byokey-proxy 1.0.0

Bring Your Own Keys — AI subscription-to-API proxy gateway
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
//! `AmpCode` provider-specific API route handlers.
//!
//! `AmpCode` routes AI requests to provider-namespaced endpoints instead of
//! the generic `/v1/chat/completions`:
//!
//! | `AmpCode` route | Handler |
//! |---|---|
//! | `POST /api/provider/anthropic/v1/messages` | [`messages::anthropic_messages`] (aliased) |
//! | `POST /api/provider/openai/v1/chat/completions` | [`chat::chat_completions`] (aliased) |
//! | `POST /api/provider/openai/v1/responses` | [`codex_responses_passthrough`] |
//! | `POST /api/provider/google/v1beta/models/{action}` | [`gemini_native_passthrough`] |
//!
//! Management routes (`/api/auth`, `/api/threads`, etc.) are forwarded to
//! `ampcode.com` verbatim via [`ampcode_proxy`].

use axum::{
    extract::{Path, Query, State},
    http::{Method, StatusCode},
    response::{IntoResponse, Response},
};
use byokey_types::{ByokError, ProviderId};
use bytes::Bytes;
use futures_util::TryStreamExt as _;
use serde_json::Value;
use std::{collections::HashMap, sync::Arc};

use crate::middleware::forward::ForwardedHeaders;
use crate::util::stream::{CodexParser, GeminiParser, response_to_stream, tap_usage_stream};
use crate::util::{bad_gateway, extract_usage, sse_response, upstream_error};
use crate::{AppState, error::ApiError};

const CODEX_RESPONSES_URL: &str = "https://chatgpt.com/backend-api/codex/responses";
const OPENAI_RESPONSES_URL: &str = "https://api.openai.com/v1/responses";
const CODEX_VERSION: &str = "0.120.0";
const CODEX_USER_AGENT: &str = "codex-tui/0.120.0 (Mac OS 26.0.1; arm64) Apple_Terminal/464";
const GEMINI_MODELS_BASE: &str = "https://generativelanguage.googleapis.com/v1beta/models";
const AMP_BACKEND: &str = "https://ampcode.com";

/// Handles `POST /api/provider/openai/v1/responses`.
///
/// `AmpCode` sends requests already formatted as `OpenAI` Responses API objects
/// (used for Oracle and Deep modes: `GPT-5.2`, `GPT-5.3 Codex`).
///
/// Routing:
/// - **OAuth token** → `chatgpt.com/backend-api/codex/responses` (Codex CLI endpoint)
/// - **API key** → `api.openai.com/v1/responses` (public `OpenAI` Responses API)
///
/// # Errors
///
/// Returns [`ApiError`] if Codex auth fails, the upstream returns a non-2xx
/// status, or the upstream JSON cannot be parsed.
///
/// # Panics
///
/// Panics if `axum::Response::builder` somehow fails to build a valid SSE
/// response (only possible if the constant headers above are malformed).
#[allow(clippy::too_many_lines)]
pub async fn codex_responses_passthrough(
    State(state): State<Arc<AppState>>,
    axum::extract::Json(body): axum::extract::Json<Value>,
) -> Result<Response, ApiError> {
    let mut body = body;

    // The Codex Responses API requires `instructions`; inject an empty default
    // when the client (e.g. AmpCode) omits it.
    if body.get("instructions").is_none() {
        body["instructions"] = Value::String(String::new());
    }

    let config = state.config.load();
    let api_key = config
        .providers
        .get(&ProviderId::Codex)
        .and_then(|pc| pc.api_key.clone());

    let model_name = body
        .get("model")
        .and_then(Value::as_str)
        .unwrap_or("unknown")
        .to_string();

    let (is_oauth, token, account_id) = if let Some(key) = api_key {
        (false, key, byokey_types::DEFAULT_ACCOUNT.to_string())
    } else {
        let (account_id, tok) = state
            .auth
            .get_token_with_account(&ProviderId::Codex)
            .await
            .map_err(ApiError::from)?;
        (true, tok.access_token, account_id)
    };

