Skip to content

Strix 1.1.0 NVIDIA Build Compatibility Bug Report #804

Description

@VanarpSuriv

Strix 1.1.0 NVIDIA Build Compatibility Bug Report
Summary
Strix 1.1.0 fails to use NVIDIA Build's OpenAI-compatible API. The same endpoint, API key, and model work via a direct curl request, but Strix fails during LLM warm-up with OpenAI NotFoundError (404).

Environment
OS: Windows 11
Python: 3.13
strix-agent: 1.1.0
openai: 2.44.0
litellm: 1.92.0
openai-agents: 0.14.6
Provider Configuration
Provider: NVIDIA Build
Base URL: https://integrate.api.nvidia.com/v1
Model: openai/gpt-oss-120b
Authentication: Bearer API key

Environment:
OPENAI_API_KEY=
OPENAI_BASE_URL=https://integrate.api.nvidia.com/v1
LLM_API_BASE=https://integrate.api.nvidia.com/v1
STRIX_LLM=openai/gpt-oss-120b
Expected Behavior
Strix should complete LLM warm-up and begin the security scan.
Actual Behavior
LLM warm-up fails immediately with openai.NotFoundError: 404 page not found.
Investigation
Verified:
• - Base URL is correct.
• - API key is valid.
• - Endpoint is reachable.
• - OpenAI SDK honors OPENAI_BASE_URL.
• - Request reaches NVIDIA.

Control experiment:
A manual curl request to the same endpoint using model 'openai/gpt-oss-120b' succeeds and returns a valid completion.
Observations
Debug logs indicate Strix sends:
model: gpt-oss-120b

Manual curl sends:
model: openai/gpt-oss-120b

The OpenAI chat completions implementation forwards self.model directly, suggesting the model identifier is normalized before reaching the SDK.
Relevant Source Files
strix/config/models.py
agents/models/openai_chatcompletions.py
agents/models/openai_provider.py
Hypothesis
Strix appears to normalize or strip the provider prefix from the configured model before sending the request. NVIDIA Build may require the fully-qualified model identifier (openai/gpt-oss-120b), resulting in the observed 404 response.
Questions for Maintainers

  1. Is NVIDIA Build officially supported?
  2. Should the OpenAI provider preserve the full model identifier?
  3. Is a different configuration expected for NVIDIA Build?
  4. Is this a known compatibility issue with the OpenAI Agents SDK?
    Reproduction
    set OPENAI_API_KEY=<API_KEY>
    set OPENAI_BASE_URL=https://integrate.api.nvidia.com/v1
    set STRIX_LLM=openai/gpt-oss-120b

strix --target https://example.com -n

Strix_NVIDIA_Build_Bug_Report.docx

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions