[Bug]: GENERATION_FAILED (fp: 018597eb)

[thundertone]

Description

Mac, v0.1.4
Custom endpoint: DeepInfra (https://api.deepinfra.com/v1/openai)
Wire protocol: OpenAI Chat
Every model fails with 422 status code (no body)
Image generation disabled

Steps to reproduce

Custom endpoint: DeepInfra (https://api.deepinfra.com/v1/openai)

Expected behavior

Custom endpoint should work

Actual behavior

Error invoking remote method 'codesign:v1:generate': CodesignError: 422 status code (no body)

open-codesign version

0.1.4

Platform

macOS

OS version

24.6.0

Model provider (if relevant)

None

Error code

GENERATION_FAILED

Diagnostics bundle

Bundle saved locally at ~/Downloads/open-codesign-diagnostics-2026-05-01T13-35-26.zip. Attach it to this issue after submitting.

Logs

[2026-05-01 15:33:32.084] [info]   (main:ipc) [generate] step=send_request {
  runId: 'momye7ml-lsp25e',
  generationId: 'momye7ml-lsp25e',
  provider: 'custom-deepinfra-xa43',
  modelId: 'deepseek-ai/DeepSeek-V4-Flash'
}
[2026-05-01 15:33:32.084] [info]   (main:ipc) agent.turn_start { runId: 'momye7ml-lsp25e', generationId: 'momye7ml-lsp25e' }
[2026-05-01 15:33:32.084] [info]   (main:ipc) [context-prune] step=caps {
  runId: 'momye7ml-lsp25e',
  generationId: 'momye7ml-lsp25e',
  messages: 4,
  before: 227,
  after: 227,
  textLimit: 8192,
  toolInputLimit: 24576,
  toolResultLimit: 8192,
  window: 3
}
[2026-05-01 15:33:32.788] [info]   (main:ipc) agent.turn_end {
  runId: 'momye7ml-lsp25e',
  generationId: 'momye7ml-lsp25e',
  deltas: 0,
  tools: 0
}
[2026-05-01 15:33:32.789] [info]   (main:ipc) agent.end { runId: 'momye7ml-lsp25e', generationId: 'momye7ml-lsp25e' }
[2026-05-01 15:33:32.790] [error]  (main:ipc) [generate] step=send_request.fail {
  runId: 'momye7ml-lsp25e',
  generationId: 'momye7ml-lsp25e',
  provider: 'custom-deepinfra-xa43',
  modelId: 'deepseek-ai/DeepSeek-V4-Flash',
  ms: 706,
  stopReason: 'error'
}
[2026-05-01 15:33:32.790] [error]  (main:ipc) generate.fail {
  runId: 'momye7ml-lsp25e',
  generationId: 'momye7ml-lsp25e',
  ms: 709,
  provider: 'custom-deepinfra-xa43',
  modelId: 'deepseek-ai/DeepSeek-V4-Flash',
  baseUrl: '[url omitted]',
  status: 404,
  message: '404 status code (no body)',
  code: 'PROVIDER_ERROR'
}
[2026-05-01 15:33:32.793] [error]  (renderer:console) [store] applyGenerateError {"generationId":"momye7ml-lsp25e","designId":"6f177afb-d9fe-42b6-ab08-641de07734f4","message":"Error invoking remote method 'codesign:v1:generate': CodesignError: 404 status code (no body)"} {}
[2026-05-01 15:33:32.795] [info]   (chat-messages-ipc) chat.append {
  designId: '6f177afb-d9fe-42b6-ab08-641de07734f4',
  seq: 7,
  kind: 'error'
}

[github-actions]

Thanks for the report @thundertone, and for including logs!

Looking at the logs you provided, the error for the DeepSeek model attempt is actually a 404 (not 422). The request URL appears to be constructed correctly from your base URL (https://api.deepinfra.com/v1/openai), so a 404 from DeepInfra typically means the model ID is not found on their platform.

deepseek-ai/DeepSeek-V4-Flash does not appear to be a real model (DeepSeek V4 doesn’t exist yet – current models are V3, R1, etc.). That would explain the 404.

Next steps:

1. Use the Test Connection button in the provider settings (under the Models tab) to validate your endpoint and model list. DeepInfra returns a list of valid models when probed correctly.
2. Pick a model from DeepInfra’s supported text-generation models – for example deepseek-ai/DeepSeek-R1 or meta-llama/Llama-3.3-70B-Instruct.
3. If you still see 422 for other models (the description mentions that), the error may be due to message validation issues (e.g., missing required fields in the request). Please share the diagnostics bundle (attached to the issue) and any Test Connection output so we can narrow it down.

Relevant code context (v0.1.4):

• The app uses pi-ai (via @mariozechner/pi-ai) to construct requests to OpenAI-compatible endpoints. For custom providers with OpenAI Chat wire, it appends /chat/completions to your base URL.
• The generate-failure-diagnose-bridge changeset added diagnostics for 404s in v0.1.4 that might suggest checking the base URL – you can ignore that if your base URL already contains /v1 (it does).

Let us know what you find after testing a valid model!

---

Open-CoDesign Bot

[thundertone]

I believe the 404 only happened when testing the other OpenAI Wire Protocol.

• Connection test succeeds and consumes tokens
• Generation immediately fails with 422
• Therefore the issue is assumed to be in the tool-calling payload, not auth or basic connectivity
• deepseek-ai/DeepSeek-V4-Flash indeed exists