[FEATURE]: Add logging of LLM HTTP Requests (v2) #56066

Activity

1. opencode-agent commented on Oct 16, 2026

   

   opencode-agentBot

   on Oct 16, 2026 – with OpenCode Agent

   Contributor

   More actions

   Thanks for the request! You'd like OpenCode v2 to log outbound LLM HTTP requests (URL, headers and optionally payloads) so failed provider calls are easier to debug. The maintainers will consider it.

   Related issues:

   • [FEATURE]: Log LLM API request/response body at DEBUG log level #29186: older request for request/response body logging at debug level (written against the v1 code)
   • [FEATURE]: propagate W3C traceparent on outbound LLM HTTP requests #49038: propagating traceparent on outbound LLM requests
   • [FEATURE]: Plugin hook to intercept and override LLM call parameters (llm.request.before) #45764: plugin hook to intercept LLM call parameters

   In the meantime, you can do this yourself with a plugin. The v2 plugin API has http.request and http.response session hooks that see every HTTP request sent to a model provider. Here is a small plugin that appends each request and response as one JSON line to a file. It hides auth headers, logs request bodies, and logs response bodies only for error responses, because successful responses are streams and are left alone. I've checked that it type-checks against the plugin API, but I haven't run it inside OpenCode yet.

   Save it as .opencode/plugins/llm-http-log/index.ts in your project (plugins there are loaded automatically), or put it anywhere and add its path to plugins in opencode.json:

   import { appendFile, mkdir } from "node:fs/promises"
   import { homedir } from "node:os"
   import { dirname, join } from "node:path"
   import { Plugin } from "@opencode/plugin"
   
   const SECRET = /^(authorization|x-api-key|api-key|x-goog-api-key|cookie|proxy-authorization)$/i
   
   function headers(input: Headers) {
     const out: Record<string, string> = {}
     input.forEach((value, key) => {
       out[key] = SECRET.test(key) ? "[redacted]" : value
     })
     return out
   }
   
   export default Plugin.define({
     id: "llm-http-log",
     async setup(ctx) {
       const file =
         typeof ctx.options.file === "string" ? ctx.options.file : join(homedir(), ".local/share/opencode/llm-http.jsonl")
       const bodies = ctx.options.bodies !== false
       await mkdir(dirname(file), { recursive: true })
       const write = (entry: Record<string, unknown>) =>
         appendFile(file, JSON.stringify({ time: new Date().toISOString(), ...entry }) + "\n").catch(() => {})
   
       await ctx.session.hook("http.request", async (event) => {
         const req = event.request
         await write({
           type: "request",
           sessionID: event.sessionID,
           kind: event.kind,
           model: event.model,
           method: req.method,
           url: req.url,
           headers: headers(req.headers),
           body: bodies && req.body ? await req.clone().text() : undefined,
         })
       })
   
       await ctx.session.hook("http.response", async (event) => {
         const res = event.response
         await write({
           type: "response",
           sessionID: event.sessionID,
           kind: event.kind,
           model: event.model,
           url: event.request.url,
           status: res.status,
           headers: headers(res.headers),
           // Only error bodies are read; successful bodies are streams and are left untouched.
           body: !res.ok ? await res.clone().text() : undefined,
         })
       })
     },
   })

   The options are optional. Use them to change the log file or turn off request bodies:

   {
     "plugins": [
       { "package": "./.opencode/plugins/llm-http-log", "options": { "file": "/tmp/llm-http.jsonl", "bodies": false } }
     ]
   }

   Then run tail -f ~/.local/share/opencode/llm-http.jsonl while you reproduce the failing request. Request bodies can contain your prompts and file contents, so check the log before sharing it. Providers that use WebSocket transport won't show up here, because these hooks only cover HTTP.
2. neriousy commented on Oct 20, 2026

   

   neriousy

   on Oct 20, 2026

   Member

   More actions

   Thanks for the suggestion. We're tracking built-in LLM request and response logging for debugging in #29186, so I'm closing this as a duplicate. Please follow that issue for updates.

   👍React with 👍1Reacted by martincmb
3. neriousy

   closed this as a duplicateof [FEATURE]: Log LLM API request/response body at DEBUG log level #29186 on Oct 20, 2026
4. mschfh commented on Nov 2, 2026

   

   mschfh

   on Nov 2, 2026

   More actions

   @neriousy The linked issue was closed as stale by you. Will there be further updates? Is this available in v2?