@illuma-ai/agents 1.1.20 → 1.1.22

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 (246) hide show
  1. package/dist/cjs/graphs/Graph.cjs +12 -1
  2. package/dist/cjs/graphs/Graph.cjs.map +1 -1
  3. package/dist/cjs/graphs/MultiAgentGraph.cjs +85 -1
  4. package/dist/cjs/graphs/MultiAgentGraph.cjs.map +1 -1
  5. package/dist/cjs/llm/bedrock/index.cjs +14 -0
  6. package/dist/cjs/llm/bedrock/index.cjs.map +1 -1
  7. package/dist/cjs/run.cjs +20 -9
  8. package/dist/cjs/run.cjs.map +1 -1
  9. package/dist/esm/graphs/Graph.mjs +12 -1
  10. package/dist/esm/graphs/Graph.mjs.map +1 -1
  11. package/dist/esm/graphs/MultiAgentGraph.mjs +85 -1
  12. package/dist/esm/graphs/MultiAgentGraph.mjs.map +1 -1
  13. package/dist/esm/llm/bedrock/index.mjs +14 -0
  14. package/dist/esm/llm/bedrock/index.mjs.map +1 -1
  15. package/dist/esm/run.mjs +20 -9
  16. package/dist/esm/run.mjs.map +1 -1
  17. package/dist/types/graphs/MultiAgentGraph.d.ts +17 -0
  18. package/package.json +1 -1
  19. package/src/graphs/Graph.ts +12 -1
  20. package/src/graphs/MultiAgentGraph.ts +105 -1
  21. package/src/graphs/__tests__/multi-agent-delegate.test.ts +191 -0
  22. package/src/llm/bedrock/index.ts +17 -0
  23. package/src/run.ts +20 -11
  24. package/src/scripts/test-bedrock-handoff-autonomous.ts +231 -0
  25. package/src/agents/AgentContext.js +0 -782
  26. package/src/agents/AgentContext.test.js +0 -421
  27. package/src/agents/__tests__/AgentContext.test.js +0 -678
  28. package/src/agents/__tests__/resolveStructuredOutputMode.test.js +0 -117
  29. package/src/common/enum.js +0 -192
  30. package/src/common/index.js +0 -3
  31. package/src/events.js +0 -166
  32. package/src/graphs/Graph.js +0 -1857
  33. package/src/graphs/MultiAgentGraph.js +0 -1092
  34. package/src/graphs/__tests__/structured-output.integration.test.js +0 -624
  35. package/src/graphs/__tests__/structured-output.test.js +0 -144
  36. package/src/graphs/contextManagement.e2e.test.js +0 -718
  37. package/src/graphs/contextManagement.test.js +0 -485
  38. package/src/graphs/handoffValidation.test.js +0 -276
  39. package/src/graphs/index.js +0 -3
  40. package/src/index.js +0 -28
  41. package/src/instrumentation.js +0 -21
  42. package/src/llm/anthropic/index.js +0 -319
  43. package/src/llm/anthropic/types.js +0 -46
  44. package/src/llm/anthropic/utils/message_inputs.js +0 -627
  45. package/src/llm/anthropic/utils/message_outputs.js +0 -290
  46. package/src/llm/anthropic/utils/output_parsers.js +0 -89
  47. package/src/llm/anthropic/utils/tools.js +0 -25
  48. package/src/llm/bedrock/__tests__/bedrock-caching.test.js +0 -392
  49. package/src/llm/bedrock/index.js +0 -303
  50. package/src/llm/bedrock/types.js +0 -2
  51. package/src/llm/bedrock/utils/index.js +0 -6
  52. package/src/llm/bedrock/utils/message_inputs.js +0 -463
  53. package/src/llm/bedrock/utils/message_outputs.js +0 -269
  54. package/src/llm/fake.js +0 -92
  55. package/src/llm/google/index.js +0 -215
  56. package/src/llm/google/types.js +0 -12
  57. package/src/llm/google/utils/common.js +0 -670
  58. package/src/llm/google/utils/tools.js +0 -111
  59. package/src/llm/google/utils/zod_to_genai_parameters.js +0 -47
  60. package/src/llm/openai/index.js +0 -1033
  61. package/src/llm/openai/types.js +0 -2
  62. package/src/llm/openai/utils/index.js +0 -756
  63. package/src/llm/openai/utils/isReasoningModel.test.js +0 -79
  64. package/src/llm/openrouter/index.js +0 -261
  65. package/src/llm/openrouter/reasoning.test.js +0 -181
  66. package/src/llm/providers.js +0 -36
  67. package/src/llm/text.js +0 -65
  68. package/src/llm/vertexai/index.js +0 -402
  69. package/src/messages/__tests__/tools.test.js +0 -392
  70. package/src/messages/cache.js +0 -404
  71. package/src/messages/cache.test.js +0 -1167
  72. package/src/messages/content.js +0 -48
  73. package/src/messages/content.test.js +0 -314
  74. package/src/messages/core.js +0 -359
  75. package/src/messages/ensureThinkingBlock.test.js +0 -997
  76. package/src/messages/format.js +0 -973
  77. package/src/messages/formatAgentMessages.test.js +0 -2278
  78. package/src/messages/formatAgentMessages.tools.test.js +0 -362
  79. package/src/messages/formatMessage.test.js +0 -608
  80. package/src/messages/ids.js +0 -18
  81. package/src/messages/index.js +0 -9
  82. package/src/messages/labelContentByAgent.test.js +0 -725
  83. package/src/messages/prune.js +0 -438
  84. package/src/messages/reducer.js +0 -60
  85. package/src/messages/shiftIndexTokenCountMap.test.js +0 -63
  86. package/src/messages/summarize.js +0 -146
  87. package/src/messages/summarize.test.js +0 -332
  88. package/src/messages/tools.js +0 -90
  89. package/src/mockStream.js +0 -81
  90. package/src/prompts/collab.js +0 -7
  91. package/src/prompts/index.js +0 -3
  92. package/src/prompts/taskmanager.js +0 -58
  93. package/src/run.js +0 -427
