fix(telemetry): restore userId and sessionId metadata in experimental_telemetry

[Raviguntakala]

What does this PR do?

Fixes #8193

Restores userId and sessionId metadata to experimental_telemetry in llm.ts, which was removed during LLM refactoring.

Originally added in #5279.

How did you verify your code works?

Tested with Langfuse - traces correctly group by session and display user attribution.

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

No duplicate PRs found

[nonepage]

Wow, he's very useful. Some models' cache hits also depend on this. Why hasn't anyone merged them😂

[Raviguntakala]

Hey @rekram1-node , can you take a quick look at this PR when you have time? Simple regression fix.

[SXGC]

This would be a great addition to the experimental telemetry. Hoping to see this merged soon!

[SimonLiu423]

I'm experiencing the same issue from #8193 and would benefit from this fix.

[tisoz]

please merged , i also need this

[tisoz]

What does this PR do?

Fixes #8193修复内容 #8193

Restores userId and sessionId metadata to experimental_telemetry in llm.ts, which was removed during LLM refactoring.恢复在 llm.ts 中 userId 和 sessionId 元数据到 experimental_telemetry，这些数据在 LLM 重构期间被移除。

Originally added in #5279.  最初在 #5279 中添加。

How did you verify your code works?

Tested with Langfuse - traces correctly group by session and display user attribution.使用 Langfuse 进行测试 - 跟踪记录能够正确按会话分组并显示用户归属。

hi bro ， can i got your help , i tried modify in source code .
but the code is not work .

do you have any idea

this is log :

INFO  2026-01-29T11:31:52 +2ms service=llm providerID=anthropic modelID=claude-haiku-4-5 sessionID=ses_3f679d82cffeN5IVnupdZJDlYU small=true agent=title mode=primary experimental_telemetry={"isEnabled":true,"metadata":{"userId":"tisoz","sessionId":"ses_3f679d82cffeN5IVnupdZJDlYU"}} tisoz_test2
ERROR 2026-01-29T11:31:53 +287ms service=llm providerID=anthropic modelID=claude-haiku-4-5 sessionID=ses_3f679d82cffeN5IVnupdZJDlYU small=true agent=title mode=primary error={"error":{"name":"AI_APICallError","url":"https://code.newcli.com/claude/aws/v1/messages","requestBodyValues":{"model":"claude-haiku-4-5","max_tokens":32000,"temperature":0.5,"system":[{"type":"text","text":"You are Claude Code, Anthropic's official CLI for Claude.","cache_control":{"type":"ephemeral"}},{"type":"text","text":"You are Claude Code, Anthropic's official CLI for Claude.\n\nYou are a title generator. You output ONLY a thread title. Nothing else.\r\n\r\n<task>\r\nGenerate a brief title that would help the user find this conversation later.\r\n\r\nFollow all rules in <rules>\r\nUse the <examples> so you know what a good title looks like.\r\nYour output must be:\r\n- A single line\r\n- ≤50 characters\r\n- No explanations\r\n</task>\r\n\r\n<rules>\r\n- you MUST use the same language as the user message you are summarizing\r\n- Title must be grammatically correct and read naturally - no word salad\r\n- Never include tool names in the title (e.g. \"read tool\", \"bash tool\", \"edit tool\")\r\n- Focus on the main topic or question the user needs to retrieve\r\n- Vary your phrasing - avoid repetitive patterns like always starting with \"Analyzing\"\r\n- When a file is mentioned, focus on WHAT the user wants to do WITH the file, not just that they shared it\r\n- Keep exact: technical terms, numbers, filenames, HTTP codes\r\n- Remove: the, this, my, a, an\r\n- Never assume tech stack\r\n- Never use tools\r\n- NEVER respond to questions, just generate a title for the conversation\r\n- The title should NEVER include \"summarizing\" or \"generating\" when generating a title\r\n- DO NOT SAY YOU CANNOT GENERATE A TITLE OR COMPLAIN ABOUT THE INPUT\r\n- Always output something meaningful, even if the input is minimal.\r\n- If the user message is short or conversational (e.g. \"hello\", \"lol\", \"what's up\", \"hey\"):\r\n  → create a title that reflects the user's tone or intent (such as Greeting, Quick check-in, Light chat, Intro message, etc.)\r\n</rules>\r\n\r\n<examples>\r\n\"debug 500 errors in production\" → Debugging production 500 errors\r\n\"refactor user service\" → Refactoring user service\r\n\"why is app.js failing\" → app.js failure investigation\r\n\"implement rate limiting\" → Rate limiting implementation\r\n\"how do I connect postgres to my API\" → Postgres API connection\r\n\"best practices for React hooks\" → React hooks best practices\r\n\"@src/auth.ts can you add refresh token support\" → Auth refresh token support\r\n\"@utils/parser.ts this is broken\" → Parser bug fix\r\n\"look at @config.json\" → Config review\r\n\"@App.tsx add dark mode toggle\" → Dark mode toggle in App\r\n</examples>\r\n","cache_control":{"type":"ephemeral"}}],"messages":[{"role":"user","content":[{"type":"text","text":"Generate a title for this conversation:\n","cache_control":{"type":"ephemeral"}},{"type":"text","text":"hi","cache_control":{"type":"ephemeral"}}]}],"stream":true},"statusCode":400,"responseHeaders":{"cf-cache-status":"DYNAMIC","cf-ray":"9c5867df390ee9df-LAX","connection":"keep-alive","content-length":"138","content-type":"text/plain; charset=utf-8","date":"Thu, 29 Jan 2026 11:31:53 GMT","nel":"{\"report_to\":\"cf-nel\",\"success_fraction\":0.0,\"max_age\":604800}","report-to":"{\"group\":\"cf-nel\",\"max_age\":604800,\"endpoints\":[{\"url\":\"https://a.nel.cloudflare.com/report/v4?s=68aaYRodj07P9ql4CGzuYxsX8ZA2k8li3w69OnuwEClMtoPP01bogeHOWBeaO%2FTNVbbKUc9C5fWSHjN5zHJvq3ycQBy1vnMShQ0KfuWS\"}]}","server":"cloudflare","x-content-type-options":"nosniff"},"responseBody":"{\"error\":{\"code\":\"model_not_found\",\"message\":\"模型 claude-haiku-4-5 未开放，可通过/model切换到sonnet4.5模型\",\"type\":\"404\"}}\n","isRetryable":false}} stream error

i noticed the request body not take metadata params

[Raviguntakala]

@tisoz The Vercel AI SDK doesn't send any metadata values present experimental_telemetry to LLM provider APIs. The experimental_telemetry parameter is only used for telemetry purposes - these values are NOT sent to the LLM request body

[tisoz]

@tisoz The Vercel AI SDK doesn't send any metadata values present experimental_telemetry to LLM provider APIs. The experimental_telemetry parameter is only used for telemetry purposes - these values are NOT sent to the LLM request bodyVercel AI SDK 不会向 LLM 提供商 API 发送任何包含 experimental_telemetry 的元数据值。experimental_telemetry 参数仅用于遥测目的——这些值不会被发送到 LLM 请求体中。

thanks bro , some proviver use metadata to enabled cache fetures.

maybe i need found a custom the way . looks not good :(