Error: 4 of 5 requests failed: config.providers: Unexpected server error.

[chrissound]

Description

When starting the app via opencode executable:

Error: 4 of 5 requests failed: config.providers: Unexpected server error. Check server logs for details.; provider.list: Unexpected server error. Check server logs for details.; app.agents: Unexpected server error. Check server logs for details.; config.get: Unexpected server error. Check server logs for details.
    at qG (/$bunfs/root/chunk-gt0me4ve.js:469:625)
    at <anonymous> (/$bunfs/root/chunk-gt0me4ve.js:469:6032)
    at processTicksAndRejections (native:7:39)

Plugins

No response

OpenCode version

(version: 1.14.49)

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response

[Smurf81]

same here after i shutdown opencode with "Ctrl+C" when he was in "Building"

ERROR 2026-05-15T12:02:02 +2ms service=server error=ConfigInvalidError cause=ConfigInvalidError: ConfigInvalidError
    at zj (/$bunfs/root/chunk-k835vey5.js:710:82)
    at <anonymous> (/$bunfs/root/chunk-k835vey5.js:712:8940)
    at Config.loadInstanceState (/$bunfs/root/chunk-k835vey5.js:714:5025)
    at Config.loadInstanceState (definition) (/$bunfs/root/chunk-k835vey5.js:714:81)
    at Config.state (/$bunfs/root/chunk-k835vey5.js:675:20296)
    at Config.state (definition) (/$bunfs/root/chunk-k835vey5.js:714:4981)
    at Config.get (/$bunfs/root/chunk-nxmfg3zm.js:37:6483)
    at Config.get (definition) (/$bunfs/root/chunk-k835vey5.js:714:5052)
    at InstanceBootstrap (/$bunfs/root/chunk-nxmfg3zm.js:37:6712)
    at InstanceStore.boot (/$bunfs/root/chunk-ar3awd7v.js:2:1490)
    at InstanceStore.load (/$bunfs/root/chunk-ar3awd7v.js:2:2651) failed
``̀

[chrissound]

A workaround solution... Because I keep being unable to start opencode for past weeks.. Is just run HOME=/tmp/opencodetmp123 opencode... Of course it'll not persist your sessions etc using this.

[Smurf81]

A workaround solution... Because I keep being unable to start opencode for past weeks.. Is just run HOME=/tmp/opencodetmp123 opencode... Of course it'll not persist your sessions etc using this.

same error for me

Try to uninstall & re-install but same thing .....

FIX : in opencode.jsonc i was using

  "agents": {
    "context": [
      "./AGENTS.md"
    ]
  }

this is not working anymore and crash opencode on version 1.15.0, need to change to

"instructions": ["AGENTS.md"]

[Qesire]

Same issue comes out of me trying start the Opencode in VSCode after my accidentally closing the bash when Opencode building something directly. I've tried reinstall, upgrade but effortless.

[Zsolt-R]

Same, accidentally closed the terminal while it was building now I am getting the same error.

[gaoxingyi1024]

1、uninstall opencode
2、remove these files below
3、install opencode

%USERPROFILE%.cache\opencode
%USERPROFILE%.local\share\opencode
%USERPROFILE%.config\opencode

[SimplyJoshi]

Root cause: auth.json corrupted with whitespace-only contents

I hit this on Windows 11 with opencode 1.15.5 and traced it to a specific failure mode that I think explains every "Ctrl+C while building" / "closed the terminal mid-build" report above. Posting the diagnosis in case it helps the maintainers and the next person who lands here.

What the server is actually crashing on

The TUI-facing error (4 of 5 requests failed: config.providers, provider.list, app.agents, config.get) is downstream noise. The real failure in the server log is a JSON parse error inside Auth.all():

ERROR ... service=server error=JSON Parse error: Unrecognized token ' '
    cause=SyntaxError: JSON Parse error: Unrecognized token ' '
    at FileSystem.readJson (chunk-y6vv15b7.js)
    at Auth.all (chunk-y6vv15b7.js)
    at Config.loadInstanceState (chunk-y6vv15b7.js)
    at Config.state (chunk-y6vv15b7.js)
    at Config.get (chunk-hqfz04v1.js)

Every config-dependent request goes through Config.get -> Config.loadInstanceState -> Auth.all -> readJson(auth.json). When that parse throws, every one of those requests fails with the same generic "Unexpected server error", which is why the user sees four failures instead of one. Logs live at ~/.local/share/opencode/log/*.log (or %USERPROFILE%\.local\share\opencode\log\ on Windows) — searching them for Auth.all surfaces the cause immediately.

What the files looked like

On my machine, both auth.json files existed and were non-empty, but contained only whitespace — no JSON, not a single {:

Path	Size	Contents
%LOCALAPPDATA%\opencode\auth.json	327 bytes	327 space characters
%USERPROFILE%\.local\share\opencode\auth.json	543 bytes	543 space characters

Both files had identical modification timestamps down to the second.

Why this happens

The byte counts strongly suggest a non-atomic write. The file is opened with O_TRUNC, the buffer is padded with spaces, and the actual JSON payload is supposed to overwrite that padding before the file is closed — but if the process is killed (Ctrl+C, terminal closed, OS shutdown) before the write completes, the file is left at its full size, filled with 0x20. The filesystem still sees a "valid" non-empty file, so nothing flags it as broken.

