fix(ai): make custom output token budget configurable

This commit is contained in:
Freud authored and GitHub committed 2026-09-03 21:07:37 +08:00
1 parent 56f85a2c77
commit 1c170396d3
28 files changed
+249 -48

No files matched your search

@@ -56,6 +56,8 @@ import {
useSettingsStore, useSettingsStore,
AI_PROVIDER_PRESETS, AI_PROVIDER_PRESETS,
AI_PROVIDER_PARTNER_PRESETS, AI_PROVIDER_PARTNER_PRESETS,
AI_OUTPUT_TOKENS_MAX,
AI_OUTPUT_TOKENS_MIN,
EDITOR_THEMES, EDITOR_THEMES,
DEFAULT_EDITOR_SETTINGS, DEFAULT_EDITOR_SETTINGS,
DEFAULT_DESKTOP_SETTINGS, DEFAULT_DESKTOP_SETTINGS,
@@ -3524,6 +3526,7 @@ const aiEditProxyEnabled = ref(false);
const aiEditProxyUrl = ref(""); const aiEditProxyUrl = ref("");
const aiEditEnableThinking = ref(true); const aiEditEnableThinking = ref(true);
const aiEditReasoningLevel = ref<AiReasoningLevel>("default"); const aiEditReasoningLevel = ref<AiReasoningLevel>("default");
const aiEditMaxOutputTokens = ref<number | undefined>(undefined);
const aiEditContextWindow = ref<number | undefined>(undefined); const aiEditContextWindow = ref<number | undefined>(undefined);
const aiEditCodexCliPath = ref(""); const aiEditCodexCliPath = ref("");
const aiEditCodexCliEnvRows = ref<AiEnvRow[]>([]); const aiEditCodexCliEnvRows = ref<AiEnvRow[]>([]);
@@ -3808,6 +3811,7 @@ function currentAiEditConfig() {
proxyUrl: aiEditProxyUrl.value, proxyUrl: aiEditProxyUrl.value,
enableThinking: aiEditEnableThinking.value, enableThinking: aiEditEnableThinking.value,
reasoningLevel: aiEditReasoningLevel.value, reasoningLevel: aiEditReasoningLevel.value,
maxOutputTokens: aiEditMaxOutputTokens.value || undefined,
contextWindow: aiEditContextWindow.value || undefined, contextWindow: aiEditContextWindow.value || undefined,
codexCliPath: aiEditCodexCliPath.value.trim() || undefined, codexCliPath: aiEditCodexCliPath.value.trim() || undefined,
codexCliEnv: aiIsCodexCli.value ? cliEnvFromRows(aiEditCodexCliEnvRows.value) : {}, codexCliEnv: aiIsCodexCli.value ? cliEnvFromRows(aiEditCodexCliEnvRows.value) : {},
@@ -3894,6 +3898,7 @@ function aiEnterEditMode(configId?: string) {
aiEditProxyUrl.value = config.proxyUrl ?? ""; aiEditProxyUrl.value = config.proxyUrl ?? "";
aiEditEnableThinking.value = config.enableThinking ?? true; aiEditEnableThinking.value = config.enableThinking ?? true;
aiEditReasoningLevel.value = config.reasoningLevel ?? "default"; aiEditReasoningLevel.value = config.reasoningLevel ?? "default";
aiEditMaxOutputTokens.value = config.maxOutputTokens;
aiEditContextWindow.value = config.contextWindow; aiEditContextWindow.value = config.contextWindow;
aiEditCodexCliPath.value = config.codexCliPath ?? ""; aiEditCodexCliPath.value = config.codexCliPath ?? "";
aiEditCodexCliEnvRows.value = aiEnvRowsFromConfig(config.codexCliEnv); aiEditCodexCliEnvRows.value = aiEnvRowsFromConfig(config.codexCliEnv);
@@ -3927,6 +3932,7 @@ function aiEnterEditMode(configId?: string) {
aiEditProxyUrl.value = ""; aiEditProxyUrl.value = "";
aiEditEnableThinking.value = true; aiEditEnableThinking.value = true;
aiEditReasoningLevel.value = "default"; aiEditReasoningLevel.value = "default";
aiEditMaxOutputTokens.value = undefined;
aiEditContextWindow.value = undefined; aiEditContextWindow.value = undefined;
aiEditCodexCliPath.value = ""; aiEditCodexCliPath.value = "";
aiEditCodexCliEnvRows.value = []; aiEditCodexCliEnvRows.value = [];
@@ -7735,6 +7741,16 @@ onUnmounted(() => {
</p> </p>
</div> </div>
</div> </div>
<!-- Maximum Output Tokens -->
<div v-if="!aiIsCliProvider" class="grid grid-cols-3 items-start gap-3">
<Label class="text-right text-xs">{{ t("ai.maxOutputTokens") }}</Label>
<div class="col-span-2">
<Input v-model.number="aiEditMaxOutputTokens" type="number" :min="AI_OUTPUT_TOKENS_MIN" :max="AI_OUTPUT_TOKENS_MAX" step="1000" class="h-8 text-xs" :placeholder="t('ai.maxOutputTokensAuto')" />
<p class="mt-1 text-xs text-muted-foreground">
{{ t("ai.maxOutputTokensHint", { min: AI_OUTPUT_TOKENS_MIN, max: AI_OUTPUT_TOKENS_MAX }) }}
</p>
</div>
</div>
<!-- Proxy --> <!-- Proxy -->
<div v-if="!aiIsCliProvider" class="grid grid-cols-3 items-center gap-3"> <div v-if="!aiIsCliProvider" class="grid grid-cols-3 items-center gap-3">
+3
View File
@@ -2777,6 +2777,9 @@ export default {
contextWindow: "Context Window", contextWindow: "Context Window",
