caesar-agent 0.4.28__tar.gz → 0.4.30__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 (149) hide show
  1. {caesar_agent-0.4.28/caesar_agent.egg-info → caesar_agent-0.4.30}/PKG-INFO +3 -3
  2. {caesar_agent-0.4.28 → caesar_agent-0.4.30/caesar_agent.egg-info}/PKG-INFO +3 -3
  3. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/pyproject.toml +1 -1
  4. caesar_agent-0.4.30/requirements.txt +71 -0
  5. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/llm_handler.py +36 -30
  6. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_compressor.py +1 -1
  7. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_llm_handler.py +3 -3
  8. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_reasoning_model_gating.py +3 -4
  9. caesar_agent-0.4.28/requirements.txt +0 -64
  10. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/LICENSE +0 -0
  11. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/NOTICE +0 -0
  12. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/README.md +0 -0
  13. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/__init__.py +0 -0
  14. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/analysis/judge_analysis.py +0 -0
  15. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/analysis/llm_as_judge.py +0 -0
  16. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/analysis/prepare_artifact.py +0 -0
  17. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/analysis/transfer_configs.py +0 -0
  18. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/artifact_synthesis.py +0 -0
  19. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/arxiv_explorer.py +0 -0
  20. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/brave_search.py +0 -0
  21. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/caesar_agent.py +0 -0
  22. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/caesar_config.py +0 -0
  23. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/checkpoint.py +0 -0
  24. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_agi/improve_transformer.yaml +0 -0
  25. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_agi/intelligence_bottleneck.yaml +0 -0
  26. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_agi/single_agent_agi.yaml +0 -0
  27. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_agi/solve_arcagi.yaml +0 -0
  28. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_creative/constrained_creativity.yaml +0 -0
  29. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_creative/counterfactual_reasoning.yaml +0 -0
  30. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_creative/crossdomain_synthesis.yaml +0 -0
  31. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_creative/meta_creativity.yaml +0 -0
  32. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_creative/openended_creativity.yaml +0 -0
  33. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_preset/arxiv.yaml +0 -0
  34. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_preset/mini.yaml +0 -0
  35. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_preset/nano.yaml +0 -0
  36. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_preset/regular.yaml +0 -0
  37. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_test/batch_test.jsonl +0 -0
  38. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_test/batch_test.yaml +0 -0
  39. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_test/single_agent_test.yaml +0 -0
  40. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_wikipedia/single_agent_ai.yaml +0 -0
  41. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_wikipedia/single_agent_evo.yaml +0 -0
  42. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/config_wikipedia/single_agent_sci.yaml +0 -0
  43. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/custom_role/self_transcend_insights_v1.txt +0 -0
  44. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/custom_role/self_transcend_insights_v2.txt +0 -0
  45. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/custom_role/self_transcend_insights_v3.txt +0 -0
  46. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/custom_role/self_transcend_insights_v4.txt +0 -0
  47. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/custom_role/self_transcend_role_v1.txt +0 -0
  48. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/custom_role/self_transcend_role_v2.txt +0 -0
  49. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/custom_role/self_transcend_role_v3.txt +0 -0
  50. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/agent_names/agent_names_all.txt +0 -0
  51. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/agent_names/agent_names_caesar.txt +0 -0
  52. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/human_prompts/human_eval_prompt.txt +0 -0
  53. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/human_prompts/human_eval_prompt2.txt +0 -0
  54. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/llm_prompts/eli5_prompt_long.txt +0 -0
  55. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/llm_prompts/eli5_prompt_medium.txt +0 -0
  56. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/llm_prompts/eli5_prompt_short.txt +0 -0
  57. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/llm_prompts/judge_csv_prompt.txt +0 -0
  58. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/llm_prompts/judge_summary_prompt.txt +0 -0
  59. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/llm_prompts/nus_rubric_10pt_api.txt +0 -0
  60. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/llm_prompts/nus_rubric_10pt_new.txt +0 -0
  61. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/llm_prompts/nus_rubric_10pt_old.txt +0 -0
  62. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/llm_prompts/nus_rubric_4pt_api.txt +0 -0
  63. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/llm_prompts/query_prompt_deep.txt +0 -0
  64. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/llm_prompts/query_prompt_shallow.txt +0 -0
  65. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/queries/queries_n50.json +0 -0
  66. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/queries/queries_n50.jsonl +0 -0
  67. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/templates/answer_bat_ca1.txt +0 -0
  68. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/templates/answer_bear_ca3.txt +0 -0
  69. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/templates/answer_cat_cam.txt +0 -0
  70. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/templates/answer_cow_cld.txt +0 -0
  71. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/templates/answer_dog_cls.txt +0 -0
  72. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/templates/answer_fox_ged.txt +0 -0
  73. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/templates/answer_lion_ges.txt +0 -0
  74. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/templates/answer_owl_gpd.txt +0 -0
  75. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/templates/answer_pig_gps.txt +0 -0
  76. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/templates/judge_claude.txt +0 -0
  77. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/templates/judge_gemini.txt +0 -0
  78. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/templates/judge_gpt.txt +0 -0
  79. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/config/llm_as_judge/templates/query.txt +0 -0
  80. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/image_generator.py +0 -0
  81. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/query_result/1_10_all_agent_answers/nus_rubric_study/analysis/agg5.py +0 -0
  82. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/query_result/1_10_all_agent_answers/nus_rubric_study/analysis/build5.py +0 -0
  83. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/query_result/1_10_all_agent_answers/nus_rubric_study/analysis/cmp.py +0 -0
  84. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/query_result/1_10_all_agent_answers/nus_rubric_study/analysis/compare.py +0 -0
  85. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/query_result/1_10_all_agent_answers/nus_rubric_study/analysis/diag.py +0 -0
  86. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/query_result/1_10_all_agent_answers/nus_rubric_study/analysis/summ.py +0 -0
  87. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/query_result/1_10_all_agent_answers/nus_rubric_study/analysis/tally.py +0 -0
  88. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/query_result/1_10_all_agent_answers/nus_rubric_study/analysis/v3an.py +0 -0
  89. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/query_result/1_10_all_agent_answers/nus_rubric_study/analysis/v4an.py +0 -0
  90. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/query_result/1_10_all_agent_answers/nus_rubric_study/analysis/v5an.py +0 -0
  91. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/query_result/1_10_all_agent_answers/nus_rubric_study/analysis/why.py +0 -0
  92. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/run_agent.py +0 -0
  93. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/semantic_scholar.py +0 -0
  94. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar/web_explorer.py +0 -0
  95. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar_agent.egg-info/SOURCES.txt +0 -0
  96. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar_agent.egg-info/dependency_links.txt +0 -0
  97. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar_agent.egg-info/entry_points.txt +0 -0
  98. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar_agent.egg-info/requires.txt +2 -2
  99. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/caesar_agent.egg-info/top_level.txt +0 -0
  100. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/__init__.py +0 -0
  101. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/action/__init__.py +0 -0
  102. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/action/action.py +0 -0
  103. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/action/advanced_reset_action.py +0 -0
  104. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/action/edit_code_action.py +0 -0
  105. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/action/edit_test_action.py +0 -0
  106. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/action/exec_code_action.py +0 -0
  107. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/action/priority_search_action.py +0 -0
  108. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/action/reset_action.py +0 -0
  109. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/action/revert_code_action.py +0 -0
  110. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/action/save_kb_action.py +0 -0
  111. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/action/tournament_search_action.py +0 -0
  112. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/action/transition_action.py +0 -0
  113. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/action_selector.py +0 -0
  114. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/agent.py +0 -0
  115. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/agent_api.py +0 -0
  116. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/agent_memory.py +0 -0
  117. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/base_agent.py +0 -0
  118. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/config.py +0 -0
  119. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/database.py +0 -0
  120. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/executor.py +0 -0
  121. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/fsm.py +0 -0
  122. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/fsm_selector.py +0 -0
  123. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/history.py +0 -0
  124. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/kb_client.py +0 -0
  125. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/kb_server.py +0 -0
  126. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/logger.py +0 -0
  127. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/metadata.py +0 -0
  128. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/multi_agent.py +0 -0
  129. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/parsing.py +0 -0
  130. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/process.py +0 -0
  131. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/repository.py +0 -0
  132. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/rome/state.py +0 -0
  133. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/setup.cfg +0 -0
  134. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_arxiv_explorer.py +0 -0
  135. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_arxiv_integration_manual.py +0 -0
  136. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_database.py +0 -0
  137. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_evalplus_agent.py +0 -0
  138. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_executor.py +0 -0
  139. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_fsm.py +0 -0
  140. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_kb_rerank_fallback.py +0 -0
  141. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_knowledge_base.py +0 -0
  142. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_logger.py +0 -0
  143. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_metadata.py +0 -0
  144. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_multi_agent.py +0 -0
  145. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_parsing.py +0 -0
  146. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_public_mode_key.py +0 -0
  147. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_search.py +0 -0
  148. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_single_agent.py +0 -0
  149. {caesar_agent-0.4.28 → caesar_agent-0.4.30}/test/test_url_guard.py +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: caesar-agent
3
- Version: 0.4.28
3
+ Version: 0.4.30
4
4
  Summary: Caesar: autonomous AI research agent with graph-based deep web exploration and adversarial answer synthesis.
5
5
  Author: Jason Liang, Elliot Meyerson, Risto Miikkulainen
6
6
  Maintainer-email: Jason Liang <jasonzliang@utexas.edu>
@@ -36,11 +36,11 @@ Requires-Dist: scipy
36
36
  Requires-Dist: tinydb
37
37
  Requires-Dist: chromadb
38
38
  Requires-Dist: llama-index
39
- Requires-Dist: llama-index-vector-stores-chroma
40
39
  Requires-Dist: llama-index-embeddings-openai
41
40
  Requires-Dist: llama-index-llms-openai
42
- Requires-Dist: pillow
41
+ Requires-Dist: llama-index-vector-stores-chroma
43
42
  Requires-Dist: mem0ai
43
+ Requires-Dist: pillow
44
44
  Requires-Dist: anthropic
45
45
  Requires-Dist: beautifulsoup4
46
46
  Requires-Dist: curl_cffi>=0.16.3
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: caesar-agent
3
- Version: 0.4.28
3
+ Version: 0.4.30
4
4
  Summary: Caesar: autonomous AI research agent with graph-based deep web exploration and adversarial answer synthesis.
5
5
  Author: Jason Liang, Elliot Meyerson, Risto Miikkulainen
6
6
  Maintainer-email: Jason Liang <jasonzliang@utexas.edu>
@@ -36,11 +36,11 @@ Requires-Dist: scipy
36
36
  Requires-Dist: tinydb
37
37
  Requires-Dist: chromadb
38
38
  Requires-Dist: llama-index
39
- Requires-Dist: llama-index-vector-stores-chroma
40
39
  Requires-Dist: llama-index-embeddings-openai
41
40
  Requires-Dist: llama-index-llms-openai
42
- Requires-Dist: pillow
41
+ Requires-Dist: llama-index-vector-stores-chroma
43
42
  Requires-Dist: mem0ai
43
+ Requires-Dist: pillow
44
44
  Requires-Dist: anthropic
45
45
  Requires-Dist: beautifulsoup4
46
46
  Requires-Dist: curl_cffi>=0.16.3
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
4
4
 
