config: compatibility.supportsStore silently dropped from opencode.jsonc #52617

Activity

1. github-actions

   assigned

   neriousy

   on Oct 2, 2026
2. github-actions commented on Oct 2, 2026

   

   github-actionsBot

   on Oct 2, 2026 – with GitHub Actions

   Contributor

   More actions

   This issue appears to follow the bug-report template and includes substantive description, reproduction steps, and environment details.

   ---

   This issue may be related to existing reports about unsupported OpenAI-compatible request fields and compatibility controls. Please check:

   • provider: v2 always sends prompt_cache_key to OpenAI-compatible providers; setCacheKey option ignored #45113: OpenAI-compatible requests always send prompt_cache_key while the setCacheKey opt-out is ignored.
   • providers: nvidia rejects unsupported prompt_cache_key sent by V2 (OpenAI-compatible runtime) #49240: NVIDIA rejects the unsupported prompt_cache_key sent by the OpenAI-compatible runtime.
3. mperreir commented on Oct 5, 2026

   

   mperreir

   on Oct 5, 2026

   More actions

   Hello,
   Until this bug is fixed, here is a workaround using the plugin idea you proposed (it works correctly on my side).
   strip-store.ts

   🚀React with 🚀3Reacted by plenka, Gabriel Araujo and Gary Bowen
4. fihaaade

   added 2 commits that reference this issue on Oct 13, 2026

   fix(schema): keep openai-compatible compatibility flags from config

   ...

   This issue will close once commit 8736a25 is merged into the 'dev' branch.

   8736a25

   fix(schema): keep openai-compatible compatibility flags from config

   ...

   This issue will close once commit 4024e6e is merged into the 'dev' branch.

   4024e6e
5. fihaaade commented on Oct 24, 2026

   

   fihaaade

   on Oct 24, 2026

   Contributor

   More actions

   This should be fixed on dev by #57239: Chat Completions now only sends store to OpenAI and Azure by default, so strict gateways such as LiteLLM no longer get "store": false. I think this can be closed once a release includes it.
6. opencodejr commented on Oct 28, 2026

   

   opencodejrBot

   on Oct 28, 2026

   More actions

   Thanks for the detailed report. I reproduced this, and part of it is already fixed on dev.

   I pointed two openai-compatible providers at a local endpoint that records request bodies. Both used your model config (compatibility: { supportsStore: false, maxTokensField: "max_tokens" }). One baseURL looks like a LiteLLM gateway; the other contains .openai.azure.com, so OpenCode's built-in detection treats it as Azure OpenAI.

   • 2.0.21: every request for both providers contains "store": false, while max_tokens is used. This matches what you described.
   • Latest dev (891a020): the gateway-style provider no longer gets store. Since v2.1.0, store is only sent by default to OpenAI and Azure endpoints (fix(ai): only send store on chat completions for OpenAI and Azure #57239), so upgrading should fix the LiteLLM UnsupportedParamsError for you. The setting itself is still ignored, though: the Azure-looking provider still gets "store": false despite supportsStore: false. The config schema has no supportsStore field, so the key is dropped when the config loads, as you found.

   The script below reproduces it. Related:

   • fix(ai): only send store on chat completions for OpenAI and Azure #57239: the merged change to the default store behaviour
   • fix(schema): keep openai-compatible compatibility flags from config #53890: a closed PR that added supportsStore to the config schema
   • provider: openai-compatible always injects stream_options and store, breaking non-OpenAI-compatible APIs #57230: a report of the same problem, closed as a duplicate of this one

   Reproduction script (repro.sh)

   #!/usr/bin/env bash
   # Reproduces #52617: model-level compatibility.supportsStore in opencode.jsonc is
   # dropped at config load, so the openai-compatible chat protocol ignores it.
   # Usage: ./repro.sh            -> runs `opencode` from PATH
   #        OC_DEV=/path/to/repo ./repro.sh -> runs the CLI from a source checkout
   set -euo pipefail
   PORT=${PORT:-4517}
   WORK=$(mktemp -d /tmp/opencode/repro52617.XXXX)
   mkdir -p "$WORK/project" "$WORK/home"
   BODIES="$WORK/bodies.jsonl"
   