contextWindowAuto: "Auto (detect from model name)", contextWindowAuto: "Auto (detect from model name)",
contextWindowHint: "Tokens. Leave empty to auto-detect. Set manually for local/custom models.", contextWindowHint: "Tokens. Leave empty to auto-detect. Set manually for local/custom models.",
maxOutputTokens: "Max Output Tokens",
maxOutputTokensAuto: "Auto (per action)",
maxOutputTokensHint: "Optional per-configuration limit sent to the provider. Range: {min}–{max}. Thinking requests still use at least 8192 tokens.",
maxRetries: "Max Retries", maxRetries: "Max Retries",
maxRetriesHint: "Retry on rate limits, timeouts, and temporary network errors. 0 = never, max 10.", maxRetriesHint: "Retry on rate limits, timeouts, and temporary network errors. 0 = never, max 10.",
maxRetriesGlobal: "Max Retries", maxRetriesGlobal: "Max Retries",
+3
View File
@@ -2847,6 +2847,9 @@ export default withEnglishFallback({
contextWindow: "Ventana de contexto", contextWindow: "Ventana de contexto",
contextWindowAuto: "Auto (detectar desde el nombre del modelo)", contextWindowAuto: "Auto (detectar desde el nombre del modelo)",
contextWindowHint: "Tokens. Déjalo vacío para detección automática. Configúralo manualmente para modelos locales/personalizados.", contextWindowHint: "Tokens. Déjalo vacío para detección automática. Configúralo manualmente para modelos locales/personalizados.",
maxOutputTokens: "Tokens máximos de salida",
maxOutputTokensAuto: "Auto (por acción)",
maxOutputTokensHint: "Límite opcional por configuración enviado al proveedor. Rango: {min}–{max}. Las solicitudes con razonamiento usan al menos 8192 tokens.",
maxRetriesGlobal: "Reintentos máximos", maxRetriesGlobal: "Reintentos máximos",
maxRetriesGlobalDescription: "Número de reintentos automáticos en errores transitorios de API (límite de tasa, timeout, fallos de red). Se aplica a todos los proveedores de IA por API. Los proveedores CLI no se ven afectados.", maxRetriesGlobalDescription: "Número de reintentos automáticos en errores transitorios de API (límite de tasa, timeout, fallos de red). Se aplica a todos los proveedores de IA por API. Los proveedores CLI no se ven afectados.",
maxRetriesRange: "{min}–{max}, predeterminado {default}", maxRetriesRange: "{min}–{max}, predeterminado {default}",
+3
View File
@@ -2741,6 +2741,9 @@ export default withEnglishFallback({
contextWindow: "Finestra di Contesto", contextWindow: "Finestra di Contesto",
contextWindowAuto: "Auto (rileva dal nome del modello)", contextWindowAuto: "Auto (rileva dal nome del modello)",
contextWindowHint: "Token. Lascia vuoto per il rilevamento automatico. Imposta manualmente per modelli locali/personalizzati.", contextWindowHint: "Token. Lascia vuoto per il rilevamento automatico. Imposta manualmente per modelli locali/personalizzati.",
maxOutputTokens: "Token di output massimi",
maxOutputTokensAuto: "Automatico (per azione)",
maxOutputTokensHint: "Limite opzionale per configurazione inviato al provider. Intervallo: {min}–{max}. Le richieste con ragionamento usano comunque almeno 8192 token.",
maxRetriesGlobal: "Tentativi massimi", maxRetriesGlobal: "Tentativi massimi",
maxRetriesGlobalDescription: "Numero di tentativi automatici su errori API transitori (limite di velocità, timeout, problemi di rete). Si applica a tutti i provider IA via API. I provider CLI non sono interessati.", maxRetriesGlobalDescription: "Numero di tentativi automatici su errori API transitori (limite di velocità, timeout, problemi di rete). Si applica a tutti i provider IA via API. I provider CLI non sono interessati.",
maxRetriesRange: "{min}–{max}, predefinito {default}", maxRetriesRange: "{min}–{max}, predefinito {default}",
+3
View File
@@ -2541,6 +2541,9 @@ export default withEnglishFallback({
contextWindow: "コンテキストウィンドウ", contextWindow: "コンテキストウィンドウ",
contextWindowAuto: "自動(モデル名から検出)", contextWindowAuto: "自動(モデル名から検出)",
contextWindowHint: "トークン数。空欄で自動検出。ローカル/カスタムモデルは手動設定してください。", contextWindowHint: "トークン数。空欄で自動検出。ローカル/カスタムモデルは手動設定してください。",
maxOutputTokens: "最大出力トークン",
maxOutputTokensAuto: "自動(アクションごと)",
maxOutputTokensHint: "プロバイダーに送信する設定ごとの任意の出力上限。範囲: {min}–{max}。思考を有効にしたリクエストでは少なくとも8192トークンを使用します。",
maxRetriesGlobal: "最大再試行回数", maxRetriesGlobal: "最大再試行回数",