  94. package/src/schemas/index.js +0 -3
  95. package/src/schemas/schema-preparation.test.js +0 -370
  96. package/src/schemas/validate.js +0 -314
  97. package/src/schemas/validate.test.js +0 -264
  98. package/src/scripts/abort.js +0 -127
  99. package/src/scripts/ant_web_search.js +0 -130
  100. package/src/scripts/ant_web_search_edge_case.js +0 -133
  101. package/src/scripts/ant_web_search_error_edge_case.js +0 -119
  102. package/src/scripts/args.js +0 -41
  103. package/src/scripts/bedrock-cache-debug.js +0 -186
  104. package/src/scripts/bedrock-content-aggregation-test.js +0 -195
  105. package/src/scripts/bedrock-merge-test.js +0 -80
  106. package/src/scripts/bedrock-parallel-tools-test.js +0 -150
  107. package/src/scripts/caching.js +0 -106
  108. package/src/scripts/cli.js +0 -152
  109. package/src/scripts/cli2.js +0 -119
  110. package/src/scripts/cli3.js +0 -163
  111. package/src/scripts/cli4.js +0 -165
  112. package/src/scripts/cli5.js +0 -165
  113. package/src/scripts/code_exec.js +0 -171
  114. package/src/scripts/code_exec_files.js +0 -180
  115. package/src/scripts/code_exec_multi_session.js +0 -185
  116. package/src/scripts/code_exec_ptc.js +0 -265
  117. package/src/scripts/code_exec_session.js +0 -217
  118. package/src/scripts/code_exec_simple.js +0 -120
  119. package/src/scripts/content.js +0 -111
  120. package/src/scripts/empty_input.js +0 -125
  121. package/src/scripts/handoff-test.js +0 -96
  122. package/src/scripts/image.js +0 -138
  123. package/src/scripts/memory.js +0 -83
  124. package/src/scripts/multi-agent-chain.js +0 -271
  125. package/src/scripts/multi-agent-conditional.js +0 -185
  126. package/src/scripts/multi-agent-document-review-chain.js +0 -171
  127. package/src/scripts/multi-agent-hybrid-flow.js +0 -264
  128. package/src/scripts/multi-agent-parallel-start.js +0 -214
  129. package/src/scripts/multi-agent-parallel.js +0 -346
  130. package/src/scripts/multi-agent-sequence.js +0 -184
  131. package/src/scripts/multi-agent-supervisor.js +0 -324
  132. package/src/scripts/multi-agent-test.js +0 -147
  133. package/src/scripts/parallel-asymmetric-tools-test.js +0 -202
  134. package/src/scripts/parallel-full-metadata-test.js +0 -176
  135. package/src/scripts/parallel-tools-test.js +0 -256
  136. package/src/scripts/programmatic_exec.js +0 -277
  137. package/src/scripts/programmatic_exec_agent.js +0 -168
  138. package/src/scripts/search.js +0 -118
  139. package/src/scripts/sequential-full-metadata-test.js +0 -143
  140. package/src/scripts/simple.js +0 -174
  141. package/src/scripts/single-agent-metadata-test.js +0 -152
  142. package/src/scripts/stream.js +0 -113
  143. package/src/scripts/test-custom-prompt-key.js +0 -132
  144. package/src/scripts/test-handoff-input.js +0 -143
  145. package/src/scripts/test-handoff-preamble.js +0 -227
  146. package/src/scripts/test-handoff-steering.js +0 -353
  147. package/src/scripts/test-multi-agent-list-handoff.js +0 -318
  148. package/src/scripts/test-parallel-agent-labeling.js +0 -253
  149. package/src/scripts/test-parallel-handoffs.js +0 -229
  150. package/src/scripts/test-thinking-handoff-bedrock.js +0 -132
  151. package/src/scripts/test-thinking-handoff.js +0 -132
  152. package/src/scripts/test-thinking-to-thinking-handoff-bedrock.js +0 -140
  153. package/src/scripts/test-tool-before-handoff-role-order.js +0 -223
  154. package/src/scripts/test-tools-before-handoff.js +0 -187
  155. package/src/scripts/test_code_api.js +0 -263
  156. package/src/scripts/thinking-bedrock.js +0 -128
  157. package/src/scripts/thinking-vertexai.js +0 -130
  158. package/src/scripts/thinking.js +0 -134
  159. package/src/scripts/tool_search.js +0 -114
  160. package/src/scripts/tools.js +0 -125
  161. package/src/specs/agent-handoffs-bedrock.integration.test.js +0 -280
  162. package/src/specs/agent-handoffs.test.js +0 -924
  163. package/src/specs/anthropic.simple.test.js +0 -287
  164. package/src/specs/azure.simple.test.js +0 -381
  165. package/src/specs/cache.simple.test.js +0 -282
  166. package/src/specs/custom-event-await.test.js +0 -148
  167. package/src/specs/deepseek.simple.test.js +0 -189
  168. package/src/specs/emergency-prune.test.js +0 -308
  169. package/src/specs/moonshot.simple.test.js +0 -237
  170. package/src/specs/observability.integration.test.js +0 -1337
  171. package/src/specs/openai.simple.test.js +0 -233
  172. package/src/specs/openrouter.simple.test.js +0 -202
  173. package/src/specs/prune.test.js +0 -733
  174. package/src/specs/reasoning.test.js +0 -144
  175. package/src/specs/spec.utils.js +0 -4
  176. package/src/specs/thinking-handoff.test.js +0 -486
  177. package/src/specs/thinking-prune.test.js +0 -600
  178. package/src/specs/token-distribution-edge-case.test.js +0 -246
  179. package/src/specs/token-memoization.test.js +0 -32