    // chatgpt.com/backend-api/codex/responses rejects sampling and limit
    // parameters that the public OpenAI Responses API accepts.
    if is_oauth && let Some(obj) = body.as_object_mut() {
        obj.remove("max_output_tokens");
        obj.remove("temperature");
        obj.remove("top_p");
    }

    let upstream_url = if is_oauth {
        CODEX_RESPONSES_URL
    } else {
        OPENAI_RESPONSES_URL
    };
    let auth_mode = if is_oauth { "oauth" } else { "api_key" };

    tracing::info!(
        model = %model_name,
        auth_mode,
        upstream_url,
        "codex responses: sending request to upstream"
    );

    let start = std::time::Instant::now();

    let resp = if is_oauth {
        state
            .http
            .post(CODEX_RESPONSES_URL)
            .header("authorization", format!("Bearer {token}"))
            .header("content-type", "application/json")
            .header("Version", CODEX_VERSION)
            .header("User-Agent", CODEX_USER_AGENT)
            .header("Originator", "codex_cli_rs")
            .header("Accept", "text/event-stream")
            .json(&body)
            .send()
            .await
    } else {
        state
            .http
            .post(OPENAI_RESPONSES_URL)
            .header("authorization", format!("Bearer {token}"))
            .header("content-type", "application/json")
            .json(&body)
            .send()
            .await
    };

    let elapsed = start.elapsed();

    let resp = resp.map_err(|e| {
        tracing::error!(
            model = %model_name,
            auth_mode,
            upstream_url,
            ?elapsed,
            error = %e,
            "codex responses: transport error (DNS/TLS/connection)"
        );
        ApiError(ByokError::from(e))
    })?;

    let provider = "codex";
    let upstream_status = resp.status().as_u16();
    let status = StatusCode::from_u16(upstream_status).unwrap_or(StatusCode::BAD_GATEWAY);

    if !status.is_success() {
        let headers_dbg = format!("{:?}", resp.headers());
        let text = resp.text().await.unwrap_or_default();
        tracing::error!(
            model = %model_name,
            auth_mode,
            upstream_url,
            upstream_status,
            ?elapsed,
            response_headers = %headers_dbg,
            response_body = %text,
            "codex responses: upstream returned non-2xx"
        );
        return Err(upstream_error(
            status,
            text,
            &state.usage,
            &model_name,
            provider,
            &account_id,
        ));
    }

    let content_type = resp
        .headers()
        .get("content-type")
        .and_then(|v| v.to_str().ok())
        .unwrap_or("")
        .to_string();

    tracing::info!(
        model = %model_name,
        upstream_status,
        ?elapsed,
        content_type = %content_type,
        "codex responses: upstream returned success"
    );

    // chatgpt.com may omit Content-Type entirely for SSE responses;
    // default to streaming when the header is absent or unrecognised.
    let is_sse = content_type.is_empty()
        || content_type.contains("text/event-stream")
        || content_type.contains("application/x-ndjson");

    if is_sse {
        let tapped = tap_usage_stream(
            response_to_stream(resp),
            state.usage.clone(),
            model_name.clone(),
            provider.to_string(),
            account_id.clone(),
            CodexParser::new(),
        );
        let stream_model = model_name;
        let mapped = tapped.map_err(move |e| {
            tracing::error!(
                model = %stream_model,
                error = %e,
                "codex responses: SSE stream error mid-transfer"
            );
            std::io::Error::other(e.to_string())
        });
        Ok(sse_response(status, mapped))
    } else {
        let json: Value = resp
            .json()
            .await
            .map_err(|e| ApiError(ByokError::from(e)))?;
        let (input, output) = extract_usage(&json, "/usage/input_tokens", "/usage/output_tokens");
        state
            .usage
            .record_success_for(&model_name, provider, &account_id, input, output);
        Ok((status, axum::Json(json)).into_response())
    }
}