maxRetriesGlobalDescription: "一時的なAPIエラー(レート制限、タイムアウト、ネットワーク障害)発生時の自動再試行回数。すべてのAPI対応AIプロバイダーに適用されます。CLIプロバイダーは影響を受けません。", maxRetriesGlobalDescription: "一時的なAPIエラー(レート制限、タイムアウト、ネットワーク障害)発生時の自動再試行回数。すべてのAPI対応AIプロバイダーに適用されます。CLIプロバイダーは影響を受けません。",
maxRetriesRange: "{min}–{max}、デフォルト {default}", maxRetriesRange: "{min}–{max}、デフォルト {default}",
+3
View File
@@ -2592,6 +2592,9 @@ export default withEnglishFallback({
contextWindow: "컨텍스트 창", contextWindow: "컨텍스트 창",
contextWindowAuto: "자동 (모델 이름에서 감지)", contextWindowAuto: "자동 (모델 이름에서 감지)",
contextWindowHint: "토큰. 자동 감지하려면 비워 두세요. 로컬/사용자 지정 모델은 수동으로 설정하세요.", contextWindowHint: "토큰. 자동 감지하려면 비워 두세요. 로컬/사용자 지정 모델은 수동으로 설정하세요.",
maxOutputTokens: "최대 출력 토큰",
maxOutputTokensAuto: "자동 (모델 기본값)",
maxOutputTokensHint: "공급자에게 전송되는 선택적 구성별 출력 토큰 제한입니다. 범위: {min}–{max}. 사고 요청은 여전히 최소 8192 토큰을 사용합니다.",
codexCliPath: "Codex CLI 경로", codexCliPath: "Codex CLI 경로",
codexCliPathHint: "PATH의 codex를 사용하려면 비워 두세요. codex login으로 별도로 로그인하세요.", codexCliPathHint: "PATH의 codex를 사용하려면 비워 두세요. codex login으로 별도로 로그인하세요.",
codexCliPathEnvError: "Codex CLI 경로에는 실행 파일 경로만 포함해야 합니다. 환경 변수는 아래에 추가하세요.", codexCliPathEnvError: "Codex CLI 경로에는 실행 파일 경로만 포함해야 합니다. 환경 변수는 아래에 추가하세요.",
+3
View File
@@ -2845,6 +2845,9 @@ export default withEnglishFallback({
contextWindow: "Janela de Contexto", contextWindow: "Janela de Contexto",
contextWindowAuto: "Automático (detectar pelo nome do modelo)", contextWindowAuto: "Automático (detectar pelo nome do modelo)",
contextWindowHint: "Tokens. Deixe vazio para detecção automática. Defina manualmente para modelos locais/personalizados.", contextWindowHint: "Tokens. Deixe vazio para detecção automática. Defina manualmente para modelos locais/personalizados.",
maxOutputTokens: "Máximo de tokens de saída",
maxOutputTokensAuto: "Automático (por ação)",
maxOutputTokensHint: "Limite opcional por configuração enviado ao provedor. Intervalo: {min}–{max}. Solicitações com raciocínio ainda usam pelo menos 8192 tokens.",
maxRetriesGlobal: "Máximo de Tentativas", maxRetriesGlobal: "Máximo de Tentativas",
maxRetriesGlobalDescription: "Número de tentativas automáticas em erros transitórios de API (limite de taxa, timeout, falhas de rede). Aplica-se a todos os provedores de IA via API. Provedores CLI não são afetados.", maxRetriesGlobalDescription: "Número de tentativas automáticas em erros transitórios de API (limite de taxa, timeout, falhas de rede). Aplica-se a todos os provedores de IA via API. Provedores CLI não são afetados.",
maxRetriesRange: "{min}–{max}, padrão {default}", maxRetriesRange: "{min}–{max}, padrão {default}",
+3
View File
@@ -2700,6 +2700,9 @@ export default withEnglishFallback({
contextWindow: "上下文窗口", contextWindow: "上下文窗口",
contextWindowAuto: "自动(根据模型名称推断)", contextWindowAuto: "自动(根据模型名称推断)",
contextWindowHint: "单位:token。留空自动推断,本地/自定义模型建议手动设置。", contextWindowHint: "单位:token。留空自动推断,本地/自定义模型建议手动设置。",
maxOutputTokens: "最大输出 Token 数",
maxOutputTokensAuto: "自动(按操作类型)",
maxOutputTokensHint: "可选的单配置输出上限,会发送给 AI 服务商。范围:{min}–{max}。启用思考时仍至少使用 8192 token。",
maxRetries: "最大重试次数", maxRetries: "最大重试次数",
maxRetriesHint: "遇到限流、超时或临时网络错误时自动重试。0 = 不重试,最大 10。", maxRetriesHint: "遇到限流、超时或临时网络错误时自动重试。0 = 不重试,最大 10。",
maxRetriesGlobal: "最大重试次数", maxRetriesGlobal: "最大重试次数",
+3
View File
@@ -2740,6 +2740,9 @@ export default withEnglishFallback({
contextWindow: "Context Window", contextWindow: "Context Window",
contextWindowAuto: "自動(從模型名稱偵測)", contextWindowAuto: "自動(從模型名稱偵測)",
contextWindowHint: "Token 數。留空以自動偵測。本機/自訂模型請手動設定。", contextWindowHint: "Token 數。留空以自動偵測。本機/自訂模型請手動設定。",
maxOutputTokens: "最大輸出 Token 數",
maxOutputTokensAuto: "自動(依操作類型)",
maxOutputTokensHint: "可選的單一設定輸出上限,會傳送給 AI 服務商。範圍:{min}–{max}。啟用思考時仍至少使用 8192 token。",
maxRetries: "最大重試次數", maxRetries: "最大重試次數",
maxRetriesHint: "遇到限流、超時或暫時性網路錯誤時自動重試。0 = 不重試,最大 10。", maxRetriesHint: "遇到限流、超時或暫時性網路錯誤時自動重試。0 = 不重試,最大 10。",
maxRetriesGlobal: "最大重試次數", maxRetriesGlobal: "最大重試次數",
+38
View File
@@ -158,6 +158,44 @@ describe("AI SQL dialect prompt", () => {
expect(prompt).toContain("Ask mode"); expect(prompt).toContain("Ask mode");
}); });
it("uses the configured output budget for non-thinking requests", () => {
const request = buildAgentRequest({
config: {
provider: "deepseek",
apiKey: "test",
apiUrl: "https://example.invalid",
model: "deepseek-v4-flash",
enableThinking: false,
maxOutputTokens: 1024,
