@trigger.dev/sdk 4.8.5 → 4.8.6

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 (187) hide show
  1. package/dist/commonjs/v3/ai.d.ts +29 -5
  2. package/dist/commonjs/v3/ai.js +160 -114
  3. package/dist/commonjs/v3/ai.js.map +1 -1
  4. package/dist/commonjs/v3/auth.d.ts +4 -4
  5. package/dist/commonjs/v3/auth.js +21 -75
  6. package/dist/commonjs/v3/auth.js.map +1 -1
  7. package/dist/commonjs/v3/chat-client.d.ts +1 -1
  8. package/dist/commonjs/v3/chat.d.ts +1 -1
  9. package/dist/commonjs/v3/prompt.js +6 -8
  10. package/dist/commonjs/v3/prompt.js.map +1 -1
  11. package/dist/commonjs/v3/runs.d.ts +5 -1
  12. package/dist/commonjs/v3/runs.js.map +1 -1
  13. package/dist/commonjs/v3/sessions.js +10 -7
  14. package/dist/commonjs/v3/sessions.js.map +1 -1
  15. package/dist/commonjs/v3/shared.d.ts +2 -2
  16. package/dist/commonjs/v3/shared.js +16 -15
  17. package/dist/commonjs/v3/shared.js.map +1 -1
  18. package/dist/commonjs/v3/steeringContext.d.ts +35 -10
  19. package/dist/commonjs/v3/steeringContext.js +35 -12
  20. package/dist/commonjs/v3/steeringContext.js.map +1 -1
  21. package/dist/commonjs/v3/streams.js +10 -8
  22. package/dist/commonjs/v3/streams.js.map +1 -1
  23. package/dist/commonjs/v3/tasks.d.ts +2 -2
  24. package/dist/commonjs/v3/tasks.js.map +1 -1
  25. package/dist/commonjs/v3/test/mock-chat-agent.js +24 -13
  26. package/dist/commonjs/v3/test/mock-chat-agent.js.map +1 -1
  27. package/dist/commonjs/v3/webhooks.d.ts +8 -2
  28. package/dist/commonjs/v3/webhooks.js +20 -11
  29. package/dist/commonjs/v3/webhooks.js.map +1 -1
  30. package/dist/commonjs/version.js +1 -1
  31. package/dist/esm/v3/ai.d.ts +29 -5
  32. package/dist/esm/v3/ai.js +162 -117
  33. package/dist/esm/v3/ai.js.map +1 -1
  34. package/dist/esm/v3/auth.d.ts +4 -4
  35. package/dist/esm/v3/auth.js +22 -76
  36. package/dist/esm/v3/auth.js.map +1 -1
  37. package/dist/esm/v3/chat-client.d.ts +1 -1
  38. package/dist/esm/v3/chat.d.ts +1 -1
  39. package/dist/esm/v3/prompt.js +6 -8
  40. package/dist/esm/v3/prompt.js.map +1 -1
  41. package/dist/esm/v3/runs.d.ts +5 -1
  42. package/dist/esm/v3/runs.js.map +1 -1
  43. package/dist/esm/v3/sessions.js +10 -7
  44. package/dist/esm/v3/sessions.js.map +1 -1
  45. package/dist/esm/v3/shared.d.ts +2 -2
  46. package/dist/esm/v3/shared.js +16 -15
  47. package/dist/esm/v3/shared.js.map +1 -1
  48. package/dist/esm/v3/steeringContext.d.ts +35 -10
  49. package/dist/esm/v3/steeringContext.js +34 -12
  50. package/dist/esm/v3/steeringContext.js.map +1 -1
  51. package/dist/esm/v3/streams.js +10 -8
  52. package/dist/esm/v3/streams.js.map +1 -1
  53. package/dist/esm/v3/tasks.d.ts +2 -2
  54. package/dist/esm/v3/tasks.js.map +1 -1
  55. package/dist/esm/v3/test/mock-chat-agent.js +25 -14
  56. package/dist/esm/v3/test/mock-chat-agent.js.map +1 -1
  57. package/dist/esm/v3/webhooks.d.ts +8 -2
  58. package/dist/esm/v3/webhooks.js +20 -11
  59. package/dist/esm/v3/webhooks.js.map +1 -1
  60. package/dist/esm/version.js +1 -1
  61. package/docs/ai/observability.mdx +25 -19
  62. package/docs/ai/prompts.mdx +25 -19
  63. package/docs/ai-chat/actions.mdx +1 -1
  64. package/docs/ai-chat/anatomy.mdx +1 -1
  65. package/docs/ai-chat/backend.mdx +24 -24
  66. package/docs/ai-chat/background-injection.mdx +5 -5
  67. package/docs/ai-chat/chat-local.mdx +2 -2
  68. package/docs/ai-chat/compaction.mdx +38 -15
  69. package/docs/ai-chat/custom-agents.mdx +4 -4
  70. package/docs/ai-chat/error-handling.mdx +4 -4
  71. package/docs/ai-chat/fast-starts.mdx +8 -7
  72. package/docs/ai-chat/frontend.mdx +6 -2
  73. package/docs/ai-chat/lifecycle-hooks.mdx +11 -11
  74. package/docs/ai-chat/migrating-from-a-route-handler.mdx +4 -4
  75. package/docs/ai-chat/migrating-from-hydrate-messages.mdx +1 -1
  76. package/docs/ai-chat/overview.mdx +1 -1
  77. package/docs/ai-chat/patterns/branching-conversations.mdx +1 -1
  78. package/docs/ai-chat/patterns/database-persistence.mdx +2 -2
  79. package/docs/ai-chat/patterns/human-in-the-loop.mdx +1 -1
  80. package/docs/ai-chat/patterns/native-compaction.mdx +29 -9
  81. package/docs/ai-chat/patterns/persistence-and-replay.mdx +1 -1
  82. package/docs/ai-chat/patterns/redacting-telemetry.mdx +2 -2
  83. package/docs/ai-chat/patterns/skills.mdx +2 -2
  84. package/docs/ai-chat/patterns/sub-agents.mdx +1 -1
  85. package/docs/ai-chat/patterns/tool-result-auditing.mdx +1 -1
  86. package/docs/ai-chat/patterns/trusted-edge-signals.mdx +2 -2
  87. package/docs/ai-chat/patterns/version-upgrades.mdx +2 -2
  88. package/docs/ai-chat/pending-messages.mdx +5 -5
  89. package/docs/ai-chat/prompt-caching.mdx +7 -7
  90. package/docs/ai-chat/quick-start.mdx +2 -2