/// Handles `POST /api/provider/google/v1beta/models/{action}`.
///
/// `AmpCode` sends requests in Google's native `generateContent` /
/// `streamGenerateContent` format (used for Review, Search, Look At, Handoff,
/// Topics, and Painter modes).
///
/// The `{action}` path segment contains `{model}:{method}`, e.g.
/// `gemini-3-pro:generateContent` or `gemini-3-flash:streamGenerateContent`.
/// Query parameters (e.g. `?alt=sse`) are forwarded verbatim to the upstream.
///
/// When the Gemini provider has `backend` configured (e.g. `backend: copilot`),
/// the request is translated from Google native format to `OpenAI` format,
/// sent to the backend provider, and the response is translated back.
///
/// # Errors
///
/// Returns [`ApiError`] if Gemini auth fails, the upstream returns a non-2xx
/// status, the upstream JSON cannot be parsed, or backend translation fails.
///
/// # Panics
///
/// Panics if `axum::Response::builder` somehow fails to build a valid SSE
/// response (only possible if the constant headers above are malformed).
#[allow(clippy::implicit_hasher, clippy::too_many_lines)] // Axum extractors fix the HashMap hasher; long due to auth+SSE branching.
pub async fn gemini_native_passthrough(
    State(state): State<Arc<AppState>>,
    Path(action): Path<String>,
    Query(query_params): Query<HashMap<String, String>>,
    axum::extract::Json(body): axum::extract::Json<Value>,
) -> Result<Response, ApiError> {
    let config = state.config.load();
    let gemini_config = config
        .providers
        .get(&ProviderId::Gemini)
        .cloned()
        .unwrap_or_default();

    // Extract model name from action (e.g. "gemini-3-pro:generateContent" → "gemini-3-pro")
    let model_name = action
        .split_once(':')
        .map_or(action.as_str(), |(model, _)| model);

    // If a backend override is configured, translate and route through it.
    if let Some(backend_id) = &gemini_config.backend {
        return gemini_native_via_backend(
            &state,
            &action,
            &query_params,
            body,
            model_name,
            backend_id,
        )
        .await;
    }

    // Direct passthrough to Gemini API.
    let api_key = gemini_config.api_key;

    // Rebuild query string from parsed params.
    let qs: String = query_params
        .iter()
        .map(|(k, v)| format!("{k}={v}"))
        .collect::<Vec<_>>()
        .join("&");
    let url = if qs.is_empty() {
        format!("{GEMINI_MODELS_BASE}/{action}")
    } else {
        format!("{GEMINI_MODELS_BASE}/{action}?{qs}")
    };

    // API key → `x-goog-api-key`; OAuth token → `Authorization: Bearer`.
    let (auth_name, auth_value, account_id): (&'static str, String, String) =
        if let Some(key) = api_key {
            (
                "x-goog-api-key",
                key,
                byokey_types::DEFAULT_ACCOUNT.to_string(),
            )
        } else {
            let (account_id, token) = state
                .auth
                .get_token_with_account(&ProviderId::Gemini)
                .await
                .map_err(ApiError::from)?;
            (
                "authorization",
                format!("Bearer {}", token.access_token),
                account_id,
            )
        };

    let resp = state
        .http
        .post(&url)
        .header(auth_name, auth_value)
        .header("content-type", "application/json")
        .json(&body)
        .send()
        .await
        .map_err(|e| ApiError(ByokError::from(e)))?;

    let provider = "gemini";
    let status = StatusCode::from_u16(resp.status().as_u16()).unwrap_or(StatusCode::BAD_GATEWAY);
    if !status.is_success() {
        let text = resp.text().await.unwrap_or_default();
        return Err(upstream_error(
            status,
            text,
            &state.usage,
            model_name,
            provider,
            &account_id,
        ));
    }

    let is_sse = resp
        .headers()
        .get("content-type")
        .and_then(|v| v.to_str().ok())
        .is_some_and(|ct| ct.contains("text/event-stream"));

    if is_sse {
        let tapped = tap_usage_stream(
            response_to_stream(resp),
            state.usage.clone(),
            model_name.to_string(),
            provider.to_string(),
            account_id,
            GeminiParser::new(),
        );
        let mapped = tapped.map_err(|e| std::io::Error::other(e.to_string()));
        Ok(sse_response(status, mapped))
    } else {
        let json: Value = resp
            .json()
            .await
            .map_err(|e| ApiError(ByokError::from(e)))?;
        let (input, output) = extract_usage(
            &json,
            "/usageMetadata/promptTokenCount",
            "/usageMetadata/candidatesTokenCount",
        );
        state
            .usage
            .record_success_for(model_name, provider, &account_id, input, output);
        Ok((status, axum::Json(json)).into_response())
    }
}