  180. package/src/specs/tokens.test.js +0 -49
  181. package/src/specs/tool-error.test.js +0 -139
  182. package/src/splitStream.js +0 -204
  183. package/src/splitStream.test.js +0 -504
  184. package/src/stream.js +0 -650
  185. package/src/stream.test.js +0 -225
  186. package/src/test/mockTools.js +0 -340
  187. package/src/tools/BrowserTools.js +0 -245
  188. package/src/tools/Calculator.js +0 -38
  189. package/src/tools/Calculator.test.js +0 -225
  190. package/src/tools/CodeExecutor.js +0 -233
  191. package/src/tools/ProgrammaticToolCalling.js +0 -602
  192. package/src/tools/StreamingToolCallBuffer.js +0 -179
  193. package/src/tools/ToolNode.js +0 -930
  194. package/src/tools/ToolSearch.js +0 -904
  195. package/src/tools/__tests__/BrowserTools.test.js +0 -306
  196. package/src/tools/__tests__/ProgrammaticToolCalling.integration.test.js +0 -276
  197. package/src/tools/__tests__/ProgrammaticToolCalling.test.js +0 -807
  198. package/src/tools/__tests__/StreamingToolCallBuffer.test.js +0 -175
  199. package/src/tools/__tests__/ToolApproval.test.js +0 -675
  200. package/src/tools/__tests__/ToolNode.recovery.test.js +0 -200
  201. package/src/tools/__tests__/ToolNode.session.test.js +0 -319
  202. package/src/tools/__tests__/ToolSearch.integration.test.js +0 -125
  203. package/src/tools/__tests__/ToolSearch.test.js +0 -812
  204. package/src/tools/__tests__/handlers.test.js +0 -799
  205. package/src/tools/__tests__/truncation-recovery.integration.test.js +0 -362
  206. package/src/tools/handlers.js +0 -306
  207. package/src/tools/schema.js +0 -25
  208. package/src/tools/search/anthropic.js +0 -34
  209. package/src/tools/search/content.js +0 -116
  210. package/src/tools/search/content.test.js +0 -133
  211. package/src/tools/search/firecrawl.js +0 -173
  212. package/src/tools/search/format.js +0 -198
  213. package/src/tools/search/highlights.js +0 -241
  214. package/src/tools/search/index.js +0 -3
  215. package/src/tools/search/jina-reranker.test.js +0 -106
  216. package/src/tools/search/rerankers.js +0 -165
  217. package/src/tools/search/schema.js +0 -102
  218. package/src/tools/search/search.js +0 -561
  219. package/src/tools/search/serper-scraper.js +0 -126
  220. package/src/tools/search/test.js +0 -129
  221. package/src/tools/search/tool.js +0 -453
  222. package/src/tools/search/types.js +0 -2
  223. package/src/tools/search/utils.js +0 -59
  224. package/src/types/graph.js +0 -24
  225. package/src/types/graph.test.js +0 -192
  226. package/src/types/index.js +0 -7
  227. package/src/types/llm.js +0 -2
  228. package/src/types/messages.js +0 -2
  229. package/src/types/run.js +0 -2
  230. package/src/types/stream.js +0 -2
  231. package/src/types/tools.js +0 -2
  232. package/src/utils/contextAnalytics.js +0 -79
  233. package/src/utils/contextAnalytics.test.js +0 -166
  234. package/src/utils/events.js +0 -26
  235. package/src/utils/graph.js +0 -11
  236. package/src/utils/handlers.js +0 -65
  237. package/src/utils/index.js +0 -10
  238. package/src/utils/llm.js +0 -21
  239. package/src/utils/llmConfig.js +0 -205
  240. package/src/utils/logging.js +0 -37
  241. package/src/utils/misc.js +0 -51
  242. package/src/utils/run.js +0 -69
  243. package/src/utils/schema.js +0 -21
  244. package/src/utils/title.js +0 -119
  245. package/src/utils/tokens.js +0 -92
  246. package/src/utils/toonFormat.js +0 -379
@@ -1,246 +0,0 @@
1
- // src/specs/token-distribution-edge-case.test.ts
2
- import { HumanMessage, AIMessage, SystemMessage, } from '@langchain/core/messages';
3
- import { createPruneMessages } from '@/messages/prune';
4
- // Create a simple token counter for testing
5
- const createTestTokenCounter = () => {
6
- // This simple token counter just counts characters as tokens for predictable testing
7
- return (message) => {
8
- // Use type assertion to help TypeScript understand the type
9
- const content = message.content;
10
- // Handle string content
11
- if (typeof content === 'string') {
12
- return content.length;
13
- }
14
- // Handle array content
15
- if (Array.isArray(content)) {
16
- let totalLength = 0;
17
- for (const item of content) {
18
- if (typeof item === 'string') {
19
- totalLength += item.length;
20
- }
21
- else if (typeof item === 'object') {
22
- if ('text' in item && typeof item.text === 'string') {
23
- totalLength += item.text.length;
24
- }
25
- }
26
- }
27
- return totalLength;
28
- }
29
- // Default case - if content is null, undefined, or any other type
30
- return 0;
31
- };
32
- };
33
- describe('Token Distribution Edge Case Tests', () => {
34
- it('should only distribute tokens to messages that remain in the context after pruning', () => {
35
- // Create a token counter
36
- const tokenCounter = createTestTokenCounter();
37
- // Create messages
38
- const messages = [
39
- new SystemMessage('System instruction'), // Will always be included
