fix(app): save custom OpenAI-compatible providers by css521 · Pull Request #57997 · anomalyco/opencode

css521 · 2026-10-22T02:35:22Z

Issue for this PR

Type of change

Bug fix
New feature
Refactor / code improvement
Documentation

What does this PR do?

Restore custom OpenAI-compatible provider setup in the current integration picker and fix the failed submission reported in #50650. Save the provider definition in global JSONC while preserving other settings. Store literal API keys through the credential API and clean them up if the config write fails.

Wait for the configured models to load before reporting a connection. Rejected writes keep the form available to retry; saved configurations whose models do not load get a separate message.

How did you verify your code works?

bun run check: lint and all 35 package checks passed. App and client bun run typecheck:tsc also passed.
App form/save tests: 7 passed. Server config tests: 3 passed, including JSONC preservation, credential binding and reloading an already-open directory.
Provider browser cases: 15 passed. The 3 custom-provider cases passed 30/30 repetitions on a fresh build with 5 workers and no retries.
bun run check:generated: regeneration produces no diff.

The fully parallel settings suite passed 46/53. All seven shortcut/activity-layout failures also reproduced on unmodified dev at 86ce19d96e.

Visual

<svg xmlns="http://www.w3.org/2000/svg" width="240" height="72" viewBox="0 0 240 72" fill="none" role="img" aria-label="Save a custom provider and load its models">
  <style>
    .flow { animation: flow 1.8s linear infinite; }
    .ready { animation: ready 3.6s ease-in-out infinite; }
    @keyframes flow { to { stroke-dashoffset: -18; } }
    @keyframes ready { 0%, 100% { opacity: 1; } 50% { opacity: .45; } }
    @media (prefers-reduced-motion: reduce) { .flow, .ready { animation: none; } }
  </style>
  <g stroke="currentColor" stroke-width="1.5" stroke-linecap="round" stroke-linejoin="round">
    <rect x="8" y="10" width="74" height="52" rx="7" opacity=".45"/>
    <path d="M19 23h31M19 33h51M19 43h26" opacity=".55"/>
    <rect x="55" y="43" width="15" height="9" rx="3" opacity=".75"/>
    <path class="flow" d="M94 36h36" stroke-dasharray="4 5" opacity=".7"/>
    <path d="m124 30 6 6-6 6" opacity=".7"/>
    <rect x="146" y="16" width="86" height="40" rx="7" opacity=".45"/>
    <path d="M182 30h37M182 40h27" opacity=".6"/>
    <g class="ready">
      <circle cx="163" cy="36" r="8" opacity=".6"/>
      <path d="m159 36 3 3 6-6"/>
    </g>
  </g>
</svg>

Screenshots / recordings

Captured with local test fixtures. The before image shows the reported failure; the after image shows the saved model in the current picker.

Reported failure	Current model picker

Checklist

I have tested my changes locally
I have not included unrelated changes in this PR

Added by the opencode bot (AI-generated).

Related: #51527, #57365, #51031, #57325

Public API: makes shell optional and adds providers to the PATCH /api/experimental/config patch (and config.update in @opencode/client); backwards compatible

opencode-agent

The server side of the fix works. Run from source, the base rejects a providers-only config patch with 400 (Missing key ["shell"]), and the form always throws provider.custom.unavailable. With this change the same credential + config patch writes the provider to the global JSONC (comments and other keys kept, no key in the file), and the provider and model show up as enabled in /api/provider and /api/model within a few seconds. The credentials tests and the server config test pass, and app, core, schema and client typecheck. I didn't drive the desktop/web form itself, so the UI part relies on your screenshots and the unit tests.

Public API impact, all backwards compatible:

Config.Patch / PATCH /api/experimental/config: shell goes from required to optional. A new optional providers record (ConfigProvider.Info by ID) replaces each named entry in the global document.
@opencode/client: config.update(input) becomes config.update(input?), and ConfigUpdateInput gains providers. The effect client now exports ConfigProvider types.
Existing callers that send { shell } keep working.

