Hmbown/CodeWhale · error · anyhow::Error

DS4 /v1/models could not be reached at

Error message

DS4 /v1/models could not be reached at {} (is ds4-server running?)

What it means

Produced by ds4_probe_error during doctor's probe_ds4_models: list_models failed with a connection-level error that contains no HTTP status token — the request never got an HTTP response. The message explicitly asks whether ds4-server is running and prints the redacted endpoint.

Solutions

Start ds4-server and confirm it listens on the configured host and port
Verify deepseek_base_url (scheme, host, port) matches the server's actual bind address
Test reachability directly, e.g. curl the /v1/models URL from the same machine

Example fix

# before
deepseek_base_url = "http://127.0.0.1:9000"

# after (server listens on 8080)
deepseek_base_url = "http://127.0.0.1:8080"

Defensive patterns

Strategy: retry

Validate before calling

// Cheap pre-check that something is listening before the probe:
let url = config.deepseek_base_url();
let host_port = parse_host_port(&url)?;
if !tcp_connectable(host_port, std::time::Duration::from_secs(1)).await {
    return Err(anyhow::anyhow!("ds4 endpoint {host_port} is not accepting connections; start ds4-server"));
}

Try / catch

match probe_ds4_models(&config).await {
    Ok(()) => {}
    Err(error) if error.to_string().contains("could not be reached") => {
        // start ds4-server (or fix the URL), then retry the probe once
    }
    Err(error) => return Err(error),
}

Prevention

Start ds4-server before running doctor and verify the port matches deepseek_base_url
Prefer a loopback base URL for local servers to avoid firewall/DNS surprises
Keep one canonical base URL in config instead of per-tool copies that drift

When it happens

Trigger: Nothing listening on the configured host:port (connection refused), DNS resolution failure for the base URL host, or a TLS handshake failure on an https base URL.

Common situations: ds4-server not started or crashed; wrong port in deepseek_base_url; firewall or VPN blocking localhost/remote endpoints; https pointed at an http listener.

Understand the failure class

Background: ECONNREFUSED and "connection refused" / "could not connect to server" errors: what they mean and how to fix them — this error's family across 44 libraries.

Related errors

AI-assisted analysis of Hmbown/CodeWhale@0c42157ee5 (2026-08-20). Data as JSON: /api/errors/4e9b0343b32b8d61. Report an issue: GitHub.

Appendix: source

Thrown at crates/tui/src/doctor.rs:583

}

/// Probe DS4 through its cheap `/v1/models` contract instead of waking the
/// model for a completion. The selected model must be advertised.
pub(crate) async fn probe_ds4_models(config: &crate::config::Config) -> anyhow::Result<()> {
    use crate::client::DeepSeekClient;
    use crate::core::model_client::ModelClient;

    let endpoint = crate::client::redact_url_for_display(&config.deepseek_base_url());
    let client = DeepSeekClient::new(config)?;
    let configured_alias = client.model().to_string();
    let models = match tokio::time::timeout(
        std::time::Duration::from_secs(15),
        client.list_models(),
    )
    .await
    {
        Ok(Ok(models)) => models,
        Ok(Err(error)) => anyhow::bail!(ds4_probe_error(config, &error.to_string())),
        Err(_) => anyhow::bail!("DS4 /v1/models timed out after 15 seconds at {}", endpoint),
    };
    if !models
        .iter()
        .any(|available| available.id == configured_alias)
    {
        let advertised = models
            .iter()
            .take(8)
            .map(|available| available.id.as_str())
            .collect::<Vec<_>>()
            .join(", ");
        anyhow::bail!(
            "DS4 /v1/models at {} did not list configured alias '{configured_alias}' (advertised: {})",
            endpoint,
            if advertised.is_empty() {
                "none"
            } else {

View on GitHub (pinned to 0c42157ee5)