  91. package/docs/ai-chat/testing.mdx +9 -9
  92. package/docs/ai-chat/tools.mdx +9 -5
  93. package/docs/ai-chat/transcript-storage.mdx +2 -2
  94. package/docs/ai-chat/types.mdx +10 -3
  95. package/docs/apikeys.mdx +2 -1
  96. package/docs/cli-login-commands.mdx +12 -0
  97. package/docs/concurrency.mdx +40 -0
  98. package/docs/database-connections.mdx +1 -1
  99. package/docs/deploy-environment-variables.mdx +84 -2
  100. package/docs/deployment/trusted-publishing.mdx +4 -0
  101. package/docs/errors-retrying.mdx +25 -18
  102. package/docs/github-actions.mdx +1 -1
  103. package/docs/github-integration.mdx +21 -0
  104. package/docs/how-it-works.mdx +4 -2
  105. package/docs/introduction.mdx +1 -1
  106. package/docs/mcp-introduction.mdx +7 -1
  107. package/docs/mcp-remote.mdx +90 -0
  108. package/docs/oidc/azure.mdx +81 -0
  109. package/docs/oidc/databricks.mdx +99 -0
  110. package/docs/oidc/doppler.mdx +60 -0
  111. package/docs/oidc/gcp.mdx +109 -0
  112. package/docs/oidc/hugging-face.mdx +90 -0
  113. package/docs/oidc/openrouter.mdx +91 -0
  114. package/docs/oidc/snowflake.mdx +91 -0
  115. package/docs/oidc/vault.mdx +71 -0
  116. package/docs/oidc.mdx +2 -0
  117. package/docs/runs/metadata.mdx +1 -1
  118. package/docs/self-hosting/docker.mdx +2 -0
  119. package/docs/self-hosting/env/webapp.mdx +4 -1
  120. package/docs/self-hosting/kubernetes.mdx +2 -0
  121. package/docs/tasks/schemaTask.mdx +26 -13
  122. package/docs/tasks/streams.mdx +2 -2
  123. package/docs/troubleshooting.mdx +7 -0
  124. package/docs/webhooks/channels.mdx +6 -6
  125. package/docs/webhooks/connect.mdx +9 -75
  126. package/docs/webhooks/custom-providers.mdx +87 -0
  127. package/docs/webhooks/deliveries.mdx +7 -7
  128. package/docs/webhooks/filters.mdx +2 -2
  129. package/docs/webhooks/how-it-works.mdx +58 -0
  130. package/docs/webhooks/human-in-the-loop.mdx +3 -3
  131. package/docs/webhooks/linear.mdx +5 -3
  132. package/docs/webhooks/overview.mdx +20 -71
  133. package/docs/webhooks/own-handler.mdx +62 -0
  134. package/docs/webhooks/pricing-and-limits.mdx +13 -0
  135. package/docs/webhooks/providers/attio.mdx +38 -0
  136. package/docs/webhooks/providers/autumn.mdx +40 -0
  137. package/docs/webhooks/providers/cal-com.mdx +41 -0
  138. package/docs/webhooks/providers/calendly.mdx +41 -0
  139. package/docs/webhooks/providers/clerk.mdx +40 -0
  140. package/docs/webhooks/providers/discord.mdx +45 -0
  141. package/docs/webhooks/providers/docusign.mdx +41 -0
  142. package/docs/webhooks/providers/elevenlabs.mdx +41 -0
  143. package/docs/webhooks/providers/firecrawl.mdx +54 -0
  144. package/docs/webhooks/providers/github.mdx +46 -0
  145. package/docs/webhooks/providers/gitlab.mdx +41 -0
  146. package/docs/webhooks/providers/granola.mdx +44 -0
  147. package/docs/webhooks/providers/hookdeck.mdx +98 -0
  148. package/docs/webhooks/providers/infisical.mdx +41 -0
  149. package/docs/webhooks/providers/jira.mdx +41 -0
  150. package/docs/webhooks/providers/langfuse.mdx +40 -0
  151. package/docs/webhooks/providers/linear.mdx +45 -0
  152. package/docs/webhooks/providers/loops.mdx +40 -0
  153. package/docs/webhooks/providers/metronome.mdx +42 -0
  154. package/docs/webhooks/providers/mux.mdx +40 -0
  155. package/docs/webhooks/providers/openai.mdx +40 -0
  156. package/docs/webhooks/providers/pagerduty.mdx +40 -0
  157. package/docs/webhooks/providers/plain.mdx +41 -0
  158. package/docs/webhooks/providers/planetscale.mdx +40 -0
  159. package/docs/webhooks/providers/recall-ai.mdx +54 -0
  160. package/docs/webhooks/providers/replicate.mdx +41 -0
  161. package/docs/webhooks/providers/resend.mdx +44 -0
  162. package/docs/webhooks/providers/retell.mdx +41 -0
  163. package/docs/webhooks/providers/sentry.mdx +41 -0
  164. package/docs/webhooks/providers/shopify.mdx +49 -0
  165. package/docs/webhooks/providers/slack.mdx +48 -0
  166. package/docs/webhooks/providers/square.mdx +40 -0
  167. package/docs/webhooks/providers/stripe.mdx +86 -0
  168. package/docs/webhooks/providers/supabase.mdx +41 -0
  169. package/docs/webhooks/providers/tailscale.mdx +43 -0
  170. package/docs/webhooks/providers/typeform.mdx +42 -0
  171. package/docs/webhooks/providers/vapi.mdx +47 -0
  172. package/docs/webhooks/providers/vercel.mdx +40 -0
  173. package/docs/webhooks/providers/whatsapp.mdx +45 -0
  174. package/docs/webhooks/providers/workos.mdx +41 -0
  175. package/docs/webhooks/providers/zendesk.mdx +41 -0
  176. package/docs/webhooks/providers.mdx +53 -0
  177. package/docs/webhooks/quick-start.mdx +131 -0
  178. package/docs/webhooks/session-routing.mdx +6 -6
  179. package/docs/webhooks/signing-secrets.mdx +80 -0