},
action: "general",
mode: "ask",
instruction: "generate a long migration",
context: context(),
});
expect(request.maxTokens).toBe(1024);
});
it("uses the configured output budget above the thinking minimum", () => {
const request = buildAgentRequest({
config: {
provider: "deepseek",
apiKey: "test",
apiUrl: "https://example.invalid",
model: "deepseek-v4-flash",
enableThinking: true,
maxOutputTokens: 32768,
},
action: "general",
mode: "ask",
instruction: "generate a long migration",
context: context(),
});
expect(request.maxTokens).toBe(32768);
});
it("MongoDB agent mode uses shell commands instead of SQL", () => { it("MongoDB agent mode uses shell commands instead of SQL", () => {
const prompt = buildSystemPrompt("general", context({ databaseType: "mongodb", connectionName: "MongoDB", database: "benchmark" }), "agent"); const prompt = buildSystemPrompt("general", context({ databaseType: "mongodb", connectionName: "MongoDB", database: "benchmark" }), "agent");
+2 -1
View File
@@ -179,7 +179,8 @@ export function buildAgentRequest(input: AiRequestInput, history?: api.AiMessage
]; ];
const params = actionParams(input.action); const params = actionParams(input.action);
const maxTokens = input.config.enableThinking ? Math.max(params.maxTokens, 8192) : params.maxTokens; const baseMaxTokens = input.config.maxOutputTokens ?? params.maxTokens;
const maxTokens = input.config.enableThinking ? Math.max(baseMaxTokens, 8192) : baseMaxTokens;
return { messages, systemPrompt, taskContract, maxTokens }; return { messages, systemPrompt, taskContract, maxTokens };
} }
@@ -669,6 +669,12 @@ describe("settingsStore AI API key normalization", () => {
it("trims API keys when normalizing loaded configurations", () => { it("trims API keys when normalizing loaded configurations", () => {
expect(normalizeAiConfig({ provider: "openai", apiKey: " secret " }).apiKey).toBe("secret"); expect(normalizeAiConfig({ provider: "openai", apiKey: " secret " }).apiKey).toBe("secret");
}); });
it("preserves and bounds a configured maximum output token budget", () => {
expect(normalizeAiConfig({ provider: "deepseek", maxOutputTokens: 32768 }).maxOutputTokens).toBe(32768);
expect(normalizeAiConfig({ provider: "deepseek", maxOutputTokens: 1_500_000 }).maxOutputTokens).toBe(1_000_000);
expect(normalizeAiConfig({ provider: "deepseek", maxOutputTokens: 128 }).maxOutputTokens).toBeUndefined();
expect(normalizeAiConfig({ provider: "deepseek", maxOutputTokens: Number.NaN }).maxOutputTokens).toBeUndefined();
});
it("provides Kimi defaults and recognizes legacy Kimi configurations", () => { it("provides Kimi defaults and recognizes legacy Kimi configurations", () => {
expect(AI_PROVIDER_PRESETS.kimi).toMatchObject({ expect(AI_PROVIDER_PRESETS.kimi).toMatchObject({
+5
View File
@@ -458,6 +458,8 @@ const defaultConfigs: Record<AiProvider, Omit<AiConfig, "apiKey">> = Object.from
) as Record<AiProvider, Omit<AiConfig, "apiKey">>; ) as Record<AiProvider, Omit<AiConfig, "apiKey">>;
const AI_REASONING_LEVELS: AiReasoningLevel[] = ["default", "minimal", "low", "medium", "high", "xhigh", "max"]; const AI_REASONING_LEVELS: AiReasoningLevel[] = ["default", "minimal", "low", "medium", "high", "xhigh", "max"];
export const AI_OUTPUT_TOKENS_MIN = 256;
export const AI_OUTPUT_TOKENS_MAX = 1_000_000;
const AI_ENV_KEY_RE = /^[A-Za-z_][A-Za-z0-9_]*$/; const AI_ENV_KEY_RE = /^[A-Za-z_][A-Za-z0-9_]*$/;
function normalizeAiReasoningLevel(value: unknown): AiReasoningLevel { function normalizeAiReasoningLevel(value: unknown): AiReasoningLevel {
@@ -488,6 +490,8 @@ export function normalizeAiHeaders(value: unknown): Record<string, string> {
export function normalizeAiConfig(config: Partial<AiConfig> | null | undefined): AiConfig { export function normalizeAiConfig(config: Partial<AiConfig> | null | undefined): AiConfig {
const provider = config?.provider && config.provider in AI_PROVIDER_PRESETS ? config.provider : inferAiProviderFromConfig(config); const provider = config?.provider && config.provider in AI_PROVIDER_PRESETS ? config.provider : inferAiProviderFromConfig(config);
const rawMaxOutputTokens = config?.maxOutputTokens;
const maxOutputTokens = typeof rawMaxOutputTokens === "number" && Number.isFinite(rawMaxOutputTokens) && rawMaxOutputTokens >= AI_OUTPUT_TOKENS_MIN ? Math.min(AI_OUTPUT_TOKENS_MAX, Math.round(rawMaxOutputTokens)) : undefined;
