Prompt submitted to cloud even when using local LLM by default

A bug in opencode sends user prompts to a public cloud endpoint at opencode.ai/zen/v1/responses even when a local LLM is configured, exposing data to OpenAI's GPT-5-nano model for title generation. The issue, filed as #15854, affects core functionality and raises privacy concerns for users expecting local-only processing.

- Notifications /login?return to=%2Fanomalyco%2Fopencode You must be signed in to change notification settings - Fork 23.4k /login?return to=%2Fanomalyco%2Fopencode Prompt submitted to cloud even when using local LLM by default 15854 Copy link Copy link Closed as not planned Labels bugSomething isn't working https://github.com/anomalyco/opencode/issues?q=state%3Aopen%20label%3A%22bug%22 Something isn't working coreAnything pertaining to core functionality of the application opencode server stuff https://github.com/anomalyco/opencode/issues?q=state%3Aopen%20label%3A%22core%22 Anything pertaining to core functionality of the application opencode server stuff Description Description When using a local model, opencode still sends your prompt to public cloud endpoint https://opencode.ai/zen/v1/responses https://opencode.ai/zen/v1/responses to get a title summary. For example: opencode run this is a test or even with an explicit model opencode -m local/Qwen/Qwen-Next-Coder run this is a test causes opencode to generate a request POST /zen/v1/responses HTTP/2 Host: opencode.ai Authorization: Bearer public Content-Type: application/json User-Agent: ai-sdk/openai/2.0.89 ai-sdk/provider-utils/3.0.20 runtime/bun/1.3.10 X-Opencode-Client: cli X-Opencode-Project: global X-Opencode-Request: msg cb41776da0013RO7pnTvyuSA0h X-Opencode-Session: ses 34be88938ffewkdc070WPvraqG Connection: keep-alive Accept: / Accept-Encoding: gzip, deflate, br Content-Length: 2574 { "model": "gpt-5-nano", "input": { "role": "developer", "content": "You are a title generator. You output ONLY a thread title. Nothing else.\n\n<task \nGenerate a brief title that would help the user find this conversation later.\n\nFollow all rules in <rules \nUse the <examples so you know what a good title looks like.\nYour output must be:\n- A single line\n- ≤50 characters\n- No explanations\n</task \n\n<rules \n- you MUST use the same language as the user message you are summarizing\n- Title must be grammatically correct and read naturally - no word salad\n- Never include tool names in the title e.g. \"read tool\", \"bash tool\", \"edit tool\" \n- Focus on the main topic or question the user needs to retrieve\n- Vary your phrasing - avoid repetitive patterns like always starting with \"Analyzing\"\n- When a file is mentioned, focus on WHAT the user wants to do WITH the file, not just that they shared it\n- Keep exact: technical terms, numbers, filenames, HTTP codes\n- Remove: the, this, my, a, an\n- Never assume tech stack\n- Never use tools\n- NEVER respond to questions, just generate a title for the conversation\n- The title should NEVER include \"summarizing\" or \"generating\" when generating a title\n- DO NOT SAY YOU CANNOT GENERATE A TITLE OR COMPLAIN ABOUT THE INPUT\n- Always output something meaningful, even if the input is minimal.\n- If the user message is short or conversational e.g. \"hello\", \"lol\", \"what's up\", \"hey\" :\n → create a title that reflects the user's tone or intent such as Greeting, Quick check-in, Light chat, Intro message, etc. \n</rules \n\n<examples \n\"debug 500 errors in production\" → Debugging production 500 errors\n\"refactor user service\" → Refactoring user service\n\"why is app.js failing\" → app.js failure investigation\n\"implement rate limiting\" → Rate limiting implementation\n\"how do I connect postgres to my API\" → Postgres API connection\n\"best practices for React hooks\" → React hooks best practices\n\"@src/auth.ts can you add refresh token support\" → Auth refresh token support\n\"@utils/parser.ts this is broken\" → Parser bug fix\n\"look at @config.json\" → Config review\n\"@App.tsx add dark mode toggle\" → Dark mode toggle in App\n</examples \n" }, { "role": "user", "content": { "type": "input text", "text": "Generate a title for this conversation:\n" } }, { "role": "user", "content": { "type": "input text", "text": "this is a test" } } , "max output tokens": 32000, "store": false, "include": "reasoning.encrypted content" , "reasoning": { "effort": "minimal" }, "stream": true } This has leaked your private prompt text to the opencode API. It can be disabled using "agent":{ "title": { "disable":true } } The default should be either disabled, or to use the model you specify. Plugins None OpenCode version 1.2.15 Steps to reproduce Configure a local model export https proxy= http://127.0.0.1:8080 http://127.0.0.1:8080 export NODE TLS REJECT UNAUTHORIZED=0 run Burp proxy or similar to capture requests run: opencode run this is a test Check Burp to see the request to opencode.ai with "this is a test" Screenshot and/or share link No response Operating System Ubuntu 24 Terminal No response Reactions are currently unavailable Metadata Metadata Assignees Labels bugSomething isn't working https://github.com/anomalyco/opencode/issues?q=state%3Aopen%20label%3A%22bug%22 Something isn't working coreAnything pertaining to core functionality of the application opencode server stuff https://github.com/anomalyco/opencode/issues?q=state%3Aopen%20label%3A%22core%22 Anything pertaining to core functionality of the application opencode server stuff Type Fields No fields configured for issues without a type.