  180. package/docs/webhooks/slack.mdx +11 -2
  181. package/docs/webhooks/verification.mdx +139 -0
  182. package/docs/webhooks/waiters.mdx +2 -2
  183. package/docs/webhooks/webhook-tasks.mdx +46 -0
  184. package/package.json +4 -5
  185. package/docs/migrating-from-v3.mdx +0 -332
  186. package/docs/webhooks/public-beta.mdx +0 -70
  187. package/docs/webhooks/sources.mdx +0 -252
@@ -81,7 +81,7 @@ export const myChat = chat.agent({
81
81
  // ... fallback initialization for non-preloaded runs
82
82
  },
83
83
  run: async ({ messages, signal, streamText }) => {
84
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
84
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
85
85
  },
86
86
  });
87
87
  ```
@@ -241,7 +241,8 @@ This is an **import-chain** problem, not a runtime one. A "we'll strip the execu
241
241
  id: "my-chat",
242
242
  run: async ({ messages, signal, streamText }) =>
243
243
  streamText({
244
- model: anthropic("claude-sonnet-4-6"),
244
+ model: anthropic("claude-sonnet-5-5"),
245
+ system: "You are a helpful assistant.",
245
246
  messages,
246
247
  tools: chatTools,
247
248
  stopWhen: stepCountIs(10),
@@ -263,7 +264,7 @@ This is an **import-chain** problem, not a runtime one. A "we'll strip the execu
263
264
  agentId: "my-chat",
264
265
  run: async ({ streamText }) =>
265
266
  streamText({
266
- model: anthropic("claude-sonnet-4-6"),
267
+ model: anthropic("claude-sonnet-5-5"),
267
268
  system: "You are a helpful assistant.",
268
269
  tools: headStartTools,
269
270
  }),
@@ -280,7 +281,7 @@ This is an **import-chain** problem, not a runtime one. A "we'll strip the execu
280
281
  </Note>
281
282
 
282
283
  <Tip>
283
- Use the **same model** on both sides (route handler and `chat.agent`) to avoid a tone or style shift between step 1 and step 2+. Your LLM provider keys stay server-side in your warm process — Trigger.dev never holds them in this design.
284
+ Use the **same model and system prompt** on both sides (route handler and `chat.agent`), with the same tool names and schemas. On Claude 5.5 models, step 1's thinking is bound to that system prompt and tool set, so a mismatch makes step 2 fail with a 400. Your LLM provider keys stay server-side in your warm process — Trigger.dev never holds them in this design.
284
285
  </Tip>
285
286
 
286
287
  Mount the handler in whatever framework you use — see [Mounting in your framework](#mounting-in-your-framework) below.
@@ -569,7 +570,7 @@ for await (const turn of session) {
569
570
  const result = turn.handover?.isFinal
570
571
  ? undefined
571
572
  : streamText({
572
- model: anthropic("claude-sonnet-4-6"),
573
+ model: anthropic("claude-sonnet-5-5"),
573
574
  messages: turn.messages,
574
575
  abortSignal: turn.signal,
575
576
  stopWhen: stepCountIs(10),
@@ -593,7 +594,7 @@ if (turn === 0 && payload.trigger === "handover-prepare") {
593
594
  // passing originalMessages so the tool output merges into the
594
595
  // handed-over assistant instead of starting a new message.
595
596
  const result = streamText({
596
- model: anthropic("claude-sonnet-4-6"),
597
+ model: anthropic("claude-sonnet-5-5"),
597
598
  messages: conversation.modelMessages,
598
599
  stopWhen: stepCountIs(10),
599
600
  });
@@ -691,7 +692,7 @@ export async function POST(req: Request) {
691
692
  messages, // first-turn user history
692
693
  run: async ({ streamText }) =>
693
694
  streamText({
694
- model: anthropic("claude-sonnet-4-6"),
695
+ model: anthropic("claude-sonnet-5-5"),
695
696
  system: "You are a helpful assistant.",
696
697
  tools: headStartTools,
697
698
  }),
@@ -267,7 +267,7 @@ Browser-supplied `clientData` is input, including after schema validation. Use i
267
267
  Pass metadata with individual messages via `sendMessage`. Per-message values are merged with transport-level client data (per-message wins on conflicts):
268
268
 
269
269
  ```ts
270
- sendMessage({ text: "Hello" }, { metadata: { model: "gpt-4o", priority: "high" } });
270
+ sendMessage({ text: "Hello" }, { metadata: { model: "gpt-6-luna", priority: "high" } });
271
271
  ```
272
272
 
273
273
  ### Typed client data with clientDataSchema
@@ -287,7 +287,7 @@ export const myChat = chat.agent({
287
287
  }),
288
288
  run: async ({ messages, clientData, signal, streamText }) => {
289
289
  return streamText({
290
- model: anthropic("claude-sonnet-4-5"),
290
+ model: anthropic("claude-sonnet-5-5"),
291
291
  system: clientData?.locale === "fr" ? "Reply in French." : "Reply in English.",
292
292
  messages,
293
293
  abortSignal: signal,
@@ -297,6 +297,10 @@ export const myChat = chat.agent({
297
297
  });
298
298
  ```
299
299
 