return { return {
...defaultConfigs[provider], ...defaultConfigs[provider],
...config, ...config,
@@ -500,6 +504,7 @@ export function normalizeAiConfig(config: Partial<AiConfig> | null | undefined):
proxyUrl: config?.proxyUrl ?? "", proxyUrl: config?.proxyUrl ?? "",
enableThinking: config?.enableThinking ?? true, enableThinking: config?.enableThinking ?? true,
reasoningLevel: normalizeAiReasoningLevel(config?.reasoningLevel), reasoningLevel: normalizeAiReasoningLevel(config?.reasoningLevel),
maxOutputTokens,
contextWindow: config?.contextWindow ?? undefined, contextWindow: config?.contextWindow ?? undefined,
codexCliPath: config?.codexCliPath?.trim() || undefined, codexCliPath: config?.codexCliPath?.trim() || undefined,
codexCliEnv: normalizeAiEnv(config?.codexCliEnv), codexCliEnv: normalizeAiEnv(config?.codexCliEnv),
+2
View File
@@ -42,6 +42,8 @@ export interface AiConfig {
proxyUrl?: string; proxyUrl?: string;
enableThinking?: boolean; enableThinking?: boolean;
reasoningLevel?: AiReasoningLevel; reasoningLevel?: AiReasoningLevel;
/** Optional per-configuration output budget sent as max_tokens/max_output_tokens. */
maxOutputTokens?: number;
contextWindow?: number; contextWindow?: number;
codexCliPath?: string | null; codexCliPath?: string | null;
codexCliEnv?: Record<string, string>; codexCliEnv?: Record<string, string>;
+142 -47
View File
@@ -399,6 +399,9 @@ pub struct AiConfig {
/// persist this field; it is attached to a runtime config clone. /// persist this field; it is attached to a runtime config clone.
#[serde(default, skip_serializing_if = "Option::is_none")] #[serde(default, skip_serializing_if = "Option::is_none")]
pub runtime_effort: Option<AiEffortSelection>, pub runtime_effort: Option<AiEffortSelection>,
/// Optional output token budget configured for this provider.
#[serde(default)]
pub max_output_tokens: Option<u32>,
#[serde(default)] #[serde(default)]
pub context_window: Option<u32>, pub context_window: Option<u32>,
/// Maximum number of automatic retries on transient API errors /// Maximum number of automatic retries on transient API errors
@@ -1214,8 +1217,12 @@ pub fn responses_stream_text(event: &serde_json::Value) -> Option<&str> {
event["delta"].as_str().filter(|s| !s.is_empty()) event["delta"].as_str().filter(|s| !s.is_empty())
} }
fn responses_max_output_tokens(max_tokens: Option<u32>) -> u32 { fn output_token_limit(max_tokens: Option<u32>, config: &AiConfig, default: u32) -> u32 {
max_tokens.unwrap_or(2048).max(16) max_tokens.or(config.max_output_tokens).unwrap_or(default)
}
fn responses_max_output_tokens(max_tokens: Option<u32>, config: &AiConfig) -> u32 {
output_token_limit(max_tokens, config, 2048).max(16)
} }
fn responses_token_usage(event: &serde_json::Value) -> Option<TokenUsage> { fn responses_token_usage(event: &serde_json::Value) -> Option<TokenUsage> {
@@ -1247,7 +1254,8 @@ fn uses_chat_completion_max_completion_tokens(config: &AiConfig) -> bool {
|| is_openai_api_config(config) && is_openai_reasoning_model(&config.model) || is_openai_api_config(config) && is_openai_reasoning_model(&config.model)
} }
fn set_chat_completion_token_limit(body: &mut serde_json::Value, config: &AiConfig, max_tokens: u32) { fn set_chat_completion_token_limit(body: &mut serde_json::Value, config: &AiConfig, max_tokens: Option<u32>) {
let max_tokens = output_token_limit(max_tokens, config, 4096);
if uses_chat_completion_max_completion_tokens(config) { if uses_chat_completion_max_completion_tokens(config) {
body["max_completion_tokens"] = json!(max_tokens); body["max_completion_tokens"] = json!(max_tokens);
} else { } else {
@@ -1261,7 +1269,7 @@ fn apply_minimax_chat_completion_fields(body: &mut serde_json::Value, config: &A
} }
} }
fn decorate_chat_completion_body(body: &mut serde_json::Value, config: &AiConfig, max_tokens: u32) { fn decorate_chat_completion_body(body: &mut serde_json::Value, config: &AiConfig, max_tokens: Option<u32>) {
set_chat_completion_token_limit(body, config, max_tokens); set_chat_completion_token_limit(body, config, max_tokens);
apply_minimax_chat_completion_fields(body, config); apply_minimax_chat_completion_fields(body, config);
apply_chat_completion_thinking_toggle(body, config); apply_chat_completion_thinking_toggle(body, config);