5
5
  [project]
6
6
  name = "caesar-agent"
7
- version = "0.4.28"
7
+ version = "0.4.30"
8
8
  authors = [
9
9
  {name = "Jason Liang"},
10
10
  {name = "Elliot Meyerson"},
@@ -0,0 +1,71 @@
1
+ # Runtime dependencies. pyproject.toml reads this file through setuptools' dynamic
2
+ # `dependencies`, so every non-comment line must be a valid PEP 508 specifier: pip's `-r
3
+ # other-file` is not expanded by setuptools and once broke `pip install .` with
4
+ # `InvalidRequirement: Expected package name at the start of dependency specifier`.
5
+ #
6
+ # Conventions: one package per line, alphabetical within a section, a version floor only where a
7
+ # release is known to break (each floor carries its reason directly above it), and every note as
8
+ # a full-line comment above the package it describes, wrapped at 96 columns.
9
+
10
+ # Core
11
+ litellm
12
+ networkx
13
+ numpy
14
+ openai
15
+ portalocker
16
+ psutil
17
+ pytest
18
+ scipy
19
+ tinydb
20
+
21
+ # Knowledge base and memory
22
+ chromadb
23
+ llama-index
24
+ llama-index-embeddings-openai
25
+ llama-index-llms-openai
26
+ llama-index-vector-stores-chroma
27
+ # Not installed: mem0ai[graph].
28
+ mem0ai
29
+ # llama-index-core imports PIL at import time and does not always pull it in transitively.
30
+ pillow
31
+
32
+ # Caesar agent
33
+ anthropic
34
+ beautifulsoup4
35
+ # Floor, not a pin: 0.15.0 segfaulted the web API under the threaded quick-explore fetch pool
36
+ # (2026-09-24); 0.16 ships curl 8.21 and curl-impersonate 2.0.
37
+ curl_cffi>=0.16.3
38
+ ddgs
39
+ # Floor, not a pin: 2.2.0 was 14 minor versions behind on a preview surface (Interactions / Deep
40
+ # Research) whose vendor's standing advice for intermittent 400 INVALID_ARGUMENT on that
41
+ # endpoint is "update the SDK". Upgraded 2026-08-01 and re-verified every surface the adapters
42
+ # touch: the agent id and both agent_config keys are still declared, and a live shape test
43
+ # reached the server (rejected on credentials, not structure). The SDK version is not part of
44
+ # adapter_fp.
45
+ google-genai>=2.16.0
46
+ pygraphviz
47
+ pypdf
48
+ # Reached the packaging path through crest/requirements.txt until crest moved to its own repo.
49
+ # rome imports yaml directly (rome/config.py, rome/base_agent.py, web_server/api/app/config.py,
50
+ # benchmark/evalplus_multi_agent.py), so it is declared here.
51
+ pyyaml>=6.0
52
+ tenacity
53
+ # Floor, not a pin: PR #856 fixes the Document() mutable-default lxml-element race that
54
+ # segfaulted under ThreadPoolExecutor.
55
+ trafilatura>=2.1.0
56
+
57
+ # Imported at module import time, so declared here rather than relied on to arrive transitively.
58
+ # fastapi backs rome/agent_api.py, which rome/agent.py imports whenever Agent.agent_api is true;
59
+ # it defaults to true, so rome.Agent could not start from a clean install without it.
60
+ fastapi
61
+ pytz
62
+ requests
63
+ rich
64
+
65
+ # Optional features. llmlingua (prompt compression) and evalplus (benchmark harness) live in
66
+ # pyproject's [project.optional-dependencies]: llmlingua alone pulls torch and transformers,
67
+ # about 730 MB, for a feature reached through one lazy import.
68
+ graphviz
69
+ # Not installed: ribs.
70
+ tiktoken
71
+ xxhash
@@ -6,6 +6,15 @@ import threading
6
6
  import warnings
7
7
  from typing import Dict, Optional, Any, Union, List
8
8
 