300
+ <Note>
301
+ On Claude 5.5 models, each thinking block is bound to the system prompt it was produced with, so changing `system` partway through a chat makes Anthropic reject the next request with a 400 (`Invalid signature in thinking block`). If users can switch language mid-chat, start a new chat when the locale changes.
302
+ </Note>
303
+
300
304
  The schema also types the `clientData` option on the frontend transport:
301
305
 
302
306
  ```ts
@@ -62,7 +62,7 @@ export const myChat = chat.agent({
62
62
  }
63
63
  },
64
64
  run: async ({ messages, signal }) => {
65
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
65
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
66
66
  },
67
67
  });
68
68
  ```
@@ -101,7 +101,7 @@ export const myChat = chat.agent({
101
101
  // (splice partial + first user into chain, dispatch the rest).
102
102
  },
103
103
  run: async ({ messages, signal }) =>
104
- streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal }),
104
+ streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal }),
105
105
  });
106
106
  ```
107
107
 
@@ -157,7 +157,7 @@ export const myChat = chat.agent({
157
157
  // ... non-preloaded chat-row initialization
158
158
  },
159
159
  run: async ({ messages, signal }) => {
160
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
160
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
161
161
  },
162
162
  });
163
163
  ```
@@ -206,7 +206,7 @@ export const myChat = chat.agent({
206
206
  });
207
207
  },
208
208
  run: async ({ messages, signal }) => {
209
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
209
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
210
210
  },
211
211
  });
212
212
  ```
@@ -245,7 +245,7 @@ export const myChat = chat.agent({
245
245
  return messages;
246
246
  },
247
247
  run: async ({ messages, signal }) => {
248
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, tools: chatTools, abortSignal: signal });
248
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, tools: chatTools, abortSignal: signal });
249
249
  },
250
250
  });
251
251
  ```
@@ -305,7 +305,7 @@ export const myChat = chat.agent({
305
305
  return stored;
306
306
  },
307
307
  run: async ({ messages, signal }) => {
308
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
308
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
309
309
  },
310
310
  });
311
311
  ```
@@ -363,7 +363,7 @@ export const myChat = chat.agent({
363
363
  });
364
364
  },
365
365
  run: async ({ messages, signal }) => {
366
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
366
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
367
367
  },
368
368
  });
369
369
  ```
@@ -398,7 +398,7 @@ export const myChat = chat.agent({
398
398
  }
399
399
  },
400
400
  run: async ({ messages, signal }) => {
401
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
401
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
402
402
  },
403
403
  });
404
404
  ```
@@ -446,7 +446,7 @@ export const myChat = chat.agent({
446
446
  ]);
447
447
  },
448
448
  run: async ({ messages, signal }) => {
449
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
449
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
450
450
  },
451
451
  });
452
452
  ```
@@ -489,7 +489,7 @@ export const myChat = chat.agent({
489
489
  logger.info("Resumed", { phase: event.phase });
490
490
  },
491
491
  run: async ({ messages, signal }) => {
492
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
492
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
493
493
  },
494
494
  });
495
495
  ```
@@ -523,7 +523,7 @@ export const myChat = chat.agent({
523
523
  await initializeChat(chatId, clientData);
524
524
  },
525
525
  run: async ({ messages, signal }) => {
526
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
526
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
527
527
  },
528
528
  });
529
529
  ```
@@ -119,7 +119,7 @@ export async function POST(req: Request) {
119
119
  if (!session) return new Response("Unauthorized", { status: 401 });
120
120
 
