core: LM Studio model discovery fails silently with decision-type models #56213

Activity

1. opencode-agent commented on Oct 16, 2026

   

   opencode-agentBot

   on Oct 16, 2026 – with OpenCode Agent

   Contributor

   More actions

   Thanks for the report. I reproduced this on 2.0.29 and on the latest release, 2.0.31.

   I used a mock LM Studio server on port 1234 whose /api/v1/models returns one LLM (qwen/qwen3-8b) plus one model with "type": "decision" (clef-flash):

   • With the decision model in the response, opencode models shows no lmstudio/... models and no error, even across several 30s discovery refreshes.
   • When the decision entry is removed, lmstudio/qwen/qwen3-8b appears after the next refresh.

   The LM Studio discovery code only accepts the model types llm and embedding, and it validates the whole response before filtering out non-LLM models. One decision model therefore makes the whole response fail, and the error is swallowed. That code is unchanged on the current dev branch (bc5ebf0), so it should behave the same there. I couldn't run dev from source to confirm. Open PR #56214 targets this.

   The script below reproduces it.

   Reproduction script (repro.sh)

   #!/usr/bin/env bash
   # Reproduces #56213: a single decision-type model in LM Studio's
   # /api/v1/models response hides every LM Studio model from `opencode models`.
   # Usage: ./repro.sh                                  (uses `opencode` from PATH)
   #        OC_DEV=/path/to/opencode/checkout ./repro.sh (runs from source)
   # Needs: bun, curl, port 1234 free. Kills any running `opencode serve --service`
   # so discovery starts from scratch.
   set -u
   DIR=$(mktemp -d)
   cd "$DIR"
   
   run_opencode() {
     if [ -n "${OC_DEV:-}" ]; then
       bun --conditions=browser --preload="$OC_DEV/packages/tui/node_modules/@opentui/solid/scripts/preload.js" "$OC_DEV/packages/cli/src/index.ts" "$@"
     else
       opencode "$@"
     fi
   }
   
   cat > server.ts <<'EOF'
   import { readFileSync } from "node:fs"
   const llm = {
     type: "llm", key: "qwen/qwen3-8b", display_name: "Qwen3 8B", architecture: "qwen3",
     loaded_instances: [], max_context_length: 32768,
     capabilities: { vision: false, trained_for_tool_use: true },
   }
   const decision = {
     type: "decision", key: "clef-flash", display_name: "Clef Flash",
     loaded_instances: [], max_context_length: 8192,
   }
   Bun.serve({
     port: 1234, hostname: "127.0.0.1",
     fetch(req) {
       const url = new URL(req.url)
       const withDecision = readFileSync("mode", "utf8").trim() === "1"
       console.error(new Date().toISOString(), "mock:", req.method, url.pathname, "decision =", withDecision)
       if (url.pathname === "/api/v1/models")
         return Response.json({ models: withDecision ? [llm, decision] : [llm] })
       return new Response("not found", { status: 404 })
     },
   })
   EOF
   
   check() {
     # discovery runs in the background service and refreshes every 30s; poll for ~40s
     for i in $(seq 1 8); do
       out=$(run_opencode models 2>&1 | grep -i '^lmstudio/' || true)
       echo "  [$(date -u +%T)] lmstudio models: ${out:-<none>}"
       sleep 5
     done
   }
   
   pkill -f 'opencode.* serve --service' 2>/dev/null; sleep 1
   echo 1 > mode
   bun server.ts 2>server.log &
   PID=$!
   until curl -sf http://127.0.0.1:1234/api/v1/models >/dev/null; do sleep 0.2; done
   
   echo "=== LM Studio returns 1 llm + 1 decision model ==="
   check
   echo 0 > mode
   echo "=== LM Studio returns only the llm model ==="
   check
   
   kill $PID
   echo "--- mock server log:"; cat server.log

   Run it: bash repro.sh
2. opencode-agent

   added

   reproduced

   on Oct 16, 2026
3. rekram1-node

   closed this as completedin #56214on Nov 10, 2026