llm/providers/codex/
provider.rs

1use super::mappers::{map_messages, map_tools};
2use super::oauth::CodexTokenManager;
3use super::streaming::{CodexStreamEvent, process_response_stream};
4use crate::provider::{LlmResponseStream, StreamingModelProvider, get_context_window};
5use crate::{Context, LlmError, Result};
6use aether_auth::OAuthCredentialStorage;
7use async_openai::types::responses::{
8    CreateResponse, IncludeEnum, InputParam, Reasoning, ReasoningSummary, ResponseTextParam,
9    TextResponseFormatConfiguration, Verbosity,
10};
11use eventsource_stream::Eventsource;
12use futures::StreamExt;
13use reqwest::header::{AUTHORIZATION, HeaderMap, HeaderValue};
14use serde_json::Value;
15use std::sync::Arc;
16use tracing::debug;
17
18const CODEX_API_BASE: &str = "https://chatgpt.com/backend-api/codex";
19
20#[derive(Clone)]
21pub struct CodexProvider {
22    base_url: String,
23    client: reqwest::Client,
24    model: String,
25    token_manager: Arc<CodexTokenManager>,
26}
27
28impl CodexProvider {
29    pub fn new(store: Arc<dyn OAuthCredentialStorage>) -> Self {
30        let token_manager = CodexTokenManager::new(store, super::PROVIDER_ID);
31        Self {
32            base_url: CODEX_API_BASE.to_string(),
33            client: reqwest::Client::new(),
34            model: "gpt-5.5".to_string(),
35            token_manager: Arc::new(token_manager),
36        }
37    }
38
39    pub fn with_connection(mut self, connection: crate::ProviderConnectionConfig) -> Self {
40        if let Some(base_url) = connection.base_url {
41            self.base_url = base_url.trim_end_matches('/').to_string();
42        }
43        self
44    }
45
46    pub fn with_model(mut self, model: &str) -> Self {
47        self.model = model.to_string();
48        self
49    }
50
51    fn build_typed_request(&self, context: &Context) -> Result<CreateResponse> {
52        let (system_prompt, input) = map_messages(context.messages())?;
53        let tools = if context.tools().is_empty() { None } else { Some(map_tools(context.tools())?) };
54
55        Ok(CreateResponse {
56            model: Some(self.model.clone()),
57            input: InputParam::Items(input),
58            instructions: system_prompt,
59            tools,
60            store: Some(false),
61            stream: Some(true),
62            reasoning: Some(Reasoning { effort: None, summary: Some(ReasoningSummary::Auto) }),
63            include: Some(vec![IncludeEnum::ReasoningEncryptedContent]),
64            text: Some(ResponseTextParam {
65                format: TextResponseFormatConfiguration::Text,
66                verbosity: Some(Verbosity::Medium),
67            }),
68            prompt_cache_key: context.prompt_cache_key().map(String::from),
69            ..Default::default()
70        })
71    }
72
73    /// The typed `async-openai` request cannot encode every effort the Codex
74    /// API accepts (e.g. `max`), so the effort is written onto the serialized
75    /// wire body instead.
76    fn build_wire_request(&self, context: &Context) -> Result<Value> {
77        let mut body = serde_json::to_value(self.build_typed_request(context)?)?;
78        let effort = context.reasoning_effort().map_or("medium", crate::ReasoningEffort::as_str);
79        body["reasoning"]["effort"] = effort.into();
80        Ok(body)
81    }
82
83    async fn build_headers(&self) -> Result<HeaderMap> {
84        let (access_token, account_id) = self.token_manager.get_valid_token().await?;
85
86        let mut headers = HeaderMap::new();
87        headers.insert(
88            AUTHORIZATION,
89            HeaderValue::from_str(&format!("Bearer {access_token}"))
90                .map_err(|e| LlmError::InvalidApiKey(e.to_string()))?,
91        );
92        headers.insert(
93            "chatgpt-account-id",
94            HeaderValue::from_str(&account_id).map_err(|e| LlmError::InvalidApiKey(e.to_string()))?,
95        );
96        headers.insert("OpenAI-Beta", HeaderValue::from_static("responses=experimental"));
97        headers.insert("originator", HeaderValue::from_static("codex_cli_rs"));
98
99        Ok(headers)
100    }
101
102    /// Send the request and return a stream of SSE lines parsed into typed events.