Concerns are inline: the 10s polling loop in the form, duplicate credentials when retrying after a 'saved but not loaded' result, and the size and timing dependence of the new server test. For maintainers: #52653 took the same approach (native provider config plus a separate credential) and was closed without a comment, so whether this design is wanted is worth confirming.

Related: #51527, #57365, #51031, #57325

opencode-agent · 2026-10-22T02:39:06Z

    return output.result
  }

+  const refresh = async (providerID: string, modelIDs: string[]) => {


The form re-reads the catalog every 250 ms for up to 10s, waiting for the server's debounced reload after the write. This ties the UI to server reload timing: a slow or large config reload shows 'saved, but its models could not be loaded' even though nothing is wrong. Could Config.update wait for the reload it requests (or the endpoint answer only once the new config is applied)? Then the form could refresh once instead of polling.

opencode-agent · 2026-10-22T02:39:06Z

+ */
+export async function saveCustomProvider(api: ServerApi, provider: CustomProvider) {
+  const credential = provider.key
+    ? await api.credential.create({


When the write succeeds but the models never load, the dialog closes and the provider is not in data.location.provider.list(). If the user opens the form again with the same ID, the duplicate check passes, a second credential is created for the same integration and the config entry is overwritten. Could the save reuse or replace an existing credential for providerID, or the duplicate check also look at the config/integration?

opencode-agent · 2026-10-22T02:39:06Z

 )

+it.live(
+  "saves a custom provider that becomes a selectable provider and model",


This test is about 160 lines, runs the real file watcher and polls for up to 15s across several endpoints, so it is slow and timing-dependent. The core behaviour is "a providers patch keeps unrelated JSONC and replaces just the named entry", which could be a short test like the existing shell one. The catalog/integration round-trip could be dropped or cut down to one check.

fix(app): save custom OpenAI-compatible providers

1ecac84

css521 requested review from Brendonovich and Hona as code owners October 22, 2026 02:35

opencode-agent Bot reviewed Oct 22, 2026

View reviewed changes

css521 mentioned this pull request Oct 22, 2026

desktop: custom provider save always throws "unavailable on this server" #50650

Open

css521 added 2 commits October 28, 2026 09:45

chore: resolve custom provider merge conflicts

124c30d

chore: merge dev for custom provider compatibility

5e72e77

Ancienttwo mentioned this pull request Nov 8, 2026

[Downstream] keys: OpenCode Go 的 key check 不带 x-opencode-session，返回 400 MissingSessionID，被归为 provider_error Ancienttwo/byok-sdk#1547

Closed

opencodejr Bot mentioned this pull request Nov 8, 2026

Desktop app is a pure garbage with no functionality - why make gui what don't even proviede a way to make a custom endpoint from it? #62752

Closed

joe-ieta mentioned this pull request Nov 14, 2026

[FEATURE]: Feature Request: Add GUI Options to Edit Custom Providers and Manage Model Lists #33704

Open

1 task done

Navigation Menu

Repository navigation

fix(app): save custom OpenAI-compatible providers - #57997

fix(app): save custom OpenAI-compatible providers#57997
css521 wants to merge 3 commits into
anomalyco:devfrom
css521:custom-provider-save

css521 commented Oct 22, 2026 •

edited

Loading

Uh oh!

opencode-agent Bot left a comment

Uh oh!

opencode-agent Bot Oct 22, 2026

Uh oh!

opencode-agent Bot Oct 22, 2026

Uh oh!

opencode-agent Bot Oct 22, 2026

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

css521 commented Oct 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Issue for this PR

Type of change

What does this PR do?

How did you verify your code works?

Visual

Screenshots / recordings

Checklist

Uh oh!

opencode-agent Bot left a comment

Choose a reason for hiding this comment

Uh oh!

opencode-agent Bot Oct 22, 2026

Choose a reason for hiding this comment

Uh oh!

opencode-agent Bot Oct 22, 2026

Choose a reason for hiding this comment

Uh oh!

opencode-agent Bot Oct 22, 2026

Choose a reason for hiding this comment

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

css521 commented Oct 22, 2026 •

edited

Loading