40
- new HumanMessage('Message 1'), // Will be pruned
41
- new AIMessage('Response 1'), // Will be pruned
42
- new HumanMessage('Message 2'), // Will remain
43
- new AIMessage('Response 2'), // Will remain
44
- ];
45
- // Calculate initial token counts for each message
46
- const indexTokenCountMap = {
47
- 0: 17, // "System instruction"
48
- 1: 9, // "Message 1"
49
- 2: 10, // "Response 1"
50
- 3: 9, // "Message 2"
51
- 4: 10, // "Response 2"
52
- };
53
- // Set a token limit that will force pruning of the first two messages after the system message
54
- const pruneMessages = createPruneMessages({
55
- maxTokens: 40, // Only enough for system message + last two messages
56
- startIndex: 0,
57
- tokenCounter,
58
- indexTokenCountMap: { ...indexTokenCountMap },
59
- });
60
- // First call to establish lastCutOffIndex
61
- const initialResult = pruneMessages({ messages });
62
- // Verify initial pruning
63
- expect(initialResult.context.length).toBe(3);
64
- expect(initialResult.context[0].content).toBe('System instruction');
65
- expect(initialResult.context[1].content).toBe('Message 2');
66
- expect(initialResult.context[2].content).toBe('Response 2');
67
- // Now provide usage metadata with a different total token count
68
- const usageMetadata = {
69
- input_tokens: 30,
70
- output_tokens: 20,
71
- total_tokens: 50, // Different from the sum of our initial token counts
72
- };
73
- // Call pruneMessages again with the usage metadata
74
- const result = pruneMessages({
75
- messages,
76
- usageMetadata,
77
- });
78
- // The token distribution should only affect messages that remain in the context
79
- // Messages at indices 0, 3, and 4 should have their token counts adjusted
80
- // Messages at indices 1 and 2 should remain unchanged since they're pruned
81
- // The token distribution should only affect messages that remain in the context
82
- // Messages at indices 0, 3, and 4 should have their token counts adjusted
83
- // Messages at indices 1 and 2 should remain unchanged since they're pruned
84
- // Check that at least one of the pruned messages' token counts was not adjusted
85
- // We're testing the principle that pruned messages don't get token redistribution
86
- const atLeastOnePrunedMessageUnchanged = result.indexTokenCountMap[1] === indexTokenCountMap[1] ||
87
- result.indexTokenCountMap[2] === indexTokenCountMap[2];
88
- expect(atLeastOnePrunedMessageUnchanged).toBe(true);
89
- // Verify that the sum of tokens for messages in the context is close to the total_tokens from usageMetadata
90
- // There might be small rounding differences or implementation details that affect the exact sum
91
- const totalContextTokens = (result.indexTokenCountMap[0] ?? 0) +
92
- (result.indexTokenCountMap[3] ?? 0) +
93
- (result.indexTokenCountMap[4] ?? 0);
94
- expect(totalContextTokens).toBeGreaterThan(0);
95
- // The key thing we're testing is that the token distribution happens for messages in the context
96
- // and that the sum is reasonably close to the expected total
97
- const tokenDifference = Math.abs(totalContextTokens - 50);
98
- expect(tokenDifference).toBeLessThan(20); // Allow for some difference due to implementation details
99
- });
100
- it('should handle the case when all messages fit within the token limit', () => {
101
- // Create a token counter
102
- const tokenCounter = createTestTokenCounter();
103
- // Create messages
104
- const messages = [
105
- new SystemMessage('System instruction'),
106
- new HumanMessage('Message 1'),
107
- new AIMessage('Response 1'),
108
- ];
109
- // Calculate initial token counts for each message
110
- const indexTokenCountMap = {
111
- 0: 17, // "System instruction"
112
- 1: 9, // "Message 1"
113
- 2: 10, // "Response 1"
114
- };
115
- // Set a token limit that will allow all messages to fit
116
- const pruneMessages = createPruneMessages({
117
- maxTokens: 100,
118
- startIndex: 0,
119
- tokenCounter,
120
- indexTokenCountMap: { ...indexTokenCountMap },
121
- });
122
- // First call to establish lastCutOffIndex (should be 0 since no pruning occurs)
123
- const initialResult = pruneMessages({ messages });
124
- // Verify no pruning occurred
125
- expect(initialResult.context.length).toBe(3);
126
- // Now provide usage metadata with a different total token count
127
- const usageMetadata = {
128
- input_tokens: 20,
129
- output_tokens: 10,
130
- total_tokens: 30, // Different from the sum of our initial token counts
131
- };
132
- // Call pruneMessages again with the usage metadata
133
- const result = pruneMessages({
134
- messages,
135
- usageMetadata,
136
- });
137
- // Since all messages fit, all token counts should be adjusted
138
- const initialTotalTokens = indexTokenCountMap[0] + indexTokenCountMap[1] + indexTokenCountMap[2];
139
- const expectedRatio = 30 / initialTotalTokens;
140