103    ///
104    /// Uses manual SSE parsing because the Codex API does not return a
105    /// `Content-Type: text/event-stream` header, which `reqwest_eventsource`
106    /// (used by `async-openai`'s `create_stream`) requires.
107    async fn send_request(
108        &self,
109        request: Value,
110        headers: HeaderMap,
111    ) -> Result<impl futures::Stream<Item = Result<CodexStreamEvent>>> {
112        let url = format!("{}/responses", self.base_url);
113
114        debug!("Sending request to Codex API: {url}");
115        debug!(
116            "Codex request body: {}",
117            serde_json::to_string(&request).unwrap_or_else(|_| "<failed to serialize>".to_string())
118        );
119
120        let response = self.client.post(&url).headers(headers).json(&request).send().await?;
121
122        if !response.status().is_success() {
123            let status = response.status();
124            let error_text = response.text().await.unwrap_or_else(|_| "Unknown error".to_string());
125
126            if matches!(status.as_u16(), 401 | 403) {
127                self.token_manager.clear_cache().await;
128            }
129
130            let message = format!("Codex API request failed with status {status}: {error_text}");
131            return Err(match status.as_u16() {
132                429 => LlmError::RateLimited(message),
133                s if (500..600).contains(&s) => LlmError::ServerError { status: Some(s), message },
134                _ => LlmError::ApiError(message),
135            });
136        }
137
138        let event_stream = response.bytes_stream().eventsource().filter_map(|result| {
139            std::future::ready(match result {
140                Ok(event) if event.data == "[DONE]" => None,
141                Ok(event) => match serde_json::from_str::<CodexStreamEvent>(&event.data) {
142                    Ok(parsed) => Some(Ok(parsed)),
143                    Err(e) => {
144                        debug!("Failed to parse Codex SSE line: {} - Error: {e}", event.data);
145                        None
146                    }
147                },
148                Err(e) => Some(Err(LlmError::StreamInterrupted(e.to_string()))),
149            })
150        });
151
152        Ok(event_stream)
153    }
154}
155
156impl StreamingModelProvider for CodexProvider {
157    fn model(&self) -> Option<crate::LlmModel> {
158        format!("{}:{}", super::PROVIDER_ID, self.model).parse().ok()
159    }
160
161    fn context_window(&self) -> Option<u32> {
162        get_context_window(super::PROVIDER_ID, &self.model)
163    }
164
165    fn stream_response(&self, context: &Context) -> LlmResponseStream {
166        let provider = self.clone();
167        let context = match self.model() {
168            Some(model) => context.filter_encrypted_reasoning(&model),
169            None => context.clone(),
170        };
171
172        Box::pin(async_stream::stream! {
173            let headers = match provider.build_headers().await {
174                Ok(h) => h,
175                Err(e) => {
176                    yield Err(e);
177                    return;
178                }
179            };
180
181            let request = match provider.build_wire_request(&context) {
182                Ok(r) => r,
183                Err(e) => {
184                    yield Err(e);
185                    return;
186                }
187            };
188
189            let event_stream = match provider.send_request(request, headers).await {
190                Ok(s) => s,
191                Err(e) => {
192                    yield Err(e);
193                    return;
194                }
195            };
196
197            let mut response_stream = Box::pin(process_response_stream(event_stream));
198            while let Some(result) = response_stream.next().await {
199                yield result;
200            }
201        })
202    }
203
204    fn display_name(&self) -> String {
205        format!("Codex ({})", self.model)
206    }
207}
208
209#[cfg(test)]
210mod tests {
211    use super::*;
212    use crate::ChatMessage;
213    use crate::ContentBlock;
214    use crate::ToolDefinition;
215    use crate::providers::test_capture_server::CaptureServer;
216    use crate::types::IsoString;