/// Route a Gemini native request through an `OpenAI`-compatible backend provider.
///
/// Translates: Google native → `OpenAI` → backend → `OpenAI` response → Google native.
async fn gemini_native_via_backend(
    state: &Arc<AppState>,
    action: &str,
    query_params: &HashMap<String, String>,
    body: Value,
    model: &str,
    backend_id: &ProviderId,
) -> Result<Response, ApiError> {
    let is_stream = action.contains("streamGenerateContent")
        || query_params.get("alt").is_some_and(|v| v == "sse");

    // Build the executor for the backend provider.
    let config = state.config.load();
    let backend_config = config
        .providers
        .get(backend_id)
        .cloned()
        .unwrap_or_default();
    let executor = byokey_provider::make_executor(
        backend_id,
        backend_config.api_key,
        backend_config.base_url,
        state.auth.clone(),
        state.http.clone(),
        Some(state.ratelimits.clone()),
        &state.versions,
    )
    .ok_or_else(|| {
        ApiError::from(ByokError::UnsupportedModel(format!(
            "backend {backend_id:?} has no executor"
        )))
    })?;

    // Translate Gemini native request → OpenAI format.
    let mut openai_req: Value = byokey_translate::GeminiNativeRequest { body: &body, model }
        .try_into()
        .map_err(ApiError::from)?;

    // Inject stream flag based on the Gemini action URL.
    openai_req["stream"] = Value::Bool(is_stream);

    // Build a ChatRequest from the translated OpenAI body.
    let chat_request: byokey_types::ChatRequest =
        serde_json::from_value(openai_req).map_err(|e| {
            ApiError::from(ByokError::Translation(format!(
                "failed to parse translated request: {e}"
            )))
        })?;

    let provider_name = backend_id.to_string();
    // Backend executors do their own account rotation; the specific account
    // isn't surfaced here yet, so attribute to DEFAULT_ACCOUNT for now.
    let account_id = byokey_types::DEFAULT_ACCOUNT;

    // Send through the backend executor.
    let provider_resp = match executor.chat_completion(chat_request).await {
        Ok(r) => r,
        Err(e) => {
            state
                .usage
                .record_failure_for(model, &provider_name, account_id);
            return Err(ApiError::from(e));
        }
    };

    match provider_resp {
        byokey_types::traits::ProviderResponse::Complete(openai_resp) => {
            // Extract usage from the OpenAI-format response before translating.
            let usage_obj = openai_resp.get("usage");
            let input = usage_obj
                .and_then(|u| u.get("prompt_tokens"))
                .and_then(Value::as_u64)
                .unwrap_or(0);
            let output = usage_obj
                .and_then(|u| u.get("completion_tokens"))
                .and_then(Value::as_u64)
                .unwrap_or(0);
            state
                .usage
                .record_success_for(model, &provider_name, account_id, input, output);

            let gemini_resp: Value = byokey_translate::OpenAIResponseToGemini {
                body: &openai_resp,
                model,
            }
            .try_into()
            .map_err(ApiError::from)?;
            Ok(axum::Json(gemini_resp).into_response())
        }
        byokey_types::traits::ProviderResponse::Stream(byte_stream) => {
            // Tap the OpenAI stream for usage before translating to Gemini SSE.
            let tapped = tap_usage_stream(
                byte_stream,
                state.usage.clone(),
                model.to_string(),
                provider_name,
                account_id.to_string(),
                GeminiParser::new(),
            );
            let model_owned = model.to_string();
            let translated = byte_stream_to_gemini_sse(tapped, model_owned);
            let mapped = translated.map_err(|e| std::io::Error::other(e.to_string()));
            Ok(sse_response(StatusCode::OK, mapped))
        }
    }
}