121
121
  const result = streamText({
122
- model: anthropic("claude-sonnet-4-5"),
122
+ model: anthropic("claude-sonnet-5-5"),
123
123
  system: "You are a helpful assistant.",
124
124
  messages: convertToModelMessages(messages),
125
125
  tools,
@@ -148,7 +148,7 @@ export const myChat = chat.agent({
148
148
  run: async ({ messages, tools, signal, streamText }) =>
149
149
  streamText({
150
150
  tools,
151
- model: anthropic("claude-sonnet-4-5"),
151
+ model: anthropic("claude-sonnet-5-5"),
152
152
  system: "You are a helpful assistant.",
153
153
  messages,
154
154
  abortSignal: signal,
@@ -359,7 +359,7 @@ export const myChat = chat.agent({
359
359
  run: async ({ messages, tools, signal, streamText }) =>
360
360
  streamText({
361
361
  tools,
362
- model: anthropic("claude-sonnet-4-5"),
362
+ model: anthropic("claude-sonnet-5-5"),
363
363
  system: "You are a helpful assistant.",
364
364
  messages,
365
365
  abortSignal: signal,
@@ -469,7 +469,7 @@ Head Start brings the route handler back for exactly that first turn. It runs st
469
469
  agentId: "my-chat",
470
470
  run: async ({ streamText }) =>
471
471
  streamText({
472
- model: anthropic("claude-sonnet-4-5"),
472
+ model: anthropic("claude-sonnet-5-5"),
473
473
  system: "You are a helpful assistant.",
474
474
  tools: headStartTools,
475
475
  }),
@@ -194,7 +194,7 @@ With `hydrateMessages`, your hook was the source of truth and the runtime wrote
194
194
  storage: transcriptStorage,
195
195
  clientDataSchema: z.object({ userId: z.string() }),
196
196
  run: async ({ messages, signal, streamText }) =>
197
- streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal }),
197
+ streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal }),
198
198
  });
199
199
  ```
200
200
  </Step>
@@ -19,7 +19,7 @@ export const myChat = chat.agent({
19
19
  id: "my-chat",
20
20
  run: async ({ messages, signal, streamText }) =>
21
21
  streamText({
22
- model: anthropic("claude-sonnet-4-5"),
22
+ model: anthropic("claude-sonnet-5-5"),
23
23
  messages,
24
24
  abortSignal: signal,
25
25
  stopWhen: stepCountIs(15),
@@ -97,7 +97,7 @@ export const myChat = chat.agent({
97
97
  },
98
98
  run: async ({ messages, signal, streamText }) => {
99
99
  return streamText({
100
- model: anthropic("claude-sonnet-4-5"),
100
+ model: anthropic("claude-sonnet-5-5"),
101
101
  messages,
102
102
  abortSignal: signal,
103
103
  });
@@ -227,7 +227,7 @@ export const myChat = chat.agent({
227
227
  ]);
228
228
  },
229
229
  run: async ({ messages, signal, streamText }) => {
230
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
230
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
231
231
  },
232
232
  });
233
233
  ```
@@ -298,7 +298,7 @@ export const myChat = chat.agent({
298
298
  },
299
299
  run: async ({ messages, signal, streamText }) => {
300
300
  return streamText({
301
- model: anthropic("claude-sonnet-4-5"),
301
+ model: anthropic("claude-sonnet-5-5"),
302
302
  messages,
303
303
  abortSignal: signal,
304
304
  stopWhen: stepCountIs(15),
@@ -76,7 +76,7 @@ export const myChat = chat.withUIMessage<ChatMessage>().agent({
76
76
  tools,
77
77
  run: async ({ messages, tools, signal, streamText }) => {
78
78
  return streamText({
79
- model: anthropic("claude-sonnet-4-5"),
79
+ model: anthropic("claude-sonnet-5-5"),
80
80
  messages,
81
81
  tools,
82
82
  abortSignal: signal,
@@ -50,23 +50,28 @@ const isToolPart = (p: { type?: string }) =>
50
50
  typeof p?.type === "string" && (p.type.startsWith("tool-") || p.type === "dynamic-tool");
51
51
 
52
52
  // Drop the oldest n tool parts, mirroring what the provider cleared. A tool call
53
- // and its result live in one part, so pairing stays intact.
53
+ // and its result live in one part, so pairing stays intact. Once a message is
54
+ // edited, its thinking and every later message's thinking no longer verify on
55
+ // Claude 5.5 models, so those reasoning parts are dropped too.
54
56
  function pruneOldestToolParts(messages: UIMessage[], n: number): UIMessage[] {
55
57
  let toRemove = n;
58
+ let edited = false;
56
59
  const out: UIMessage[] = [];
57
60
  for (const m of messages) {
58
- if (toRemove <= 0 || m.role !== "assistant" || !m.parts) {
61
+ if (m.role !== "assistant" || !m.parts) {
59
62
  out.push(m);
60
63
  continue;
61
64
  }
62
65
  const kept = m.parts.filter((p) => {
63
66
  if (toRemove > 0 && isToolPart(p)) {
64
67
  toRemove--;
68
+ edited = true;
65
69
  return false;
66
70
  }
67
71
  return true;
68
72
  });
69
- if (kept.length > 0) out.push({ ...m, parts: kept });
73
+ const parts = edited ? kept.filter((p) => p.type !== "reasoning") : kept;
74
+ if (parts.length > 0) out.push({ ...m, parts });
70
75
  }
71
76
  return out;
72
77
  }
@@ -77,7 +82,7 @@ export const nativePersist = chat.agent({
77
82
  tools: { fetchRecord },
78
83
  run: async ({ messages, chatId, tools, signal, streamText }) => {
79
84
  return streamText({
80
- model: anthropic("claude-sonnet-4-5"),
85
+ model: anthropic("claude-sonnet-5-5"),
81
86
  messages,
82
87
  tools,
83
88
  abortSignal: signal,
@@ -126,6 +131,10 @@ Turn 1 sends the user message and accumulates six tool results. Anthropic clears
126
131
  `onTurnComplete` is where persistence happens. Action turns fire `onAction` only, and a `chat.history.set()` inside `run()` is overwritten by the accumulator at turn end. See [Persistence and replay](/ai-chat/patterns/persistence-and-replay#action-turns-no-snapshot-write).
127
132
  </Note>
128
133
 
