zeroclaw-labs/zeroclaw · error

OpenAiCompatibleBuilder: base_url() is required

Error message

OpenAiCompatibleBuilder: base_url() is required

What it means

OpenAiCompatibleBuilder is the builder for OpenAI-compatible model providers in zeroclaw-providers. Its build() method treats display_name, base_url, and auth_style as mandatory setters and panics via expect() when any was skipped. This panic means .base_url() was never called before .build(), i.e. the builder contract was violated.

Source

Thrown at crates/zeroclaw-providers/src/compatible.rs:555

        self.auth_profile_override = profile_override;
        self
    }

    /// Finalize the builder into a ready provider. Every optional construction
    /// value must be set on this builder; the returned provider has no
    /// post-construction mutators.
    ///
    /// # Panics
    /// Panics if [`Self::display_name`], [`Self::base_url`], or
    /// [`Self::auth_style`] was not called — those three fields carry no
    /// sensible default and every real call site sets them.
    pub fn build(self) -> OpenAiCompatibleModelProvider {
        let name = self
            .name
            .expect("OpenAiCompatibleBuilder: display_name() is required");
        let base_url = self
            .base_url
            .expect("OpenAiCompatibleBuilder: base_url() is required");
        let auth_style = self
            .auth_style
            .expect("OpenAiCompatibleBuilder: auth_style() is required");
        // Either merge preset can enable the shared merge behavior.
        let merge_system_into_user =
            self.merge_system_into_user || self.merge_system_into_user_preserve_native;
        // Default `native_tool_calling` is `!merge_system_into_user_disable_native`,
        // i.e. only the "combined preset" builder setter disables it. The
        // explicit `without_native_tools()` override wins if present.
        let native_tool_calling = self
            .native_tool_calling_override
            .unwrap_or(!self.merge_system_into_user);
        // Read the PEM bytes now so later HTTP clients incur no per-request I/O.
        // A read error is logged at WARN and TLS falls back to system roots —
        // preserving the established warning-and-fallback semantics.
        let tls_ca_cert_pem =
            self.tls_ca_cert_path
                .as_deref()

View on GitHub (pinned to 88bb9c8533)

Solutions

1. Call .base_url(url) on the builder before .build() (and verify display_name() and auth_style() are also set, they panic the same way).
2. Trace the failing call through make_model_provider/http_client/streaming_http_client to find where the base URL should be injected and why it is None.
3. If the URL comes from env/config, validate its presence early and return a descriptive error instead of letting build() panic.
4. Add a construction unit test for every new provider preset so a missing setter fails in CI, not at runtime.

Example fix

// before
let provider = OpenAiCompatibleModelProvider::builder()
    .display_name("my-llm")
    .auth_style(AuthStyle::Bearer)
    .build(); // panics: base_url() is required

// after
let provider = OpenAiCompatibleModelProvider::builder()
    .display_name("my-llm")
    .base_url("https://api.example.com/v1")
    .auth_style(AuthStyle::Bearer)
    .build();

// Before building, resolve the endpoint URL and fail with a clear error:
let base_url = std::env::var("MY_PROVIDER_BASE_URL")
    .or_else(|_| config.provider_endpoint.clone().ok_or_else(|| anyhow::anyhow!("base URL missing")))?;
let provider = OpenAiCompatibleModelProvider::builder()
    .display_name("my-llm")
    .base_url(base_url)
    .auth_style(AuthStyle::Bearer)
    .build();

// Embedding code that must not abort:
let provider = std::panic::catch_unwind(|| builder.build())
    .map_err(|_| anyhow::anyhow!("provider builder contract violated: display_name/base_url/auth_style required"))?;

When it happens

Trigger: Calling OpenAiCompatibleModelProvider::builder()...build() (directly or through http_client(), streaming_http_client(), or make_model_provider()) without a prior .base_url(url) call. Typically the setter is skipped because the URL comes from config/env and an if-let around it silently falls through.

Common situations: Adding a new OpenAI-compatible provider integration and wiring api_key/model but forgetting the endpoint URL; a config field rename (e.g. endpoint vs base_url) leaving the setter unpopulated; an absent env var making a conditional setter call not execute.

Related errors

• OpenAiCompatibleBuilder: auth_style() is required
• ModelPinnedProviderBuilder: pinned_model() is required
• ModelPinnedProviderBuilder: inner() is required
• ACP returned unknown optionId: {option_id}
• {} model list failed at {url}: HTTP {status}

AI-assisted analysis of zeroclaw-labs/zeroclaw@88bb9c8533 (2026-08-23). Data as JSON: /api/errors/e35dbcf64045fcd5. Report an issue: GitHub.