   # Probe endpoint: records each request body and answers with a minimal SSE stream.
   cat > "$WORK/probe.ts" <<EOF
   const out = Bun.file("$BODIES").writer()
   Bun.serve({ port: $PORT, async fetch(req) {
     const url = new URL(req.url)
     if (!url.pathname.endsWith("/chat/completions")) return new Response("{}", { headers: { "content-type": "application/json" } })
     const body = await req.json()
     out.write(JSON.stringify({ path: url.pathname, body }) + "\n"); out.flush()
     const chunk = (o: unknown) => "data: " + JSON.stringify(o) + "\n\n"
     const base = { id: "x", object: "chat.completion.chunk", created: 0, model: body.model }
     const sse = chunk({ ...base, choices: [{ index: 0, delta: { role: "assistant", content: "ok" }, finish_reason: null }] })
       + chunk({ ...base, choices: [{ index: 0, delta: {}, finish_reason: "stop" }], usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 2 } })
       + "data: [DONE]\n\n"
     return new Response(sse, { headers: { "content-type": "text/event-stream" } })
   } })
   EOF
   bun "$WORK/probe.ts" & PROBE=$!
   trap 'kill $PROBE 2>/dev/null || true' EXIT
   sleep 1
   
   # "proxy": a LiteLLM-like gateway URL. "azproxy": URL containing ".openai.azure.com"
   # (in the path) so built-in detection says store IS supported; supportsStore:false
   # in config must override that.
   cat > "$WORK/project/opencode.jsonc" <<EOF
   {
     "providers": {
       "proxy": {
         "name": "Proxy",
         "package": "@opencode/ai/providers/openai-compatible",
         "settings": { "baseURL": "http://127.0.0.1:$PORT/v1", "apiKey": "x" },
         "models": { "claude-sonnet-5-5": { "modelID": "claude-sonnet-5-5",
           "compatibility": { "supportsStore": false, "maxTokensField": "max_tokens" } } }
       },
       "azproxy": {
         "name": "AzProxy",
         "package": "@opencode/ai/providers/openai-compatible",
         "settings": { "baseURL": "http://127.0.0.1:$PORT/.openai.azure.com/v1", "apiKey": "x" },
         "models": { "claude-sonnet-5-5": { "modelID": "claude-sonnet-5-5",
           "compatibility": { "supportsStore": false, "maxTokensField": "max_tokens" } } }
       }
     }
   }
   EOF
   
   run() {
     if [ -n "${OC_DEV:-}" ]; then
       # the standalone server is spawned as a child bun process; pass the same flags to it
       BUN_OPTIONS="--conditions=browser --preload=$OC_DEV/packages/tui/node_modules/@opentui/solid/scripts/preload.js" 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
   }
   cd "$WORK/project"
   export XDG_CONFIG_HOME="$WORK/home/config" XDG_DATA_HOME="$WORK/home/data" XDG_STATE_HOME="$WORK/home/state" XDG_CACHE_HOME="$WORK/home/cache"
   echo "version: $(run --version)"
   for m in proxy/claude-sonnet-5-5 azproxy/claude-sonnet-5-5; do
     run run --standalone -m "$m" "say ok" >/dev/null 2>"$WORK/err.log" || { echo "run $m failed"; tail -20 "$WORK/err.log"; }
   done
   echo "--- captured request bodies (path, store present?, store value, max_tokens field) ---"
   bun -e '
   for (const l of (await Bun.file(process.argv[1]).text()).trim().split("\n").filter(Boolean)) {
     const { path, body } = JSON.parse(l)
     console.log(path, "| store:", "store" in body ? JSON.stringify(body.store) : "<omitted>",
       "| max_tokens:", body.max_tokens !== undefined, "| max_completion_tokens:", body.max_completion_tokens !== undefined)
   }' "$BODIES"

   Run it: bash repro.sh
7. opencodejr

   added

   reproduced

   on Oct 28, 2026