217    use aether_auth::{FakeOAuthCredentialStore, OAuthCredential};
218    use base64::Engine;
219    use base64::engine::general_purpose::URL_SAFE_NO_PAD;
220
221    #[test]
222    fn context_window_uses_codex_subscription_limit() {
223        let provider = create_test_provider();
224        assert_eq!(provider.context_window(), Some(272_000));
225    }
226
227    #[test]
228    fn display_name_includes_model() {
229        let provider = create_test_provider();
230        assert_eq!(provider.display_name(), "Codex (gpt-5.5)");
231    }
232
233    #[tokio::test]
234    async fn stream_response_sends_max_effort_on_the_wire() {
235        let mut server = CaptureServer::start().await;
236        let provider = server_backed_provider(&server).with_model("gpt-5.6-sol");
237        let mut context = Context::new(
238            vec![
239                ChatMessage::System { content: "You are helpful".to_string(), timestamp: IsoString::now() },
240                ChatMessage::User { content: vec![ContentBlock::text("Think harder")], timestamp: IsoString::now() },
241            ],
242            vec![ToolDefinition::new(
243                "bash",
244                "Run a command",
245                r#"{"type": "object", "properties": {"cmd": {"type": "string"}}}"#,
246            )],
247        );
248        context.set_reasoning_effort(Some(crate::ReasoningEffort::Max));
249        context.set_prompt_cache_key(Some("session-abc".to_string()));
250
251        let responses = provider.stream_response(&context).collect::<Vec<_>>().await;
252        let captured = server.captured().await;
253
254        assert!(responses.iter().all(Result::is_ok), "{responses:?}");
255        assert_eq!(captured.body["reasoning"]["effort"], "max");
256        assert_eq!(captured.body["model"], "gpt-5.6-sol");
257        assert_eq!(captured.body["instructions"], "You are helpful");
258        assert_eq!(captured.body["tools"].as_array().unwrap().len(), 1);
259        assert_eq!(captured.body["prompt_cache_key"], "session-abc");
260        assert_eq!(captured.body["store"], false);
261        assert_eq!(captured.body["stream"], true);
262        assert_eq!(captured.headers["chatgpt-account-id"], "account-1");
263    }
264
265    #[tokio::test]
266    async fn stream_response_defaults_to_medium_effort_on_the_wire() {
267        let mut server = CaptureServer::start().await;
268        let provider = server_backed_provider(&server);
269        let context = Context::new(
270            vec![ChatMessage::User { content: vec![ContentBlock::text("Hello")], timestamp: IsoString::now() }],
271            vec![],
272        );
273
274        let responses = provider.stream_response(&context).collect::<Vec<_>>().await;
275        let captured = server.captured().await;
276
277        assert!(responses.iter().all(Result::is_ok), "{responses:?}");
278        assert_eq!(captured.body["reasoning"]["effort"], "medium");
279    }
280
281    fn server_backed_provider(server: &CaptureServer) -> CodexProvider {
282        let credential = OAuthCredential {
283            client_id: "test".to_string(),
284            access_token: test_jwt("account-1"),
285            refresh_token: None,
286            expires_at: Some(u64::MAX),
287            granted_scopes: Vec::new(),
288        };
289        let store: Arc<dyn OAuthCredentialStorage> =
290            Arc::new(FakeOAuthCredentialStore::new().with_credential("codex", credential));
291        CodexProvider::new(store).with_connection(crate::ProviderConnectionConfig {
292            base_url: Some(server.base_url.clone()),
293            ..Default::default()
294        })
295    }
296
297    fn test_jwt(account_id: &str) -> String {
298        let header = URL_SAFE_NO_PAD.encode(r#"{"alg":"none"}"#);
299        let payload = URL_SAFE_NO_PAD
300            .encode(serde_json::json!({"https://api.openai.com/auth": {"chatgpt_account_id": account_id}}).to_string());
301        format!("{header}.{payload}.signature")
302    }
303
304    fn create_test_provider() -> CodexProvider {
305        let store: Arc<dyn OAuthCredentialStorage> = Arc::new(FakeOAuthCredentialStore::new());
306        CodexProvider::new(store).with_model("gpt-5.5")
307    }
308}
llm/providers/codex/provider.rs

llm/providers/codex/
provider.rs