@@ -2133,7 +2141,7 @@ pub async fn resolve_model_effort_core(config: &AiConfig, model_id: &str) -> Res
pub async fn call_claude(client: &reqwest::Client, request: AiCompletionRequest) -> Result<String, String> { pub async fn call_claude(client: &reqwest::Client, request: AiCompletionRequest) -> Result<String, String> {
let mut body = json!({ let mut body = json!({
"model": claude_http_model(&request.config.model), "model": claude_http_model(&request.config.model),
"max_tokens": request.max_tokens.unwrap_or(2048), "max_tokens": output_token_limit(request.max_tokens, &request.config, 2048),
"system": claude_system_prompt(&request.system_prompt), "system": claude_system_prompt(&request.system_prompt),
"messages": claude_messages(&request.messages), "messages": claude_messages(&request.messages),
}); });
@@ -2213,7 +2221,7 @@ pub async fn call_openai_compatible(client: &reqwest::Client, request: AiComplet
"model": request.config.model, "model": request.config.model,
"messages": messages, "messages": messages,
}); });
decorate_chat_completion_body(&mut body_obj, &request.config, request.max_tokens.unwrap_or(2048)); decorate_chat_completion_body(&mut body_obj, &request.config, request.max_tokens);
let res = client let res = client
.post(resolve_endpoint(&request.config)) .post(resolve_endpoint(&request.config))
@@ -2238,7 +2246,7 @@ pub async fn call_responses_api(client: &reqwest::Client, request: AiCompletionR
let mut body = json!({ let mut body = json!({
"model": request.config.model, "model": request.config.model,
"input": build_responses_input(&request.system_prompt, &request.messages), "input": build_responses_input(&request.system_prompt, &request.messages),
"max_output_tokens": responses_max_output_tokens(request.max_tokens), "max_output_tokens": responses_max_output_tokens(request.max_tokens, &request.config),
}); });
crate::ai_effort::apply_runtime_effort(&mut body, &request.config); crate::ai_effort::apply_runtime_effort(&mut body, &request.config);
@@ -2268,7 +2276,7 @@ pub async fn call_gemini(client: &reqwest::Client, request: AiCompletionRequest)
}, },
"contents": contents, "contents": contents,
"generationConfig": { "generationConfig": {
"maxOutputTokens": request.max_tokens.unwrap_or(2048), "maxOutputTokens": output_token_limit(request.max_tokens, &request.config, 2048),
}, },
}); });
crate::ai_effort::apply_runtime_effort(&mut body, &request.config); crate::ai_effort::apply_runtime_effort(&mut body, &request.config);
@@ -2787,7 +2795,7 @@ pub async fn test_connection_core(config: &AiConfig) -> Result<AiTestConnectionR
"messages": messages, "messages": messages,
"stream": true, "stream": true,
}); });
decorate_chat_completion_body(&mut body, &config_inner, 16); decorate_chat_completion_body(&mut body, &config_inner, Some(16));
body body
}; };
let headers = maybe_bearer_headers(&config_inner)?; let headers = maybe_bearer_headers(&config_inner)?;
@@ -3167,7 +3175,7 @@ async fn stream_claude(
) -> Result<(), String> { ) -> Result<(), String> {
let mut body = json!({ let mut body = json!({
"model": claude_http_model(&request.config.model), "model": claude_http_model(&request.config.model),
"max_tokens": request.max_tokens.unwrap_or(2048), "max_tokens": output_token_limit(request.max_tokens, &request.config, 2048),
"system": claude_system_prompt(&request.system_prompt), "system": claude_system_prompt(&request.system_prompt),
"messages": claude_messages(&request.messages), "messages": claude_messages(&request.messages),
"stream": true, "stream": true,
@@ -3275,7 +3283,7 @@ async fn stream_openai(
"messages": messages, "messages": messages,
"stream": true, "stream": true,
}); });
decorate_chat_completion_body(&mut body_obj, &request.config, request.max_tokens.unwrap_or(2048)); decorate_chat_completion_body(&mut body_obj, &request.config, request.max_tokens);
let endpoint = resolve_endpoint(&request.config); let endpoint = resolve_endpoint(&request.config);
let config = request.config.clone(); let config = request.config.clone();
@@ -3406,7 +3414,7 @@ async fn stream_responses_api(
let mut body = json!({ let mut body = json!({
"model": request.config.model, "model": request.config.model,
"input": build_responses_input(&request.system_prompt, &request.messages), "input": build_responses_input(&request.system_prompt, &request.messages),
