }
925
+ )
926
+
927
+ async with self._op_lock:
928
+ conn = self._conn
929
+
930
+ assert conn is not None
931
+ cur = conn.execute(
932
+ """
933
+ INSERT INTO events (
934
+ session_id,
935
+ timestep_id,
936
+ event_type,
937
+ system_instance_id,
938
+ event_time,
939
+ message_time,
940
+ model_name,
941
+ provider,
942
+ input_tokens,
943
+ output_tokens,
944
+ total_tokens,
945
+ cost_usd,
946
+ latency_ms,
947
+ span_id,
948
+ trace_id,
949
+ call_records,
950
+ reward,
951
+ terminated,
952
+ truncated,
953
+ system_state_before,
954
+ system_state_after,
955
+ metadata,
956
+ event_metadata
957
+ )
958
+ VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
959
+ """,
960
+ (
961
+ payload["session_id"],
962
+ payload["timestep_id"],
963
+ payload.get("event_type"),
964
+ payload["system_instance_id"],
965
+ payload["event_time"],
966
+ payload["message_time"],
967
+ payload.get("model_name"),
968
+ payload.get("provider"),
969
+ payload.get("input_tokens"),
970
+ payload.get("output_tokens"),
971
+ payload.get("total_tokens"),
972
+ payload.get("cost_usd"),
973
+ payload.get("latency_ms"),
974
+ payload.get("span_id"),
975
+ payload.get("trace_id"),
976
+ _json_dumps(payload.get("call_records")),
977
+ payload.get("reward"),
978
+ payload.get("terminated"),
979
+ payload.get("truncated"),
980
+ _json_dumps(payload.get("system_state_before")),
981
+ _json_dumps(payload.get("system_state_after")),
982
+ _json_dumps(payload.get("metadata")),
983
+ _json_dumps(payload.get("event_metadata")),
984
+ ),
985
+ )
986
+ event_id = int(cur.lastrowid)
987
+ conn.execute(
988
+ """
989
+ UPDATE session_traces
990
+ SET num_events = num_events + 1
991
+ WHERE session_id = ?
992
+ """,
993
+ (session_id,),
994
+ )
995
+ if timestep_db_id is not None:
996
+ conn.execute(
997
+ """
998
+ UPDATE session_timesteps
999
+ SET num_events = num_events + 1
1000
+ WHERE id = ?
1001
+ """,
1002
+ (timestep_db_id,),
1003
+ )
1004
+ conn.commit()
1005
+ return event_id
1006
+
1007
+ async def insert_message_row(
1008
+ self,
1009
+ session_id: str,
1010
+ *,
1011
+ timestep_db_id: int | None,
1012
+ message_type: str,
1013
+ content: Any,
1014
+ event_time: float | None = None,
1015
+ message_time: int | None = None,
1016
+ metadata: dict[str, Any] | None = None,
1017
+ ) -> int:
1018
+ await self.initialize()
1019
+
1020
+ metadata_payload = dict(metadata or {})
1021
+ if isinstance(content, SessionMessageContent):
1022
+ if content.json_payload:
1023
+ metadata_payload.setdefault("json_payload", content.json_payload)
1024
+ content_value = content.json_payload
1025
+ else:
1026
+ content_value = content.as_text()
1027
+ if content.text:
1028
+ metadata_payload.setdefault("text", content.text)
1029
+ else:
1030
+ content_value = content
1031
+ if not isinstance(content_value, str):
1032
+ try:
1033
+ content_value = json.dumps(content_value, ensure_ascii=False)
1034
+ except (TypeError, ValueError):
1035
+ content_value = str(content_value)
1036
+
1037
+ async with self._op_lock:
1038
+ conn = self._conn
1039
+
1040
+ assert conn is not None
1041
+ cur = conn.execute(
1042
+ """
1043
+ INSERT INTO messages (
1044
+ session_id,
1045
+ timestep_id,
1046
+ message_type,
1047
+ content,
1048
+ event_time,
1049
+ message_time,
1050
+ metadata
1051
+ )
1052
+ VALUES (?, ?, ?, ?, ?, ?, ?)
1053
+ """,
1054
+ (
1055
+ session_id,
1056
+ timestep_db_id,
1057
+ message_type,
1058
+ content_value,
1059
+ event_time,
1060
+ message_time,
1061
+ _json_dumps(metadata_payload),
1062
+ ),
1063
+ )
1064
+ message_id = int(cur.lastrowid)
1065
+ conn.execute(
1066
+ """
1067
+ UPDATE session_traces
1068
+ SET num_messages = num_messages + 1
1069
+ WHERE session_id = ?
1070
+ """,
1071
+ (session_id,),
1072
+ )
1073
+ if timestep_db_id is not None:
1074
+ conn.execute(
1075
+ """
1076
+ UPDATE session_timesteps
1077
+ SET num_messages = num_messages + 1
1078
+ WHERE id = ?
1079
+ """,
1080
+ (timestep_db_id,),
1081
+ )
1082
+ conn.commit()
1083
+ return message_id
1084
+
1085
+ async def insert_outcome_reward(
1086
+ self,
1087
+ session_id: str,
1088
+ *,
1089
+ total_reward: int,
1090
+ achievements_count: int,
1091
+ total_steps: int,
1092
+ reward_metadata: dict | None = None,
1093
+ ) -> int:
1094
+ await self.initialize()
1095
+
1096
+ async with self._op_lock:
1097
+ conn = self._conn
1098
+
1099
+ assert conn is not None
1100
+ cur = conn.execute(
1101
+ """
1102
+ INSERT INTO outcome_rewards (
1103
+ session_id,
1104
+ total_reward,
1105
+ achievements_count,
1106
+ total_steps,
1107
+ created_at,
1108
+ reward_metadata
1109
+ )
1110
+ VALUES (?, ?, ?, ?, ?, ?)
1111
+ """,
1112
+ (
1113
+ session_id,
1114
+ total_reward,
1115
+ achievements_count,
1116
+ total_steps,
1117
+ datetime.now(UTC).isoformat(),
1118
+ _json_dumps(reward_metadata),
1119
+ ),
1120
+ )
1121
+ conn.commit()
1122
+ return int(cur.lastrowid)
1123
+
1124
+ async def insert_event_reward(
1125
+ self,
1126
+ session_id: str,
1127
+ *,
1128
+ event_id: int,
1129
+ message_id: int | None = None,
1130
+ turn_number: int | None = None,
1131
+ reward_value: float = 0.0,
1132
+ reward_type: str | None = None,
1133
+ key: str | None = None,
1134
+ annotation: dict[str, Any] | None = None,
1135
+ source: str | None = None,
1136
+ ) -> int:
1137
+ await self.initialize()
1138
+
1139
+ async with self._op_lock:
1140
+ conn = self._conn
1141
+
1142
+ assert conn is not None
1143
+ cur = conn.execute(
1144
+ """
1145
+ INSERT INTO event_rewards (
1146
+ event_id,
1147
+ session_id,
1148
+ message_id,
1149
+ turn_number,
1150
+ reward_value,
1151
+ reward_type,
1152
+ key,
1153
+ annotation,
1154
+ source,
1155
+ created_at
1156
+ )
1157
+ VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
1158
+ """,
1159
+ (
1160
+ event_id,
1161
+ session_id,
1162
+ message_id,
1163
+ turn_number,
1164
+ reward_value,
1165
+ reward_type,
1166
+ key,
1167
+ _json_dumps(annotation),
1168
+ source,
1169
+ datetime.now(UTC).isoformat(),
1170
+ ),
1171
+ )
1172
+ conn.commit()
1173
+ return int(cur.lastrowid)