9
+ # litellm reads its model price map at import: from GitHub over HTTPS, or the copy
10
+ # bundled with the installed release when this variable is "True" or the fetch fails.
11
+ # The bundled copy is the same file the fetch falls back to, and LLMHandler registers
12
+ # its own price table over either (see _register_known_models), so nothing here
13
+ # depends on the remote copy. Defaulting to the bundle saves one network round trip
14
+ # per process start and lets an offline start behave like an online one; an explicit
15
+ # LITELLM_LOCAL_MODEL_COST_MAP=False in the environment still restores the fetch.
16
+ os.environ.setdefault("LITELLM_LOCAL_MODEL_COST_MAP", "True")
17
+
9
18
  # Mute litellm's `asyncio.get_event_loop()` DeprecationWarning fired at import
10
19
  # time on Python 3.12+. Upstream issue; filter is scoped to the litellm import
11
20
  # so we don't muffle anyone else's deprecation warnings.
@@ -99,13 +108,6 @@ class LLMHandler:
99
108
  # OpenAI GPT-4o series
100
109
  "gpt-4o": {"input": 2.5, "output": 10.0},
101
110
  "gpt-4o-mini": {"input": 0.15, "output": 0.6},
102
- # OpenAI o-series reasoning models
103
- "o1": {"input": 15.0, "output": 60.0},
104
- "o1-mini": {"input": 1.10, "output": 4.40},
105
- "o1-pro": {"input": 150.0, "output": 600.0},
106
- "o3": {"input": 2.0, "output": 8.0},
107
- "o3-mini": {"input": 1.10, "output": 4.40},
108
- "o4-mini": {"input": 1.10, "output": 4.40},
109
111
  # OpenAI Realtime models
110
112
  "gpt-realtime": {"input": 4.0, "output": 16.0},
111
113
  "gpt-realtime-mini": {"input": 0.60, "output": 2.40},
@@ -142,13 +144,6 @@ class LLMHandler:
142
144
  # OpenAI GPT-4o series
143
145
  "gpt-4o": 128000,
144
146
  "gpt-4o-mini": 128000,
145
- # OpenAI o-series reasoning models
146
- "o1": 200000,
147
- "o1-mini": 128000,
148
- "o1-pro": 200000,
149
- "o3": 200000,
150
- "o3-mini": 200000,
151
- "o4-mini": 200000,
152
147
  # OpenAI Realtime models
153
148
  "gpt-realtime": 128000,
154
149
  "gpt-realtime-mini": 128000,
@@ -162,19 +157,18 @@ class LLMHandler:
162
157
  "gemini-3.1-flash-lite-preview": 1048576,
163
158
  }
164
159
 