"max_output_tokens": responses_max_output_tokens(request.max_tokens), "max_output_tokens": responses_max_output_tokens(request.max_tokens, &request.config),
"stream": true, "stream": true,
}); });
crate::ai_effort::apply_runtime_effort(&mut body, &request.config); crate::ai_effort::apply_runtime_effort(&mut body, &request.config);
@@ -3498,7 +3506,7 @@ async fn stream_gemini(
}, },
"contents": contents, "contents": contents,
"generationConfig": { "generationConfig": {
"maxOutputTokens": request.max_tokens.unwrap_or(2048), "maxOutputTokens": output_token_limit(request.max_tokens, &request.config, 2048),
}, },
}); });
crate::ai_effort::apply_runtime_effort(&mut body, &request.config); crate::ai_effort::apply_runtime_effort(&mut body, &request.config);
@@ -3733,7 +3741,7 @@ async fn stream_claude_with_tools(
let mut body = json!({ let mut body = json!({
"model": claude_http_model(&request.config.model), "model": claude_http_model(&request.config.model),
"max_tokens": request.max_tokens.unwrap_or(4096), "max_tokens": output_token_limit(request.max_tokens, &request.config, 4096),
"system": claude_system_prompt(&request.system_prompt), "system": claude_system_prompt(&request.system_prompt),
"messages": messages, "messages": messages,
"tools": tool_json, "tools": tool_json,
@@ -3918,7 +3926,7 @@ async fn stream_openai_with_tools(
"stream": true, "stream": true,
"stream_options": { "include_usage": true }, "stream_options": { "include_usage": true },
}); });
set_chat_completion_token_limit(&mut body, &request.config, request.max_tokens.unwrap_or(4096)); set_chat_completion_token_limit(&mut body, &request.config, request.max_tokens);
apply_minimax_chat_completion_fields(&mut body, &request.config); apply_minimax_chat_completion_fields(&mut body, &request.config);
if request.config.runtime_effort.is_none() && !request.config.enable_thinking { if request.config.runtime_effort.is_none() && !request.config.enable_thinking {
@@ -4094,7 +4102,7 @@ async fn stream_responses_with_tools(
let mut body = json!({ let mut body = json!({
"model": request.config.model, "model": request.config.model,
"input": build_responses_input_with_tools(&request.system_prompt, &request.messages), "input": build_responses_input_with_tools(&request.system_prompt, &request.messages),
"max_output_tokens": responses_max_output_tokens(request.max_tokens), "max_output_tokens": responses_max_output_tokens(request.max_tokens, &request.config),
"tools": tool_json, "tools": tool_json,
"tool_choice": "auto", "tool_choice": "auto",
"stream": true, "stream": true,
@@ -4346,7 +4354,7 @@ async fn stream_gemini_with_tools(
"systemInstruction": { "parts": [{ "text": request.system_prompt }] }, "systemInstruction": { "parts": [{ "text": request.system_prompt }] },
"tools": [{ "functionDeclarations": tool_declarations }], "tools": [{ "functionDeclarations": tool_declarations }],
"generationConfig": { "generationConfig": {
"maxOutputTokens": request.max_tokens.unwrap_or(4096), "maxOutputTokens": output_token_limit(request.max_tokens, &request.config, 4096),
} }
}); });
crate::ai_effort::apply_runtime_effort(&mut body, &request.config); crate::ai_effort::apply_runtime_effort(&mut body, &request.config);
@@ -4603,6 +4611,27 @@ mod tests {
assert_eq!(openai_message_content(&message), serde_json::json!(marker)); assert_eq!(openai_message_content(&message), serde_json::json!(marker));
} }
#[test]
fn ai_config_max_output_tokens_uses_camel_case_and_defaults_to_none() {
let configured: AiConfig = serde_json::from_value(serde_json::json!({
"provider": "deepseek",
"model": "deepseek-v4-flash",
"maxOutputTokens": 32_768
}))
.unwrap();
assert_eq!(configured.max_output_tokens, Some(32_768));
let legacy: AiConfig = serde_json::from_value(serde_json::json!({
"provider": "deepseek",
"model": "deepseek-v4-flash"
}))
.unwrap();
assert_eq!(legacy.max_output_tokens, None);
let serialized = serde_json::to_value(configured).unwrap();
assert_eq!(serialized["maxOutputTokens"], serde_json::json!(32_768));
}
#[test] #[test]
fn ai_chat_selection_default_mode_serde() { fn ai_chat_selection_default_mode_serde() {
// Old blobs without a defaultMode field load as None (no migration needed). // Old blobs without a defaultMode field load as None (no migration needed).