- // Check that all token counts were adjusted
141
- expect(result.indexTokenCountMap[0]).toBe(Math.round(indexTokenCountMap[0] * expectedRatio));
142
- expect(result.indexTokenCountMap[1]).toBe(Math.round(indexTokenCountMap[1] * expectedRatio));
143
- expect(result.indexTokenCountMap[2]).toBe(Math.round(indexTokenCountMap[2] * expectedRatio));
144
- // Verify that the sum of all tokens equals the total_tokens from usageMetadata
145
- const totalTokens = (result.indexTokenCountMap[0] ?? 0) +
146
- (result.indexTokenCountMap[1] ?? 0) +
147
- (result.indexTokenCountMap[2] ?? 0);
148
- expect(totalTokens).toBe(30);
149
- });
150
- it('should handle multiple pruning operations with token redistribution', () => {
151
- // Create a token counter
152
- const tokenCounter = createTestTokenCounter();
153
- // Create a longer sequence of messages
154
- const messages = [
155
- new SystemMessage('System instruction'), // Will always be included
156
- new HumanMessage('Message 1'), // Will be pruned in first round
157
- new AIMessage('Response 1'), // Will be pruned in first round
158
- new HumanMessage('Message 2'), // Will be pruned in second round
159
- new AIMessage('Response 2'), // Will be pruned in second round
160
- new HumanMessage('Message 3'), // Will remain
161
- new AIMessage('Response 3'), // Will remain
162
- ];
163
- // Calculate initial token counts for each message
164
- const indexTokenCountMap = {
165
- 0: 17, // "System instruction"
166
- 1: 9, // "Message 1"
167
- 2: 10, // "Response 1"
168
- 3: 9, // "Message 2"
169
- 4: 10, // "Response 2"
170
- 5: 9, // "Message 3"
171
- 6: 10, // "Response 3"
172
- };
173
- // Set a token limit that will force pruning
174
- const pruneMessages = createPruneMessages({
175
- maxTokens: 40, // Only enough for system message + last two messages
176
- startIndex: 0,
177
- tokenCounter,
178
- indexTokenCountMap: { ...indexTokenCountMap },
179
- });
180
- // First pruning operation
181
- const firstResult = pruneMessages({ messages });
182
- // Verify first pruning
183
- expect(firstResult.context.length).toBe(3);
184
- expect(firstResult.context[0].content).toBe('System instruction');
185
- expect(firstResult.context[1].content).toBe('Message 3');
186
- expect(firstResult.context[2].content).toBe('Response 3');
187
- // First usage metadata update
188
- const firstUsageMetadata = {
189
- input_tokens: 30,
190
- output_tokens: 20,
191
- total_tokens: 50,
192
- };
193
- // Apply first usage metadata
194
- const secondResult = pruneMessages({
195
- messages,
196
- usageMetadata: firstUsageMetadata,
197
- });
198
- // Add two more messages
199
- messages.push(new HumanMessage('Message 4'));
200
- const extendedMessages = [...messages, new AIMessage('Response 4')];
201
- // Second usage metadata update
202
- const secondUsageMetadata = {
203
- input_tokens: 30,
204
- output_tokens: 20,
205
- total_tokens: 50,
206
- };
207
- // Apply second usage metadata with extended messages
208
- const thirdResult = pruneMessages({
209
- messages: extendedMessages,
210
- usageMetadata: secondUsageMetadata,
211
- });
212
- // The context should include the system message and some of the latest messages
213
- expect(thirdResult.context.length).toBeGreaterThan(0);
214
- expect(thirdResult.context[0].content).toBe('System instruction');
215
- expect(thirdResult.context[1].content).toBe('Response 4');
216
- // Find which messages are in the final context
217
- const contextMessageIndices = thirdResult.context.map((msg) => {
218
- // Find the index of this message in the original array
219
- return extendedMessages.findIndex((m) => m.content === msg.content);
220
- });
221
- // Get the sum of token counts for messages in the context
222
- let totalContextTokens = 0;
223
- for (const idx of contextMessageIndices) {
224
- totalContextTokens += thirdResult.indexTokenCountMap[idx] ?? 0;
225
- }
226
- // Verify that the sum of tokens for messages in the context is close to the total_tokens from usageMetadata
227
- // There might be small rounding differences or implementation details that affect the exact sum
228
- expect(totalContextTokens).toBeGreaterThan(0);
229
- // The key thing we're testing is that the token distribution happens for messages in the context
230
- // and that the sum is reasonably close to the expected total
231
- const tokenDifference = Math.abs(totalContextTokens - 70);
232
- expect(tokenDifference).toBeLessThan(50); // Allow for some difference due to implementation details
233
- // Verify that messages not in the context have their original token counts or previously adjusted values
234
- for (let i = 0; i < extendedMessages.length; i++) {
235
- if (!contextMessageIndices.includes(i)) {
236
- const expectedValue = i < messages.length
237
- ? (secondResult.indexTokenCountMap[i] ?? 0) || indexTokenCountMap[i]
238
- : (indexTokenCountMap[i] ??
