Silent fail with incorrect AskSage URL

[rljacob]

Pre-submit Checks

• I have searched Warp bugs and there are no duplicates
• I have searched Warp known issues page and my issue is not there
• I have an issue with AI and have included the debugging ID (Optional, but helps expedite the AI quality fix). Debugging ID instructions
• I have technical issue and have included the logs (optional, but helps expedite the bug fix). Log instructions

Describe the bug

While trying to set up a Custom Endpoint with AskSage, I gave it the wrong URL for the OpenAI chat completions endpoint. I used
https://api.asksage.ai/server/v1/chat/completions.
When asking the agent a simple question, there was no error message. The prompt just returned saying "0 credits"

Debugging ID
{"request_id":"b07a6682-8b16-428a-81a2-a5a68cdd5d80","conversation_id":"18fb036a-c169-4cf7-94bb-7b671e23636f"}

The correct URL needs "/openai" after "/server"

To reproduce

If no access to AskSage, might be able to reproduce by setting up a custom endpoint with an incorrect but functioning URL.

Expected behavior

An error message of some kind.

Screenshots, videos, and logs

Operating system (OS)

macOS

Operating system and version

26.5.0

Shell Version

GNU bash, version 3.2.57(1)-release (arm64-apple-darwin25)

Current Warp version

v0.2026.07.08.17.54.preview_02

Regression

No, this bug or issue has existed throughout my experience using Warp

Recent working Warp date

No response

Additional context

No response

Does this block you from using Warp daily?

No

Is this an issue only in Warp?

Yes, I confirmed that this only happens in Warp, not other terminals.

Warp Internal (ignore): linear-label:b9d78064-c89e-4973-b153-5178a31ee54e

No response

[oz-for-oss]

@rljacob

Here's what I found while triaging this issue:

Thanks for the detailed report and debugging ID. This looks like a custom AI endpoint error-handling issue: Warp accepts the URL shape, but when the provider path is wrong it should surface an endpoint/provider error instead of silently returning to the prompt with 0 credits.

Maintainer details

I concluded that custom AI endpoint requests with an incorrect OpenAI-compatible AskSage URL appear to finish silently with a 0-credit footer instead of surfacing an endpoint/provider error..

Triage summary

This looks like a bug in custom AI endpoint error handling. The reporter configured an AskSage custom endpoint with an OpenAI-compatible URL missing the /openai path segment, then Agent Mode returned to the prompt with a 0 credits footer and no visible error.

Evidence

• The report includes the bad URL, the expected corrected URL shape, a screenshot, macOS/Warp build details, and a debugging ID.
• app/src/settings_view/custom_inference_modal.rs validates custom endpoint URLs for syntax, HTTPS, and restricted local/private hosts, but it does not validate that the route is actually an OpenAI-compatible chat completions endpoint.
• crates/ai/src/api_keys.rs packages custom endpoints as CustomModelProviders with base_url, API key, and model config keys; app/src/ai/agent/api.rs and app/src/ai/agent/api/impl.rs attach those providers to multi-agent requests.
• Response handling in app/src/ai/blocklist/controller.rs, app/src/ai/blocklist/controller/response_stream.rs, and app/src/ai/agent/mod.rs has user-visible error rendering for known stream/API failure categories. The reported 0 credits silent completion suggests the downstream custom endpoint failure is either not returned as one of those categories or is being treated as a successful/zero-cost finish.

Reproducibility

Medium. The exact AskSage repro may require AskSage access, but the report is concrete enough to investigate with the debugging ID or by configuring a custom endpoint URL that is syntactically valid but not the provider's OpenAI chat completions route.

Likely root cause

Most likely an error-classification/surfacing gap for custom inference provider failures rather than a settings validation bug alone. The client currently accepts the URL shape and forwards it; the missing behavior is a visible error when the provider route is invalid or returns a non-OpenAI-compatible response.

Duplicate search

I enumerated open issues and recently closed issues and found related custom inference/BYOK failures, but no exact AskSage, incorrect custom endpoint URL, or 0 credits silent-failure duplicate meeting the duplicate threshold.

This is my automated analysis and may be incorrect. A maintainer will verify the details.

Powered by Oz