BigPizzaV3/CodexPlusPlus · error · anyhow::Error

OpenAI 会话身份仅支持 Responses API

Error message

OpenAI 会话身份仅支持 Responses API

What it means

apply_relay_config_to_home_with_session_provider (crates/codex-plus-core/src/relay_config.rs:335) enforces that the OpenAI session identity (RelaySessionProvider::Openai, ChatGPT-style login auth) is only combined with the Responses wire protocol. Codex OpenAI-session auth works only against the Responses API, so a chat-protocol config would break at runtime; this guard fails at apply time instead.

Source

Thrown at crates/codex-plus-core/src/relay_config.rs:335

pub fn apply_relay_config_to_home_with_session_provider(
    home: &Path,
    base_url: &str,
    bearer_token: &str,
    protocol: RelayProtocol,
    proxy_port: u16,
    session_provider: RelaySessionProvider,
) -> anyhow::Result<RelayApplyResult> {
    let base_url = base_url.trim();
    if base_url.is_empty() {
        anyhow::bail!("中转 Base URL 不能为空");
    }
    let bearer_token = bearer_token.trim();
    if bearer_token.is_empty() {
        anyhow::bail!("中转 Key 不能为空");
    }
    if session_provider == RelaySessionProvider::Openai && protocol != RelayProtocol::Responses {
        anyhow::bail!("OpenAI 会话身份仅支持 Responses API");
    }
    let codex_base_url = codex_base_url_for_protocol(base_url, protocol, proxy_port);
    let updated = upsert_model_provider_config_with_session_provider(
        "",
        &codex_base_url,
        bearer_token,
        true,
        session_provider,
    )?;
    let auth_contents = serde_json::to_string_pretty(&json!({
        "OPENAI_API_KEY": bearer_token
    }))?;
    let backup_path =
        write_codex_live_atomic(home, Some(&updated), Some(auth_contents.as_bytes()))?;
    let status = relay_config_status_from_home(home);
    Ok(RelayApplyResult {
        config_path: status.config_path,
        backup_path,

View on GitHub (pinned to f2074595a2)

Solutions

1. Keep protocol = Responses whenever the session provider is Openai
2. Switch to RelaySessionProvider::Custom with an API key if the upstream only speaks chat completions
3. In the UI, disable the chat protocol option (or auto-reset the session provider to Custom) while OpenAI login is active

Example fix

// before
apply_relay_config_to_home_with_session_provider(&home, url, &token, RelayProtocol::ChatCompletions, port, RelaySessionProvider::Openai)?;

// after
apply_relay_config_to_home_with_session_provider(&home, url, &token, RelayProtocol::Responses, port, RelaySessionProvider::Openai)?;

fn session_protocol_combo_ok(provider: RelaySessionProvider, protocol: RelayProtocol) -> bool {
    provider != RelaySessionProvider::Openai || protocol == RelayProtocol::Responses
}

if !session_protocol_combo_ok(provider, protocol) {
    anyhow::bail!("OpenAI session identity requires Responses; use a Custom provider for chat");
}

if let Err(e) = apply_relay_config_to_home_with_session_provider(&home, url, token, protocol, port, provider) {
    if e.to_string().contains("OpenAI") {
        // invalid combo: pick Responses + Openai, or chat + Custom; do not retry unchanged
    }
}

When it happens

Trigger: Calling apply_relay_config_to_home_with_session_provider with session_provider = RelaySessionProvider::Openai and protocol = RelayProtocol::ChatCompletions. apply_relay_config_to_home_with_protocol always passes Custom, so this comes from the session-provider-aware entry point used by official-login flows.

Common situations: A user logs in with an OpenAI account, then switches the profile wire protocol to chat to reach a chat-only aggregator, and the UI forwards the Openai session provider together with the chat protocol.

Related errors

• Chat Completions 协议暂不支持 Responses compact 请求
• 中转 Base URL 不能为空
• 中转 Key 不能为空
• config.toml 内容不能为空
• 发送微信回复被拒绝：ret={} errcode={} {}

AI-assisted analysis of BigPizzaV3/CodexPlusPlus@f2074595a2 (2026-08-23). Data as JSON: /api/errors/deffc367da3eb5d0. Report an issue: GitHub.