165
- # Models that support reasoning_effort or equivalent thinking parameters.
166
- # is_reasoning_model() also falls back to a `gpt-5*` prefix match so any
167
- # future GPT-5.x release (e.g. 5.7) is auto-detected. That prefix rule does
168
- # NOT cover the o-series, so every o-model must be listed explicitly.
160
+ # Models that take reasoning_effort and reject temperature. is_reasoning_model()
161
+ # also accepts any `gpt-5*` name, so new GPT-5.x releases gate correctly before
162
+ # they are listed here; the explicit entries keep the older ones visible.
169
163
  #
170
164
  # This is the single source of truth. rome/kb_client.py kept a hand-maintained
171
165
  # second copy until it drifted: its copy was missing the whole GPT-5.6 family,
172
166
  # so the `mini` preset's KB model (gpt-5.6-luna) failed the membership test
173
167
  # and had its configured reasoning_effort silently dropped at construction.
174
168
  REASONING_MODELS = {
175
- "gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna", "gpt-5.5-pro", "gpt-5.5", "gpt-5.4", "gpt-5.2", "gpt-5.1", "gpt-5", "gpt-5.4-mini", "gpt-5-mini", "gpt-5-nano", "gpt-5-pro", "o1", "o1-mini", "o1-pro", "o3", "o3-mini", "o4-mini",
169
+ "gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna", "gpt-5.5-pro", "gpt-5.5", "gpt-5.4", "gpt-5.2", "gpt-5.1", "gpt-5", "gpt-5.4-mini", "gpt-5-mini", "gpt-5-nano", "gpt-5-pro",
176
170
  # Merged in from kb_client's former copy.
177
- "gpt-5.1-codex-max", "gpt-5.2-pro", "gpt-5.4-nano", "gpt-5.4-pro", "o3-pro",
171
+ "gpt-5.1-codex-max", "gpt-5.2-pro", "gpt-5.4-nano", "gpt-5.4-pro",
178
172
  }
179
173
 