134
+ <Warning>
135
+ Thinking is on by default on Claude 5.5 models, and each thinking block is bound to the conversation that came before it. Removing tool parts from earlier turns in your own code changes that history, so Anthropic can reject the next request with a 400 (`Invalid signature in thinking block`). Server-side context editing, as configured above, doesn't count as an edit. That's why `pruneOldestToolParts` also removes the reasoning parts from the turns it changes and every turn after them.
136
+ </Warning>
137
+
129
138
  ## Persist OpenAI stored responses
130
139
 
131
140
  OpenAI's `store: true` keeps the thread server-side and returns a `responseId`. Pass that back as `previousResponseId` on the next turn and send only the messages since the last assistant reply; everything before it lives on OpenAI's side.
@@ -156,7 +165,7 @@ export const openaiStore = chat.agent({
156
165
  const outbound = native ? messagesSinceLastAssistant(messages) : messages;
157
166
 
158
167
  const result = streamText({
159
- model: openai("gpt-4o"),
168
+ model: openai("gpt-6-luna"),
160
169
  messages: outbound,
161
170
  abortSignal: signal,
162
171
  stopWhen: stepCountIs(5),
@@ -197,10 +206,17 @@ const FALLBACK_ORDER: Provider[] = ["anthropic", "openai"];
197
206
  type NativeState = { provider: "openai"; previousResponseId: string };
198
207
  const nativeStore = new Map<string, NativeState>();
199
208
 
209
+ // Retained messages follow a different history after compaction, so their thinking
210
+ // blocks no longer verify on Claude 5.5 models. Drop reasoning parts before replaying them.
211
+ function stripReasoning(message: ModelMessage): ModelMessage {
212
+ if (message.role !== "assistant" || typeof message.content === "string") return message;
213
+ return { ...message, content: message.content.filter((part) => part.type !== "reasoning") };
214
+ }
215
+
200
216
  // Provider-agnostic summary: a plain string, portable across any provider.
201
217
  async function summarizeConversation(messages: ModelMessage[]): Promise<string> {
202
218
  const { text } = await generateText({
203
- model: openai("gpt-4o-mini"),
219
+ model: openai("gpt-6-luna"),
204
220
  messages: [
205
221
  ...messages,
206
222
  {
@@ -223,7 +239,7 @@ export const resilientChat = chat.agent({
223
239
  summarize: ({ messages }) => summarizeConversation(messages),
224
240
  compactModelMessages: ({ modelMessages, summary }) => [
225
241
  { role: "user", content: `Summary of the conversation so far:\n\n${summary}` },
226
- ...modelMessages.slice(-2),
242
+ ...modelMessages.slice(-2).map(stripReasoning),
227
243
  ],
228
244
  compactUIMessages: ({ uiMessages, summary }) => [
229
245
  {
@@ -251,7 +267,7 @@ export const resilientChat = chat.agent({
251
267
  // compacted baseline (summary + recent), so raw history is not re-sent.
252
268
  const useHandle = native?.provider === "openai";
253
269
  const result = streamText({
254
- model: openai("gpt-4o"),
270
+ model: openai("gpt-6-luna"),
255
271
  messages,
256
272
  abortSignal: signal,
257
273
  stopWhen: stepCountIs(5),
@@ -269,7 +285,7 @@ export const resilientChat = chat.agent({
269
285
  }
270
286
 
271
287
  return streamText({
272
- model: anthropic("claude-sonnet-4-5"),
288
+ model: anthropic("claude-sonnet-5-5"),
273
289
  messages,
274
290
  abortSignal: signal,
275
291
  stopWhen: stepCountIs(5),
@@ -292,6 +308,10 @@ export const resilientChat = chat.agent({
292
308
 
293
309
  When Anthropic is down, the loop falls through to OpenAI. Because `compaction` has already reduced `messages` to a summary plus the last couple of exchanges, the switch sends the portable baseline, not megabytes of raw transcript.
294
310
 
311
+ <Note>
312
+ The messages that `compactModelMessages` keeps after the summary now follow a different history from the one their thinking blocks were produced in. Claude 5.5 models reject those blocks with a 400 (`Invalid signature in thinking block`), which is why the example runs the retained messages through `stripReasoning`.
313
+ </Note>
314
+
295
315
  <Warning>
296
316
  Fallback here retries a turn that hasn't started streaming yet. Once a response is streaming to the client, a mid-stream provider failure can't be swapped transparently. Surface the error and let the frontend regenerate the turn. See [Error handling](/ai-chat/error-handling).
297
317
  </Warning>
@@ -163,7 +163,7 @@ export const myChat = chat.agent({
163
163
  await db.chat.update({ where: { id: chatId }, data: { messages: uiMessages } });
164
164
  },
165
165
  run: async ({ messages, signal }) => {
166
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
166
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
167
167
  },
168
168
  });
169
169
  ```
@@ -28,7 +28,7 @@ export const supportAgent = chat.agent({
28
28
  id: "support-agent",
29
29
  telemetry: { recordUserMessage: false, recordClientData: false },
30
30
  run: async ({ messages, signal, streamText }) =>
31
- streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal }),
31
+ streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal }),
32
32
  });
33
33
  ```
34
34
 