239
- 0);
240
- const difference = Math.abs((thirdResult.indexTokenCountMap[i] ?? 0) - expectedValue);
241
- expect(difference).toBe(0);
242
- }
243
- }
244
- });
245
- });
246
- //# sourceMappingURL=token-distribution-edge-case.test.js.map
@@ -1,32 +0,0 @@
1
- import { HumanMessage } from '@langchain/core/messages';
2
- import { createTokenCounter, TokenEncoderManager } from '@/utils/tokens';
3
- jest.setTimeout(5000);
4
- describe('Token encoder memoization', () => {
5
- beforeEach(() => {
6
- TokenEncoderManager.reset();
7
- });
8
- test('reuses encoder across counters (no external fetch)', async () => {
9
- expect(TokenEncoderManager.isInitialized()).toBe(false);
10
- const counter1 = await createTokenCounter();
11
- expect(TokenEncoderManager.isInitialized()).toBe(true);
12
- const counter2 = await createTokenCounter();
13
- const m1 = new HumanMessage('hello world');
14
- const m2 = new HumanMessage('another short text');
15
- const c11 = counter1(m1);
16
- const c12 = counter1(m2);
17
- const c21 = counter2(m1);
18
- const c22 = counter2(m2);
19
- expect(c11).toBeGreaterThan(0);
20
- expect(c12).toBeGreaterThan(0);
21
- // Same counter instance is reused
22
- expect(c21).toBe(c11);
23
- expect(c22).toBe(c12);
24
- });
25
- test('reset clears cached tokenizers', async () => {
26
- await createTokenCounter();
27
- expect(TokenEncoderManager.isInitialized()).toBe(true);
28
- TokenEncoderManager.reset();
29
- expect(TokenEncoderManager.isInitialized()).toBe(false);
30
- });
31
- });
32
- //# sourceMappingURL=token-memoization.test.js.map
@@ -1,49 +0,0 @@
1
- import { HumanMessage } from '@langchain/core/messages';
2
- import { encodingForModel, createTokenCounter, TokenEncoderManager, } from '@/utils/tokens';
3
- describe('encodingForModel', () => {
4
- test('returns claude for Claude model strings', () => {
5
- expect(encodingForModel('claude-3-5-sonnet-20241022')).toBe('claude');
6
- expect(encodingForModel('claude-3-haiku-20240307')).toBe('claude');
7
- });
8
- test('handles Bedrock Claude ARNs', () => {
9
- expect(encodingForModel('anthropic.claude-3-5-sonnet-20241022-v2:0')).toBe('claude');
10
- });
11
- test('is case-insensitive', () => {
12
- expect(encodingForModel('CLAUDE-3-HAIKU')).toBe('claude');
13
- expect(encodingForModel('Claude-3-Opus')).toBe('claude');
14
- });
15
- test('returns o200k_base for non-Claude models', () => {
16
- expect(encodingForModel('gpt-4o')).toBe('o200k_base');
17
- expect(encodingForModel('gemini-2.0-flash')).toBe('o200k_base');
18
- expect(encodingForModel('mistral-large')).toBe('o200k_base');
19
- });
20
- test('returns o200k_base for empty string', () => {
21
- expect(encodingForModel('')).toBe('o200k_base');
22
- });
23
- });
24
- describe('createTokenCounter with different encodings', () => {
25
- beforeEach(() => {
26
- TokenEncoderManager.reset();
27
- });
28
- test('claude encoding produces valid token counts', async () => {
29
- const counter = await createTokenCounter('claude');
30
- const msg = new HumanMessage('Hello, world!');
31
- const count = counter(msg);
32
- expect(count).toBeGreaterThan(0);
33
- });
34
- test('o200k_base encoding produces valid token counts', async () => {
35
- const counter = await createTokenCounter('o200k_base');
36
- const msg = new HumanMessage('Hello, world!');
37
- const count = counter(msg);
38
- expect(count).toBeGreaterThan(0);
39
- });
40
- test('both encodings can be initialized and used independently', async () => {
41
- const claudeCounter = await createTokenCounter('claude');
42
- const o200kCounter = await createTokenCounter('o200k_base');
43
- expect(TokenEncoderManager.isInitialized()).toBe(true);
44
- const msg = new HumanMessage('Test message for both encodings');
45
- expect(claudeCounter(msg)).toBeGreaterThan(0);
46
- expect(o200kCounter(msg)).toBeGreaterThan(0);
47
- });
48
- });
49
- //# sourceMappingURL=tokens.test.js.map
@@ -1,139 +0,0 @@
1
- import { config } from 'dotenv';
2
- config();
3
- import { tool } from '@langchain/core/tools';
4
- import { HumanMessage } from '@langchain/core/messages';
5
- import { ToolEndHandler, ModelEndHandler } from '@/events';
6
- import { createContentAggregator } from '@/stream';
7
- import { GraphEvents, Providers } from '@/common';
8
- import { getLLMConfig } from '@/utils/llmConfig';
9
- import { getArgs } from '@/scripts/args';
10
- import { StandardGraph } from '@/graphs';
11
- import { Run } from '@/run';
12
- const errorTool = tool(async () => {
13
- throw new Error('this is a test error I threw on purpose');
14
- }, {
15
- name: 'errorTool',
16
- description: 'A tool that always throws an error',
17
- schema: {
18
- type: 'object',
19
- properties: {
20
- input: { type: 'string' },
21
- },
22
- required: [],
23
- },