180
174
  # Unique identifier for chat completion requests
@@ -183,14 +177,16 @@ class LLMHandler:
183
177
  @classmethod
184
178
  def synthesis_models(cls) -> list[str]:
185
179
  """OpenAI current-generation text models offered as synthesis targets:
186
- the GPT-5.x family and the o-series, in MODEL_PRICING order. The legacy
187
- GPT-4.x line, audio/realtime models, and non-OpenAI providers are all
188
- excluded by construction, so there is no second list to keep in sync
189
- with MODEL_PRICING — add a new gpt-5.x/o-series entry there and it is
190
- offered automatically."""
180
+ the GPT-5.x family, in MODEL_PRICING order. The legacy GPT-4.x line,
181
+ audio/realtime models, and non-OpenAI providers are all excluded by
182
+ construction, so there is no second list to keep in sync with
183
+ MODEL_PRICING — add a new gpt-5.x entry there and it is offered
184
+ automatically. The o-series was dropped once OpenAI scheduled o1, o1-pro,
185
+ o3-mini and o4-mini for retirement (2026-10-23) and had already retired
186
+ o1-mini."""
191
187
  return [
192
188
  m for m in cls.MODEL_PRICING
193
- if (m.startswith("gpt-5") or re.match(r"o\d", m)) and "realtime" not in m
189
+ if m.startswith("gpt-5") and "realtime" not in m
194
190
  ]
195
191
 
196
192
  def __init__(self, config: Dict = None):
@@ -235,8 +231,8 @@ class LLMHandler:
235
231
 