@@ -47,7 +47,7 @@ export const supportAgent = chat.agent({
47
47
  telemetry: { recordUserMessage: false, recordClientData: false },
48
48
  run: async ({ messages, signal, streamText }) =>
49
49
  streamText({
50
- model: anthropic("claude-sonnet-4-5"),
50
+ model: anthropic("claude-sonnet-5-5"),
51
51
  messages,
52
52
  abortSignal: signal,
53
53
  // Replaces the default telemetry config (and chat.prompt.set()'s), so keep isEnabled on
@@ -94,7 +94,7 @@ export const agent = chat.agent({
94
94
  },
95
95
  run: async ({ messages, signal, streamText }) => {
96
96
  return streamText({
97
- model: anthropic("claude-sonnet-4-5"),
97
+ model: anthropic("claude-sonnet-5-5"),
98
98
  messages,
99
99
  abortSignal: signal,
100
100
  stopWhen: stepCountIs(15),
@@ -165,7 +165,7 @@ If you also define your own AI SDK tools, pass them through `chat.toStreamTextOp
165
165
 
166
166
  ```ts
167
167
  return streamText({
168
- model: anthropic("claude-sonnet-4-5"),
168
+ model: anthropic("claude-sonnet-5-5"),
169
169
  messages,
170
170
  abortSignal: signal,
171
171
  tools: {
@@ -301,7 +301,7 @@ export const myAgent = chat.agent({
301
301
  tools, // declare here so toModelOutput survives across turns
302
302
  run: async ({ messages, tools, stopSignal, streamText }) => {
303
303
  return streamText({
304
- model: anthropic("claude-sonnet-4-6"),
304
+ model: anthropic("claude-sonnet-5-5"),
305
305
  messages,
306
306
  tools,
307
307
  abortSignal: stopSignal,
@@ -37,7 +37,7 @@ export const myChat = chat.agent({
37
37
  return await db.getMessages(chatId);
38
38
  },
39
39
  run: async ({ messages, signal, streamText }) => {
40
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
40
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
41
41
  },
42
42
  });
43
43
  ```
@@ -247,7 +247,7 @@ export const myAgent = chat
247
247
  // Score-based routing. The values arrive from the edge proxy.
248
248
  if (clientData.__cf.botScore < 30) {
249
249
  return streamText({
250
- model: anthropic("claude-haiku-4-5"),
250
+ model: anthropic("claude-haiku-5-5"),
251
251
  messages: [{ role: "system", content: "Reject politely; do not engage." }],
252
252
  abortSignal: signal,
253
253
  stopWhen: stepCountIs(15),
@@ -255,7 +255,7 @@ export const myAgent = chat
255
255
  }
256
256
 
257
257
  return streamText({
258
- model: anthropic("claude-sonnet-4-5"),
258
+ model: anthropic("claude-sonnet-5-5"),
259
259
  messages,
260
260
  abortSignal: signal,
261
261
  // ...
@@ -122,7 +122,7 @@ export const myChat = chat
122
122
  }
123
123
  },
124
124
  run: async ({ messages, signal, streamText }) => {
125
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
125
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
126
126
  },
127
127
  });
128
128
  ```
@@ -184,7 +184,7 @@ export const myChat = chat
184
184
  }
185
185
  },
186
186
  run: async ({ messages, signal, streamText }) => {
187
- return streamText({ model: anthropic("claude-sonnet-4-5"), messages, abortSignal: signal });
187
+ return streamText({ model: anthropic("claude-sonnet-5-5"), messages, abortSignal: signal });
188
188
  },
189
189
  });
190
190
  ```
@@ -45,7 +45,7 @@ export const myChat = chat.agent({
45
45
  },
46
46
  run: async ({ messages, signal, streamText }) => {
47
47
  return streamText({
48
- model: anthropic("claude-sonnet-4-5"),
48
+ model: anthropic("claude-sonnet-5-5"),
49
49
  messages,
50
50
  tools: {
51
51
  inspectDocument: tool({
@@ -151,7 +151,7 @@ const session = chat.createSession(payload, {
151
151
 
152
152
  for await (const turn of session) {
153
153
  const result = streamText({
154
- model: anthropic("claude-sonnet-4-5"),
154
+ model: anthropic("claude-sonnet-5-5"),
155
155
  messages: turn.messages,
156
156
  abortSignal: turn.signal,
157
157
  prepareStep: turn.prepareStep(), // Handles injection + compaction
@@ -193,7 +193,7 @@ for (let turn = 0; turn < 100; turn++) {
193
193
  });
194
194
 
195
195
  const result = streamText({
196
- model: anthropic("claude-sonnet-4-5"),
196
+ model: anthropic("claude-sonnet-5-5"),
197
197
  messages,
198
198
  prepareStep: conversation.prepareStep(), // Handles injection + compaction
199
199
  stopWhen: stepCountIs(15),
@@ -248,7 +248,7 @@ function Chat({ chatId }: { chatId: string }) {
248
248
  messages,
249
249
  setMessages,
250
250
  sendMessage,
251
- metadata: { model: "gpt-4o" },
251
+ metadata: { model: "gpt-6-luna" },
252
252
  });
253
253
 