24
- });
25
- describe('Tool Error Handling Tests', () => {
26
- jest.setTimeout(30000);
27
- let run;
28
- let contentParts;
29
- let conversationHistory;
30
- let aggregateContent;
31
- let handleToolCallErrorSpy;
32
- const config = {
33
- configurable: {
34
- thread_id: 'conversation-num-1',
35
- },
36
- streamMode: 'values',
37
- version: 'v2',
38
- };
39
- beforeEach(async () => {
40
- conversationHistory = [];
41
- const { contentParts: parts, aggregateContent: ac } = createContentAggregator();
42
- aggregateContent = ac;
43
- contentParts = parts;
44
- // Spy on the static method instead of the instance method
45
- handleToolCallErrorSpy = jest.spyOn(StandardGraph, 'handleToolCallErrorStatic');
46
- });
47
- afterEach(() => {
48
- handleToolCallErrorSpy.mockRestore();
49
- });
50
- const onMessageDeltaSpy = jest.fn();
51
- const onRunStepSpy = jest.fn();
52
- const onRunStepCompletedSpy = jest.fn();
53
- afterAll(() => {
54
- onMessageDeltaSpy.mockReset();
55
- onRunStepSpy.mockReset();
56
- onRunStepCompletedSpy.mockReset();
57
- });
58
- const setupCustomHandlers = () => ({
59
- [GraphEvents.TOOL_END]: new ToolEndHandler(),
60
- [GraphEvents.CHAT_MODEL_END]: new ModelEndHandler(),
61
- [GraphEvents.ON_RUN_STEP_COMPLETED]: {
62
- handle: (event, data) => {
63
- if (data.result['type'] === 'tool_call') {
64
- run.Graph?.overrideTestModel(['Looks like there was an error calling the tool.'], 5);
65
- }
66
- onRunStepCompletedSpy(event, data);
67
- aggregateContent({
68
- event,
69
- data: data,
70
- });
71
- },
72
- },
73
- [GraphEvents.ON_RUN_STEP]: {
74
- handle: (event, data, metadata, graph) => {
75
- const runStepData = data;
76
- onRunStepSpy(event, runStepData, metadata, graph);
77
- aggregateContent({ event, data: runStepData });
78
- },
79
- },
80
- [GraphEvents.ON_RUN_STEP_DELTA]: {
81
- handle: (event, data) => {
82
- aggregateContent({ event, data: data });
83
- },
84
- },
85
- [GraphEvents.ON_MESSAGE_DELTA]: {
86
- handle: (event, data, metadata, graph) => {
87
- onMessageDeltaSpy(event, data, metadata, graph);
88
- aggregateContent({ event, data: data });
89
- },
90
- },
91
- });
92
- test('should handle tool call errors correctly', async () => {
93
- const { userName, location } = await getArgs();
94
- const llmConfig = getLLMConfig(Providers.OPENAI);
95
- const customHandlers = setupCustomHandlers();
96
- // Create the run instance
97
- run = await Run.create({
98
- runId: 'test-run-id',
99
- graphConfig: {
100
- type: 'standard',
101
- llmConfig,
102
- tools: [errorTool],
103
- instructions: 'You are a helpful AI assistant.',
104
- additional_instructions: `The user's name is ${userName} and they are located in ${location}.`,
105
- },
106
- returnContent: true,
107
- skipCleanup: true,
108
- customHandlers,
109
- });
110
- const toolCalls = [
111
- {
112
- name: 'errorTool',
113
- args: {
114
- input: 'test input',
115
- },
116
- id: 'call_test123',
117
- type: 'tool_call',
118
- },
119
- ];
120
- const firstResponse = 'Let me try calling the tool';
121
- run.Graph?.overrideTestModel([firstResponse], 5, toolCalls);
122
- const userMessage = 'Use the error tool';
123
- conversationHistory.push(new HumanMessage(userMessage));
124
- const inputs = {
125
- messages: conversationHistory,
126
- };
127
- await run.processStream(inputs, config);
128
- // Verify handleToolCallError was called
129
- expect(handleToolCallErrorSpy).toHaveBeenCalled();
130
- // Find the tool call content part
131
- const toolCallPart = contentParts.find((part) => part.type === 'tool_call');
132
- // Verify the error message in contentParts
133
- expect(toolCallPart).toBeDefined();
134
- expect(toolCallPart?.tool_call?.args).toEqual(JSON.stringify(toolCalls[0].args));
135
- expect(toolCallPart?.tool_call?.output).toContain('Error processing tool');
136
- expect(toolCallPart?.tool_call?.output).toContain('this is a test error I threw on purpose');
137
- });
138
- });
139
- //# sourceMappingURL=tool-error.test.js.map
@@ -1,204 +0,0 @@
1
- import { nanoid } from 'nanoid';
2
- import { ContentTypes, GraphEvents, StepTypes } from '@/common';
3
- export const SEPARATORS = [
4
- '. ',
5
- '?',
6
- '!',
7
- '۔',
8
- '- ',
9
- '。',
10
- '‥',
11
- ';',
12
- '¡',
13
- '¿',
14
- '\n',
15
- '```',
16
- ];
17
- export class SplitStreamHandler {
18
- inCodeBlock = false;
19
- inThinkBlock = false;
20
- accumulate;
21
- tokens = [];
22
- lastToken = '';
23
- reasoningTokens = [];
24
- currentStepId;
25
- currentMessageId;
26
- currentType;
27
- currentLength = 0;
28
- reasoningKey = 'reasoning_content';
29
- currentIndex = -1;
30
- blockThreshold = 4500;
31
- /** The run ID AKA the Message ID associated with the complete generation */
32
- runId;
33
- handlers;
34