236
232
  litellm loads its map at import: fetched from GitHub, or the copy bundled with the
237
233
  installed release when LITELLM_LOCAL_MODEL_COST_MAP=True or the fetch fails. That copy
238
- ages with the pin (1.95.0's bundle prices gpt-5.6-luna at 5x today's rate and has no
239
- o1-mini at all), and completion_cost() reads it before this handler's own table. Registering
234
+ ages with the pin (1.95.0's bundle prices gpt-5.6-luna at 5x today's rate) and may lack
235
+ a model this table carries, and completion_cost() reads it before this handler's own table. Registering
240
236
  the table makes every model here resolvable and priced the way the table says, whichever
241
237
  map was loaded. Entries litellm already has keep their other fields; only what is set here
242
238
  is overridden.
@@ -311,9 +307,19 @@ class LLMHandler:
311
307
 
312
308
  @lru_cache(maxsize=1)
313
309
  def _get_model_context_length(self) -> int:
314
- """Get context length for the current model."""
310
+ """Context window of the current model, in input tokens.
311
+
312
+ litellm.get_max_tokens() answers a different question: the largest completion the
313
+ model will produce. Read as a context window it made gpt-4o a 16k model and gpt-5.4 a
314
+ 128k one, so context management compressed and truncated far too early. The window is
315
+ get_model_info()['max_input_tokens']; the handler's own table (registered into that map
316
+ at init) is the fallback for a model litellm does not know.
317
+ """
315
318
  try:
316
- return litellm.get_max_tokens(self._get_litellm_model())
319
+ info = litellm.get_model_info(self._get_litellm_model())
320
+ window = info.get("max_input_tokens") or info.get("max_tokens")
321
+ if window:
322
+ return int(window)
317
323
  except Exception:
318
324
  pass
319
325
  base_model = self._get_base_model()
@@ -55,7 +55,7 @@ class TestRealContextCompression:
55
55
  config.update(config_overrides)
56
56
 
57
57
  with patch.dict(os.environ, {'OPENAI_API_KEY': 'test-key'}):
58
- with patch('litellm.get_max_tokens', return_value=4096):
58
+ with patch('litellm.get_model_info', return_value={'max_input_tokens': 4096}):
59
59
  return LLMHandler(config)
60
60
 
61
61
  def run_test(self, test_name: str, test_func, **kwargs):
@@ -91,7 +91,7 @@ class TestLLMHandler:
91
91
  logger.configure(log_config)
92
92
  with patch('litellm.completion', return_value=_make_mock_response()):
93
93
  with patch('litellm.token_counter', return_value=20):
94
- with patch('litellm.get_max_tokens', return_value=128000):
94
+ with patch('litellm.get_model_info', return_value={'max_input_tokens': 128000}):
95
95
  handler = LLMHandler(basic_config)
96
96
  # Patch litellm calls for subsequent test usage
97
97
  handler._completion_patch = patch('litellm.completion', return_value=_make_mock_response())
@@ -368,7 +368,7 @@ class TestLLMHandler:
368
368
  })
369
369
 
370
370
  with patch('litellm.token_counter', return_value=20):
371
- with patch('litellm.get_max_tokens', return_value=128000):
371
+ with patch('litellm.get_model_info', return_value={'max_input_tokens': 128000}):
372
372
  handler = LLMHandler(basic_config)
373
373
 
374
374
  assert handler.cost_limit is None
@@ -406,7 +406,7 @@ class TestLLMHandler:
406
406
 
407
407
  def test_map_reasoning_params_openai(self, handler):
408
408
  """Test reasoning parameter mapping for OpenAI"""
409
- kwargs = {"model": "o3", "reasoning_effort": "high", "temperature": 0.5}
409
+ kwargs = {"model": "gpt-5.4", "reasoning_effort": "high", "temperature": 0.5}
410
410
  result = handler._map_reasoning_params(kwargs)
411
411
  assert result["reasoning_effort"] == "high"
412
412
  assert "temperature" not in result
@@ -32,8 +32,7 @@ def test_kb_client_defines_no_second_copy():
32
32
  "gpt-5.6-sol",
33
33
  "gpt-5.6-terra",
34
34
  "gpt-5-mini", # the default KB model
35
- "o3-pro", # o-series: no prefix rule covers it, must be listed
36
- "o1-pro",
35
+ "gpt-5.4-pro", # listed explicitly and covered by the prefix rule
37
36
  ],
38
37
  )
39
38
  def test_reasoning_models_are_gated_in(model):
@@ -58,11 +57,11 @@ def test_provider_prefix_is_stripped():
58
57
  def test_classmethod_matches_the_instance_method():
59
58
  """Both must answer identically -- the instance form is the older API."""
60
59
  handler = LLMHandler()
61
- for model in ("gpt-5.6-luna", "o3-pro", "gpt-4o", "gpt-5.9-unreleased"):
60
+ for model in ("gpt-5.6-luna", "gpt-5.4-pro", "gpt-4o", "gpt-5.9-unreleased"):
62
61
  assert handler._is_reasoning_model(model) is LLMHandler.is_reasoning_model(model)
63
62
 