This matches every report above where the problem started after interrupting opencode while it was "Building". That's the window in which the auth file is being rewritten.

The parser then fails fatally instead of treating an unreadable auth file as "no credentials configured", which traps the user: they cannot open the TUI to run auth login, because the TUI refuses to bootstrap without a parseable auth file. Catch-22.

Recovery (no reinstall needed)

Overwrite both files with an empty JSON object, then re-authenticate. This worked instantly on my machine and preserves sessions, config, plugins, etc.

Windows (PowerShell):

Set-Content "$env:LOCALAPPDATA\opencode\auth.json" '{}'
Set-Content "$env:USERPROFILE\.local\share\opencode\auth.json" '{}'
opencode auth login

macOS / Linux:

echo '{}' > "$HOME/.local/share/opencode/auth.json"
# also check $XDG_DATA_HOME/opencode/auth.json if you've overridden it
opencode auth login

The "uninstall + nuke .cache, .config, .local/share" workaround posted earlier works because it deletes the corrupted file along with everything else. The targeted fix is just the two auth.json files — no need to lose your sessions, projects, or installed plugins.

Suggested fixes

1. Atomic write for auth.json — write to auth.json.tmp and rename() over auth.json. POSIX rename and Windows MoveFileEx with MOVEFILE_REPLACE_EXISTING are both atomic at the filesystem level: an interrupted process either leaves the old file intact or replaces it with the fully-written new one, never a half-written state. This eliminates the failure mode entirely.
2. Lenient parse on read — if auth.json is empty, whitespace-only, or otherwise malformed, log a warning and treat it as {} rather than crashing the bootstrap. The TUI should always start; an unparseable auth file is a "no providers configured" condition, not a fatal one. Users can then run opencode auth login to recover from inside the app.
3. Surface the real cause — the user-facing 4 of 5 requests failed: config.providers... message doesn't mention auth.json or the underlying SyntaxError. Including the offending file path and the parse error in the bootstrap error would let users self-diagnose without grepping log files.

Environment

• opencode 1.15.5
• Windows 11 Pro (build 26200)
• Bun runtime (bundled)

Happy to test a patch or attach more log detail if it helps.

[AlexMikhalev]

Detailed investigation with root cause analysis

I hit this exact crash on v1.15.5 (Linux) and traced it to a provider API returning 5xx during startup, which causes opencode to crash rather than degrade gracefully.

Crash log (`/tmp/opencode-debug.log`)

```
Error: 2 of 5 requests failed: Unexpected server error. Check server logs for details.
Affected startup requests: config.providers, provider.list
at jO (/$bunfs/root/chunk-5pecsz7x.js:470:117)
```

Root cause

During startup opencode fires 5 parallel requests. Two failed:

1. `config.providers` -- validates provider configurations by initialising each provider
2. `provider.list` -- fetches available models via the plugin's `provider.models` hook

Both hit the provider API (in my case Kimi API at `https://api.kimi.com/coding/v1\`) which returned a 5xx server error.

Why it crashes instead of degrading

The `opencode-kimi-full` plugin (v1.4.0) has a `loader` hook that runs when opencode instantiates the provider. It calls `ensureFresh()` which:

1. Reads auth from `~/.local/share/opencode/auth.json`
2. Checks if token is expiring (60s safety window)
3. If so, calls `refreshToken()` against the OAuth endpoint
4. Calls `listModels()` against the provider API

The `provider.models` hook also calls `listModels()` to discover model metadata. While the plugin has try-catch fallbacks, the error is thrown at a higher level in opencode's startup orchestration before those hooks can gracefully degrade.

Evidence from session database

The SQLite session database (`~/.local/share/opencode/opencode.db`) shows:

• Last successful session: 08:37 BST
• Crash time: 10:16 BST (from debug log timestamp)
• Next session: 20:44 BST (8 hours later)

The crash blocked all opencode usage for ~10.5 hours. The auth token was valid (expires 21 May 10:20 UTC), so this was not a token expiry issue.

Reproduction

1. Configure a provider that uses a plugin with a `loader` hook (e.g. `opencode-kimi-full`)
2. Make the provider API return 5xx (or have a real outage)
3. Start opencode -- it crashes with the error above

Suggested fixes

1. Lazy provider initialisation -- don't validate providers at startup; initialise them on first use
2. Graceful degradation -- if a provider fails during startup, mark it as unavailable and continue with remaining providers
3. Retry with backoff -- add retry logic for transient 5xx errors during startup
4. Better error messages -- surface which specific provider failed and why, rather than the generic "Unexpected server error"

Environment

• opencode v1.15.5
• Linux (Pop!_OS)
• Plugin: opencode-kimi-full v1.4.0
• Provider: kimi-for-coding-oauth (OpenAI-compatible via @ai-sdk/openai-compatible)

[czmhwyl]

This is because the code format was incorrect when configuring the other API call path. The solution is to clear the opencode.json and auth.json files.

[thaumkid]

I noticed a similar issue crops up when there is whitespace in my startup path, possibly with certain plugins enabled. seems one solution is to never have spaces in any important paths.