tui: client polls http://localhost:8000/health every 30s from any project #45172

Activity

1. github-actions

   assigned

   kommander

   on Aug 26, 2026
2. opencode-agent

   added

   2.0

   on Aug 26, 2026
3. maxstevens-nl commented on Sep 22, 2026

   

   maxstevens-nl

   on Sep 22, 2026

   Contributor

   More actions

   They seem to be scanning the localhost for an active vLLM instance. Workaround, turning off the vllm plugin:

   // .config/opencode/opencode.json
   {
     "$schema": "https://opencode.ai/config.json",
     "plugins": ["-opencode.provider.vllm"]
   }
4. neriousy

   marked [FEATURE]: add option to disable dev-server health checks #51551 as a duplicate of this issue on Oct 20, 2026
5. opencodejr commented on Oct 28, 2026

   

   opencodejrBot

   on Oct 28, 2026

   Contributor

   More actions

   Thanks for the detailed report and the isolation experiments!

   The requests come from OpenCode's built-in vLLM local discovery. By default it checks http://127.0.0.1:8000 every 30 seconds by calling GET /health and then /v1/models, so locally running vLLM models show up automatically. This is intended and listed in the providers docs under Runtimes. But as you found, it keeps probing whatever else is on port 8000, even after repeated 404s, and it never logs the requests.

   You can turn it off now by removing the built-in plugin in your config, then restarting the background service:

   // ~/.config/opencode/opencode.json
   {
     "$schema": "https://opencode.ai/config.json",
     "plugins": ["-opencode.provider.vllm"]
   }

   LM Studio (-opencode.provider.lmstudio, port 1234) and Ollama (-opencode.provider.ollama, port 11434) use the same discovery mechanism and can be removed the same way.

   I'm tracking your suggestions as an enhancement: backing off or stopping after non-2xx responses, logging the probes, and a documented setting to turn local runtime discovery on or off. Maintainers will consider it.

   Related issues:

   • config: add a switch to disable built-in local runtime discovery (vLLM/LM Studio/Ollama) #58094: asked for a first-class setting to disable vLLM, LM Studio and Ollama discovery (closed in favour of this issue)
   • Built-in vLLM discovery causes recurring 401 requests to oMLX on port 8000 #54347: the same vLLM probe causing recurring 401s against an oMLX server on port 8000 (closed in favour of this issue)
   • fix(core): keep configured local models available without discovery #54011: open PR to keep configured local models available without discovery
6. zumik3-del commented on Nov 11, 2026

   

   zumik3-del

   on Nov 11, 2026

   More actions

   Same discovery mechanism, plus one extra facet worth recording: when the probed port is OpenCode's own API, the probe hits itself and then reports its own 401s in its own log.

   Setup: opencode serve --port 1234 --hostname 0.0.0.0 (v2.2.7, Linux x64, systemd) protected by OPENCODE_SERVER_PASSWORD (HTTP Basic auth). LM Studio discovery targets http://127.0.0.1:1234, which is exactly the server's own listen address.

   Observed:

   • GET /api/v1/models exactly every 30.000s, unauthenticated → 401 recorded in opencode's own log (11,078 entries over 9 days before I raised the log level).
   • tcpdump -i lo shows source and destination are the same opencode process (confirmed with ss -tnp: the client socket belongs to the server pid). Headers: Accept: application/json, traceparent, b3, no Authorization, no User-Agent.
   • The same path with valid Basic auth returns 404: /api/v1/* does not exist in the 2.2.7 openapi.json, so the probe can never succeed against OpenCode itself.
   • packages/core/src/plugin/provider/lmstudio.ts (origin: "http://127.0.0.1:1234", getJson("/api/v1/models")) and the 30s interval in local.ts match this exactly.

   Impact is mostly diagnostic: a permanent unauthenticated 401 loop in the server's own log looks like a broken internal client, and it cost me a long time chasing it while investigating an unrelated "TUI hangs" report.

   Two suggestions on top of the ones already raised:

   1. Skip discovery when the target origin resolves to the server's own listen port/address (cheap guard against self-probing).
   2. Treat 401 (auth required) as terminal for that origin rather than retrying every 30s.

   Workaround applied here: "plugins": ["-opencode.provider.lmstudio"]. Verified with tcpdump -i lo that /api/v1/models requests dropped to zero over a 70s window (previously 3 per 70s), with api/info still answering in ~1.4ms.

   😕React with 😕1Reacted by Max Stevens
7. rekram1-node

   marked config: add a switch to disable built-in local runtime discovery (vLLM/LM Studio/Ollama) #58094 as a duplicate of this issue on Nov 13, 2026