[Bug]: GENERATION_FAILED (fp: 7d688101)

[ray73864]

Description

Trying to use it with a local llama.cpp instance.

Steps to reproduce

I added my local llama.cpp as a custom provider, then went to start prompting for a design to be created.

Expected behavior

It should not have errored.

Actual behavior

Error invoking remote method 'codesign:v1:generate': CodesignError: preferences.json is invalid: unsupported field "proxyUrl"

open-codesign version

0.2.0

Platform

Windows

OS version

Windows 11 (10.0.26200)

Model provider (if relevant)

Other

Error code

GENERATION_FAILED

Diagnostics bundle

Bundle saved locally at D:\Downloads\open-codesign-diagnostics-2026-08-15T04-17-28.zip. Attach it to this issue after submitting.

Logs

[2026-08-15 12:16:20.659] [warn]   (renderer:console) [i18n] missing translation key "settings.providers.custom.allowPrivateNetwork" in namespace "translation" for locale "en" {}
[2026-08-15 12:16:20.771] [warn]   (renderer:console) [i18n] missing translation key "settings.providers.custom.allowPrivateNetwork" in namespace "translation" for locale "en" {}
[2026-08-15 12:16:22.709] [warn]   (renderer:console) [i18n] missing translation key "settings.providers.custom.allowPrivateNetwork" in namespace "translation" for locale "en" {}
[2026-08-15 12:16:23.325] [warn]   (renderer:console) [i18n] missing translation key "settings.providers.custom.allowPrivateNetwork" in namespace "translation" for locale "en" {}
[2026-08-15 12:16:24.523] [warn]   (renderer:console) [i18n] missing translation key "settings.providers.custom.allowPrivateNetwork" in namespace "translation" for locale "en" {}
[2026-08-15 12:16:45.888] [info]   (main:ipc) [generate] step=load_config { runId: 'mstv6hx2-fhgdta', generationId: 'mstv6hx2-fhgdta' }
[2026-08-15 12:16:45.889] [info]   (main:ipc) [generate] step=load_config.ok {
  runId: 'mstv6hx2-fhgdta',
  generationId: 'mstv6hx2-fhgdta',
  ms: 0,
  hasApiKey: true,
  baseUrl: '[url omitted]'
}
[2026-08-15 12:16:45.889] [info]   (main:ipc) [generate] step=validate_provider {
  runId: 'mstv6hx2-fhgdta',
  generationId: 'mstv6hx2-fhgdta',
  provider: 'custom-llama-cpp-2hfj',
  modelId: 'gemma-4-26b'
}
[2026-08-15 12:16:45.889] [info]   (main:ipc) [generate] step=validate_provider.ok {
  runId: 'mstv6hx2-fhgdta',
  generationId: 'mstv6hx2-fhgdta',
  provider: 'custom-llama-cpp-2hfj'
}
[2026-08-15 12:16:45.891] [error]  (renderer:console) [store] applyGenerateError {"generationId":"mstv6hx2-fhgdta","designId":"529fa649-daf1-43b3-b96b-da3485dc16b5","message":"Error invoking remote method 'codesign:v1:generate': CodesignError: preferences.json is invalid: unsupported field \"proxyUrl\""} {}
[2026-08-15 12:17:27.186] [warn]   (main:ipc) provider.error.final {
  runId: '8cc5c016-6fe2-4a31-a42e-71d97e5762b1',
  upstream_provider: 'custom-llama-cpp-2hfj',
  upstream_code: 'PROVIDER_ERROR',
  upstream_message: 'Provider stopped before completion because the response hit the token limit',
  retry_count: 0,
  original_error_name: 'CodesignError'
}
[2026-08-15 12:17:27.187] [error]  (main:ipc) [title] step=send_request.fail {
  runId: '8cc5c016-6fe2-4a31-a42e-71d97e5762b1',
  ms: 159377,
  errorClass: 'CodesignError'
}
[2026-08-15 12:17:27.187] [error]  (main:ipc) [title] generate-title.fail {
  runId: '8cc5c016-6fe2-4a31-a42e-71d97e5762b1',
  provider: 'custom-llama-cpp-2hfj',
  modelId: 'qwen3.6-35b-a3b',
  baseUrl: '[url omitted]',
  message: 'Provider stopped before completion because the response hit the token limit',
  code: 'PROVIDER_ERROR'
}
[2026-08-15 12:17:27.191] [warn]   (renderer:store) [title] generateTitle failed, using prompt fallback {
  designId: '529fa649-daf1-43b3-b96b-da3485dc16b5',
  message: "Error invoking remote method 'codesign:v1:generate-title': CodesignError: Provider stopped before completion because the response hit the token limit"
}

[github-actions]

Thanks for the detailed report. The key error is from validation of your local config:

CodesignError: preferences.json is invalid: unsupported field "proxyUrl"

The error means the preferences.json file that stores app preferences contains a proxyUrl key on your custom provider (custom-llama-cpp-2hfj), but the preferences schema in v0.2.0 does not accept that field. The generation IPC (codesign:v1:generate) validates preferences before it sends the request, so generation fails before llama.cpp is even called.

Workaround for now:

1. Close Open CoDesign.
2. Open your app data's preferences.json (the file referenced in the error).
3. Find the custom-llama-cpp-2hfj provider entry and remove the "proxyUrl": ... line.
4. Save, restart, and try again.

For a local llama.cpp instance you normally don't need a proxy; keep baseUrl set directly to your llama.cpp server (e.g. http://127.0.0.1:8080/v1) and leave the API key blank or set it to whatever your server expects.

The missing settings.providers.custom.allowPrivateNetwork translation strings in the log look like a separate cosmetic i18n gap, but they may be related if that toggle is what writes the proxyUrl field. I wasn't able to verify the exact writing path from the public checkout, so it's worth confirming whether the field appears after you toggle that option.

Need more info (if it still fails after removing proxyUrl):

• Redacted custom provider JSON from preferences.json (mask the API key and base URL).
• Does "Test Connection" pass for the custom provider?
• If possible, attach the diagnostics bundle open-codesign-diagnostics-2026-08-15T04-17-28.zip.

Separately, the later Provider stopped before completion because the response hit the token limit error is a different failure — that happens when llama.cpp stops a response mid-stream. If you still see that after the config issue is fixed, it usually means the server's max tokens/context is too small for the prompt; raising n_predict (or context size) on the llama.cpp side should help.

Thanks again for the report — happy to dig further once we know whether removing proxyUrl resolves it.

---

Open-CoDesign Bot