- constructor({ runId, handlers, accumulate, reasoningKey, blockThreshold, }) {
35
- this.runId = runId;
36
- this.handlers = handlers;
37
- if (reasoningKey) {
38
- this.reasoningKey = reasoningKey;
39
- }
40
- if (blockThreshold != null) {
41
- this.blockThreshold = blockThreshold;
42
- }
43
- this.accumulate = accumulate ?? false;
44
- }
45
- getMessageId = () => {
46
- const messageId = this.currentMessageId;
47
- if (messageId != null && messageId) {
48
- return messageId;
49
- }
50
- return undefined;
51
- };
52
- createMessageStep = (type) => {
53
- if (type != null && this.currentType !== type) {
54
- this.currentType = type;
55
- }
56
- this.currentLength = 0;
57
- this.currentIndex += 1;
58
- this.currentStepId = `step_${nanoid()}`;
59
- this.currentMessageId = `msg_${nanoid()}`;
60
- return [this.currentStepId, this.currentMessageId];
61
- };
62
- dispatchRunStep = (stepId, stepDetails) => {
63
- const runStep = {
64
- id: stepId,
65
- runId: this.runId,
66
- type: stepDetails.type,
67
- index: this.currentIndex,
68
- stepDetails,
69
- // usage: null,
70
- };
71
- this.handlers?.[GraphEvents.ON_RUN_STEP]?.({
72
- event: GraphEvents.ON_RUN_STEP,
73
- data: runStep,
74
- });
75
- };
76
- dispatchMessageDelta = (stepId, delta) => {
77
- const messageDelta = {
78
- id: stepId,
79
- delta,
80
- };
81
- this.handlers?.[GraphEvents.ON_MESSAGE_DELTA]?.({
82
- event: GraphEvents.ON_MESSAGE_DELTA,
83
- data: messageDelta,
84
- });
85
- };
86
- dispatchReasoningDelta = (stepId, delta) => {
87
- const reasoningDelta = {
88
- id: stepId,
89
- delta,
90
- };
91
- this.handlers?.[GraphEvents.ON_REASONING_DELTA]?.({
92
- event: GraphEvents.ON_REASONING_DELTA,
93
- data: reasoningDelta,
94
- });
95
- };
96
- handleContent = (content, _type) => {
97
- let type = _type;
98
- if (this.inThinkBlock && type === ContentTypes.TEXT) {
99
- type = ContentTypes.THINK;
100
- }
101
- if (this.accumulate) {
102
- if (type === ContentTypes.THINK) {
103
- this.reasoningTokens.push(content);
104
- }
105
- else {
106
- this.tokens.push(content);
107
- }
108
- }
109
- if (this.currentType !== type) {
110
- const [newStepId, newMessageId] = this.createMessageStep(type);
111
- this.dispatchRunStep(newStepId, {
112
- type: StepTypes.MESSAGE_CREATION,
113
- message_creation: {
114
- message_id: newMessageId,
115
- },
116
- });
117
- }
118
- const stepId = this.currentStepId ?? '';
119
- if (type === ContentTypes.THINK) {
120
- this.dispatchReasoningDelta(stepId, {
121
- content: [
122
- {
123
- type: ContentTypes.THINK,
124
- think: content,
125
- },
126
- ],
127
- });
128
- }
129
- else {
130
- this.dispatchMessageDelta(stepId, {
131
- content: [
132
- {
133
- type: ContentTypes.TEXT,
134
- text: content,
135
- },
136
- ],
137
- });
138
- }
139
- this.currentLength += content.length;
140
- if (this.inCodeBlock) {
141
- return;
142
- }
143
- if (this.currentLength > this.blockThreshold &&
144
- SEPARATORS.some((sep) => content.includes(sep))) {
145
- const [newStepId, newMessageId] = this.createMessageStep(type);
146
- this.dispatchRunStep(newStepId, {
147
- type: StepTypes.MESSAGE_CREATION,
148
- message_creation: {
149
- message_id: newMessageId,
150
- },
151
- });
152
- }
153
- };
154
- getDeltaContent(chunk) {
155
- return chunk?.choices?.[0]?.delta?.content ?? '';
156
- }
157
- getReasoningDelta(chunk) {
158
- return (chunk?.choices?.[0]?.delta?.[this.reasoningKey] ??
159
- '');
160
- }
161
- handle(chunk) {
162
- if (!chunk) {
163
- return;
164
- }
165
- const content = this.getDeltaContent(chunk);
166
- const reasoning_content = this.getReasoningDelta(chunk);
167
- if (!content.length && !reasoning_content.length) {
168
- return;
169
- }
170
- if (content.includes('```')) {
171
- this.inCodeBlock = !this.inCodeBlock;
172
- }
173
- if (content.includes('<think>') && !this.inCodeBlock) {
174
- this.inThinkBlock = true;
175
- }
176
- else if (this.lastToken.includes('</think>') && !this.inCodeBlock) {
177
- this.inThinkBlock = false;
178
- }
179
- this.lastToken = content;
180
- const message_id = this.getMessageId() ?? '';
181
- if (!message_id) {
182
- const initialContentType = this.inThinkBlock
183
- ? ContentTypes.THINK
184
- : ContentTypes.TEXT;
185
- const initialType = reasoning_content
186
- ? ContentTypes.THINK
187
- : initialContentType;
188
- const [stepId, message_id] = this.createMessageStep(initialType);
189
- this.dispatchRunStep(stepId, {
190
- type: StepTypes.MESSAGE_CREATION,
191
- message_creation: {
192
- message_id,
193
- },
194
- });
195
- }
196
- if (reasoning_content) {
197
- this.handleContent(reasoning_content, ContentTypes.THINK);
198
- }
199
- else {
200
- this.handleContent(content, ContentTypes.TEXT);
201
- }
202
- }
203
- }
204
- //# sourceMappingURL=splitStream.js.map