/// Transform a stream of `OpenAI` SSE byte chunks into Gemini-native SSE chunks.
///
/// The upstream `ByteStream` yields arbitrary byte boundaries; SSE lines may be
/// split across chunks. We buffer incoming bytes and split on newlines so that
/// each line is translated individually.
fn byte_stream_to_gemini_sse(
    upstream: byokey_types::traits::ByteStream,
    model: String,
) -> impl futures_util::Stream<Item = std::result::Result<Bytes, ByokError>> {
    use futures_util::StreamExt as _;

    let mut buffer = Vec::<u8>::new();

    upstream.flat_map(move |chunk_result| {
        let mut output: Vec<std::result::Result<Bytes, ByokError>> = Vec::new();

        match chunk_result {
            Err(e) => output.push(Err(e)),
            Ok(chunk) => {
                buffer.extend_from_slice(&chunk);

                // Process complete lines from the buffer.
                while let Some(pos) = buffer.iter().position(|&b| b == b'\n') {
                    let line: Vec<u8> = buffer.drain(..=pos).collect();
                    let translated: Option<Vec<u8>> = byokey_translate::OpenAISseChunk {
                        line: &line,
                        model: &model,
                    }
                    .into();
                    if let Some(bytes) = translated {
                        output.push(Ok(Bytes::from(bytes)));
                    }
                }
            }
        }

        futures_util::stream::iter(output)
    })
}

/// Transparent proxy to `ampcode.com` — used for both `/api/{*path}` and
/// `/v0/management/{*path}`. Takes the original URI path directly so a
/// single handler covers all non-provider amp routes.
pub async fn ampcode_proxy(
    State(state): State<Arc<AppState>>,
    axum::extract::Extension(fwd): axum::extract::Extension<ForwardedHeaders>,
    method: Method,
    uri: axum::http::Uri,
    body: Bytes,
) -> Response {
    let path = uri.path();
    let url = match uri.query() {
        Some(q) => format!("{AMP_BACKEND}{path}?{q}"),
        None => format!("{AMP_BACKEND}{path}"),
    };

    let debug = path.ends_with("/internal") && tracing::enabled!(tracing::Level::DEBUG);
    if debug {
        let req_body = std::str::from_utf8(&body)
            .ok()
            .and_then(|s| serde_json::from_str::<serde_json::Value>(s).ok())
            .map_or_else(
                || format!("{body:?}"),
                |v| serde_json::to_string_pretty(&v).unwrap_or_default(),
            );
        tracing::debug!(%method, %url, body = %req_body, "ampcode proxy request");
    }

    let resp = match state
        .http
        .request(method, url)
        .headers(fwd.headers)
        .body(body)
        .send()
        .await
    {
        Ok(r) => r,
        Err(e) => return bad_gateway(e),
    };

    let status = StatusCode::from_u16(resp.status().as_u16()).unwrap_or(StatusCode::BAD_GATEWAY);

    let mut resp_headers = axum::http::HeaderMap::new();
    for (name, value) in resp.headers() {
        if let (Ok(n), Ok(v)) = (
            axum::http::HeaderName::from_bytes(name.as_ref()),
            axum::http::HeaderValue::from_bytes(value.as_bytes()),
        ) {
            resp_headers.insert(n, v);
        }
    }

    let body_bytes = resp.bytes().await.unwrap_or_default();

    if debug {
        let resp_body = std::str::from_utf8(&body_bytes)
            .ok()
            .and_then(|s| serde_json::from_str::<serde_json::Value>(s).ok())
            .map_or_else(
                || format!("{body_bytes:?}"),
                |v| serde_json::to_string_pretty(&v).unwrap_or_default(),
            );
        tracing::debug!(%status, body = %resp_body, "ampcode proxy response");
    }

    (status, resp_headers, body_bytes).into_response()
}

#[cfg(test)]
mod tests {
    #[test]
    fn test_urls_are_https() {
        assert!(super::CODEX_RESPONSES_URL.starts_with("https://"));
        assert!(super::OPENAI_RESPONSES_URL.starts_with("https://"));
        assert!(super::GEMINI_MODELS_BASE.starts_with("https://"));
    }
}