64
63
 
65
- def test_every_priced_gpt5_and_o_model_is_reasoning_capable():
64
+ def test_every_priced_gpt5_model_is_reasoning_capable():
66
65
  """Anything offered for synthesis must gate correctly."""
67
66
  for model in LLMHandler.synthesis_models():
68
67
  assert LLMHandler.is_reasoning_model(model) is True, model
@@ -1,64 +0,0 @@
1
- # pyproject reads this file via setuptools' dynamic dependencies. Do NOT use pip's `-r` to pull in
2
- # another file here: setuptools does not expand -r and passes the literal line through as a PEP 508
3
- # specifier, which made `pip install .` die with
4
- # `InvalidRequirement: Expected package name at the start of dependency specifier`.
5
-
6
- # Rome Core Packages
7
- litellm
8
- networkx
9
- numpy
10
- openai
11
- portalocker
12
- psutil
13
- pytest
14
- scipy
15
- tinydb
16
-
17
- # KB/Memory Packages
18
- chromadb
19
- llama-index
20
- llama-index-vector-stores-chroma
21
- llama-index-embeddings-openai
22
- llama-index-llms-openai
23
- pillow # llama-index-core imports PIL at import time; not always pulled transitively
24
- # mem0ai[graph]
25
- mem0ai
26
-
27
- # Casear Agent Packages
28
- anthropic
29
- beautifulsoup4
30
- curl_cffi>=0.16.3 # floor, not a pin: 0.15.0 segfaulted the web API under threaded quick-explore (2026-09-24); 0.16 ships curl 8.21 + curl-impersonate 2.0 bugfixes
31
- ddgs
32
- google-genai>=2.16.0 # floor, not a pin: 2.2.0 was 14 minor versions behind on a PREVIEW
33
- # surface (Interactions / Deep Research) whose vendor's standing advice
34
- # for intermittent 400 INVALID_ARGUMENT on that endpoint is 'update the
35
- # SDK'. Upgraded 2026-08-01 and re-verified every surface the adapters
36
- # touch: agent id and both agent_config keys are still declared, and a
37
- # live shape test reached the server (rejected on credentials, not
38
- # structure). NB the SDK version is NOT in adapter_fp -- see below.
39
- pygraphviz
40
- pypdf
41
- # Reached the packaging path via crest/requirements.txt until crest moved to its own repo. rome
42
- # imports yaml directly (rome/config.py, rome/base_agent.py, web_server/api/app/config.py,
43
- # benchmark/evalplus_multi_agent.py), so it has to be declared here now.
44
- pyyaml>=6.0
45
- tenacity
46
- trafilatura>=2.1.0 # PR #856 fixes Document() mutable-default lxml-element race that segfaulted under ThreadPoolExecutor
47
-
48
- # Imported at module import time, so they must be declared rather than relied on
49
- # to arrive transitively. fastapi backs rome/agent_api.py, which rome/agent.py
50
- # imports whenever Agent.agent_api is true -- and it defaults to true, so
51
- # rome.Agent could not start from a clean install without it.
52
- fastapi
53
- pytz
54
- requests
55
- rich
56
-
57
- # Rome Optional Packages
58
- # NB llmlingua (prompt compression) and evalplus (benchmark harness) moved to
59
- # pyproject [project.optional-dependencies]. llmlingua alone pulls torch and
60
- # transformers -- ~730MB for a feature reached through one lazy import.
61
- graphviz
62
- # ribs
63
- tiktoken
64
- xxhash
File without changes
File without changes
File without changes
@@ -9,11 +9,11 @@ scipy
9
9
  tinydb
10
10
  chromadb
11
11
  llama-index
12
- llama-index-vector-stores-chroma
13
12
  llama-index-embeddings-openai
14
13
  llama-index-llms-openai
15
- pillow
14
+ llama-index-vector-stores-chroma
16
15
  mem0ai
16
+ pillow
17
17
  anthropic
18
18
  beautifulsoup4
19
19
  curl_cffi>=0.16.3
File without changes
File without changes