254
254
  return (
@@ -345,7 +345,7 @@ const sent = await transport.sendPendingMessage(chatId, {
345
345
  id: crypto.randomUUID(),
346
346
  role: "user",
347
347
  parts: [{ type: "text", text: "and compare to vercel" }],
348
- }, { model: "gpt-4o" });
348
+ }, { model: "gpt-6-luna" });
349
349
  ```
350
350
 
351
351
  Unlike `sendMessage()` from useChat, this does NOT:
@@ -4,7 +4,7 @@ sidebarTitle: "Prompt caching"
4
4
  description: "Cache the stable prefix of your agent's prompt with Anthropic prompt caching to cut token cost and latency on every turn."
5
5
  ---
6
6
 
7
- **Prompt caching lets a provider reuse the unchanged prefix of your prompt across requests, billing it at a fraction of the input price and skipping re-processing.** With Anthropic, cache reads cost ~10% of base input tokens, so a long, stable system prompt or a growing conversation history pays full price once and reads cheaply on every turn after.
7
+ **Prompt caching lets a provider reuse the unchanged prefix of your prompt across requests, billing it at a fraction of the input price and skipping re-processing.** With Anthropic, cache reads cost ~5% of base input tokens on Claude Sonnet 5.5 and Opus 5.5, so a long, stable system prompt or a growing conversation history pays full price once and reads cheaply on every turn after.
8
8
 
9
9
  Caching is a **byte-exact prefix match**: any change in the prefix invalidates everything after it. A multi-turn agent is the ideal case — the system prompt, tools, and earlier turns are identical turn over turn, so the cacheable prefix only grows. `chat.agent` is built to keep that prefix stable across turns, suspends, and resumes; this page shows how to place the cache breakpoints and verify they're hitting.
10
10
 
@@ -46,7 +46,7 @@ export const myChat = chat.agent({
46
46
  },
47
47
  run: async ({ messages, signal, streamText }) => {
48
48
  return streamText({
49
- model: anthropic("claude-sonnet-4-6"),
49
+ model: anthropic("claude-sonnet-5-5"),
50
50
  // Caches the system block with a 5-minute breakpoint.
51
51
  messages,
52
52
  abortSignal: signal,
@@ -63,7 +63,7 @@ export const myChat = chat.agent({
63
63
  systemProviderOptions: { anthropic: { cacheControl: { type: "ephemeral" } } },
64
64
  run: async ({ messages, signal, streamText }) =>
65
65
  streamText({
66
- model: anthropic("claude-sonnet-4-6"),
66
+ model: anthropic("claude-sonnet-5-5"),
67
67
  messages,
68
68
  abortSignal: signal,
69
69
  }),
@@ -80,7 +80,7 @@ onChatStart: async () => {
80
80
  },
81
81
  run: async ({ messages, signal, streamText }) => {
82
82
  return streamText({
83
- model: anthropic("claude-sonnet-4-6"),
83
+ model: anthropic("claude-sonnet-5-5"),
84
84
  messages,
85
85
  abortSignal: signal,
86
86
  });
@@ -117,7 +117,7 @@ export const myChat = chat.agent({
117
117
  },
118
118
  run: async ({ messages, signal, streamText }) => {
119
119
  return streamText({
120
- model: anthropic("claude-sonnet-4-6"),
120
+ model: anthropic("claude-sonnet-5-5"),
121
121
  messages,
122
122
  abortSignal: signal,
123
123
  });
@@ -128,7 +128,7 @@ export const myChat = chat.agent({
128
128
  The system breakpoint and the conversation breakpoint compose: the system block is cached once for the life of the chat, and each turn extends the cached message prefix.
129
129
 
130
130
  <Note>
131
- Anthropic allows **at most 4** cache breakpoints per request, and a prefix must be at least ~1024 tokens (model-dependent) to cache at all — shorter prefixes silently don't cache. One system breakpoint plus one rolling message breakpoint is the typical setup and leaves headroom.
131
+ Anthropic allows **at most 4** cache breakpoints per request, and a prefix must reach a model-dependent minimum to cache at all (512 tokens on Claude Sonnet 5.5 and Opus 5.5) — shorter prefixes silently don't cache. One system breakpoint plus one rolling message breakpoint is the typical setup and leaves headroom.
132
132
  </Note>
133
133
 
134
134
  <Note>
@@ -171,7 +171,7 @@ run: async ({ messages, signal, previousTurnUsage, streamText }) => {
171
171
  console.log("cache write", previousTurnUsage?.inputTokenDetails?.cacheWriteTokens);
172
172
 
173
173
  return streamText({
174
- model: anthropic("claude-sonnet-4-6"),
174
+ model: anthropic("claude-sonnet-5-5"),
175
175
  messages,
176
176
  abortSignal: signal,
177
177
  });
@@ -23,7 +23,7 @@ The chat surface works with Vercel AI SDK **v5, v6, or v7**; install whichever m
23
23
  id: "my-chat",
24
24
  run: async ({ messages, signal, streamText }) => {
25
25
  return streamText({
26
- model: anthropic("claude-sonnet-4-5"),
26
+ model: anthropic("claude-sonnet-5-5"),
27
27
  messages,
28
28
  abortSignal: signal,
29
29
  stopWhen: stepCountIs(15),
@@ -76,7 +76,7 @@ The chat surface works with Vercel AI SDK **v5, v6, or v7**; install whichever m
76
76
 
77
77
  `requireChatOwner` is your application helper: authenticate the request, load the chat by ID and owner, and throw if it doesn't belong to that user. Create the chat record on your server before rendering the frontend, and pass its ID into `Chat`. Check ownership in both actions, including token refresh.
78
78
 
79
- Set `TRIGGER_SECRET_KEY` and your model provider key in the server and worker environments. Keep both keys out of the browser.
79
+ Set `TRIGGER_SECRET_KEY` on your server and your model provider key in the worker environment. Keep both keys out of the browser.
80
80
 
81
81
  </Step>
82
82