What causes "OpenAI API key is missing." in abi/screenshot-to-code?

Set OPENAI_API_KEY in backend/.env and restart the backend.

abi/screenshot-to-code · critical · Exception

OpenAI API key is missing.

Error message

OpenAI API key is missing.

What it means

Raised by create_provider_session in the provider factory when the selected model is in OPENAI_MODELS but no OpenAI API key is available. The factory reads the key from explicit parameter then the OPENAI_API_KEY env var; the check is fail-fast before any client is constructed.

Source

Thrown at backend/agent/providers/factory.py:43

    anthropic_api_key: Optional[str],
    gemini_api_key: Optional[str],
    replicate_api_key: Optional[str],
    should_extract_assets: bool = True,
    recorder: Optional[AgentRunRecorder] = None,
) -> ProviderSession:
    canonical_tools = canonical_tool_definitions(
        image_generation_enabled=should_generate_images,
        # The edit_images tool calls Replicate, so don't offer it without a key.
        image_editing_enabled=bool(replicate_api_key or REPLICATE_API_KEY),
        # The extract_assets tool calls Gemini, so don't offer it without a key.
        asset_extraction_enabled=should_extract_assets and bool(gemini_api_key),
        # screenshot_preview needs headless Chromium; skip it if it can't launch.
        screenshot_enabled=is_screenshot_preview_available(),
    )

    if model in OPENAI_MODELS:
        if not openai_api_key:
            raise Exception("OpenAI API key is missing.")

        client = AsyncOpenAI(api_key=openai_api_key, base_url=openai_base_url)
        return OpenAIProviderSession(
            client=client,
            model=model,
            prompt_messages=prompt_messages,
            tools=serialize_openai_tools(canonical_tools),
            recorder=recorder,
        )

    if model in ANTHROPIC_MODELS:
        if not anthropic_api_key:
            raise Exception("Anthropic API key is missing.")

        client = AsyncAnthropic(api_key=anthropic_api_key)
        return AnthropicProviderSession(
            client=client,
            model=model,

View on GitHub (pinned to d026163f58)

Solutions

Set OPENAI_API_KEY in backend/.env and restart the backend.
Or enter the key in the in-app Settings dialog so it is passed through the generation route.
Verify with: cd backend && poetry run python -c "import os; print(bool(os.environ['OPENAI_API_KEY']))".
Or switch to a model family whose key you already have (Anthropic/Gemini).

Example fix

# backend/.env
OPENAI_API_KEY=sk-...

Defensive patterns

Strategy: validation

Validate before calling

import os
if model in OPENAI_MODELS and not os.environ.get("OPENAI_API_KEY"):
    raise SystemExit("Set OPENAI_API_KEY in backend/.env before using OpenAI models")

Try / catch

try:
    session = create_provider_session(model=model, ...)
except Exception as e:
    if "API key is missing" in str(e):
        # prompt the user to open Settings / configure .env
        ...

Prevention

Configure backend/.env at deploy time with the keys for every model family you expose in the UI.
Restart uvicorn after editing .env — env changes are not hot-reloaded.
Expose a startup check that logs which providers have keys.

When it happens

Trigger: Requesting generation with an OpenAI model (e.g. gpt-4o family) while neither the function argument openai_api_key nor the OPENAI_API_KEY environment variable is set.

Common situations: Fresh clone without backend/.env, key set in the frontend Settings but not passed through this code path, env var set in a different shell than the one running uvicorn, or typo in the variable name.

Related errors

AI-assisted analysis of abi/screenshot-to-code@d026163f58 (2026-08-14). Data as JSON: /api/errors/247c26bd65f58845. Report an issue: GitHub.