@@ -4842,6 +4871,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: Some(1_000_000), context_window: Some(1_000_000),
max_retries: None, max_retries: None,
@@ -5505,6 +5535,7 @@ mod tests {
proxy_url: "not a proxy url".to_string(), proxy_url: "not a proxy url".to_string(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -5546,6 +5577,7 @@ mod tests {
proxy_url: "127.0.0.1:7890".to_string(), proxy_url: "127.0.0.1:7890".to_string(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -5585,6 +5617,7 @@ mod tests {
proxy_url: "not a proxy url".to_string(), proxy_url: "not a proxy url".to_string(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -5624,6 +5657,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -5667,6 +5701,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -5707,6 +5742,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -5731,7 +5767,7 @@ mod tests {
for provider in for provider in
[AiProvider::Ollama, AiProvider::AnthropicCompatible, AiProvider::OpenaiCompatible, AiProvider::Custom] [AiProvider::Ollama, AiProvider::AnthropicCompatible, AiProvider::OpenaiCompatible, AiProvider::Custom]
{ {
let config = AiConfig { provider, ..base.clone() }; let config = AiConfig { max_output_tokens: None, provider, ..base.clone() };
assert!(validate_config(&config).is_ok()); assert!(validate_config(&config).is_ok());
assert!(validate_model_list_config(&config).is_ok()); assert!(validate_model_list_config(&config).is_ok());
assert!(maybe_bearer_headers(&config).unwrap().get(AUTHORIZATION).is_none()); assert!(maybe_bearer_headers(&config).unwrap().get(AUTHORIZATION).is_none());
@@ -5745,7 +5781,7 @@ mod tests {
AiProvider::Qwen, AiProvider::Qwen,
AiProvider::MiniMax, AiProvider::MiniMax,
] { ] {
let config = AiConfig { provider, ..base.clone() }; let config = AiConfig { max_output_tokens: None, provider, ..base.clone() };
assert_eq!(validate_config(&config).unwrap_err(), "API key is required"); assert_eq!(validate_config(&config).unwrap_err(), "API key is required");
assert_eq!(validate_model_list_config(&config).unwrap_err(), "API key is required"); assert_eq!(validate_model_list_config(&config).unwrap_err(), "API key is required");
} }
@@ -5766,6 +5802,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -5801,6 +5838,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -5839,6 +5877,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -5864,19 +5903,24 @@ mod tests {
assert_eq!(resolve_endpoint(&config), "https://gateway.example.com/anthropic/v1/messages"); assert_eq!(resolve_endpoint(&config), "https://gateway.example.com/anthropic/v1/messages");
assert_eq!(resolve_model_list_endpoint(&config).unwrap(), "https://gateway.example.com/anthropic/v1/models"); assert_eq!(resolve_model_list_endpoint(&config).unwrap(), "https://gateway.example.com/anthropic/v1/models");
let full_messages = let full_messages = AiConfig {
AiConfig { endpoint: "https://gateway.example.com/anthropic/v1/messages".to_string(), ..config.clone() }; max_output_tokens: None,
endpoint: "https://gateway.example.com/anthropic/v1/messages".to_string(),
..config.clone()
};
assert_eq!(resolve_endpoint(&full_messages), "https://gateway.example.com/anthropic/v1/messages"); assert_eq!(resolve_endpoint(&full_messages), "https://gateway.example.com/anthropic/v1/messages");
assert_eq!( assert_eq!(
resolve_model_list_endpoint(&full_messages).unwrap(), resolve_model_list_endpoint(&full_messages).unwrap(),
"https://gateway.example.com/anthropic/v1/models" "https://gateway.example.com/anthropic/v1/models"
); );
let bare_origin = AiConfig { endpoint: "https://gateway.example.com".to_string(), ..config.clone() }; let bare_origin =
AiConfig { max_output_tokens: None, endpoint: "https://gateway.example.com".to_string(), ..config.clone() };
assert_eq!(resolve_endpoint(&bare_origin), "https://gateway.example.com/v1/messages"); assert_eq!(resolve_endpoint(&bare_origin), "https://gateway.example.com/v1/messages");
assert_eq!(resolve_model_list_endpoint(&bare_origin).unwrap(), "https://gateway.example.com/v1/models"); assert_eq!(resolve_model_list_endpoint(&bare_origin).unwrap(), "https://gateway.example.com/v1/models");
let kimi_coding = AiConfig { let kimi_coding = AiConfig {
max_output_tokens: None,
endpoint: "https://api.kimi.com/coding/".to_string(), endpoint: "https://api.kimi.com/coding/".to_string(),
model: "kimi-for-coding".to_string(), model: "kimi-for-coding".to_string(),
..config.clone() ..config.clone()
@@ -5893,16 +5937,26 @@ mod tests {
assert_eq!(resolve_endpoint(&config), "https://gateway.example.com/v1/messages"); assert_eq!(resolve_endpoint(&config), "https://gateway.example.com/v1/messages");
assert_eq!(resolve_model_list_endpoint(&config).unwrap(), "https://gateway.example.com/v1/models"); assert_eq!(resolve_model_list_endpoint(&config).unwrap(), "https://gateway.example.com/v1/models");
let v1 = AiConfig { endpoint: "https://gateway.example.com/v1".to_string(), ..config.clone() }; let v1 = AiConfig {
max_output_tokens: None,
endpoint: "https://gateway.example.com/v1".to_string(),
..config.clone()
};
assert_eq!(resolve_endpoint(&v1), "https://gateway.example.com/v1/messages"); assert_eq!(resolve_endpoint(&v1), "https://gateway.example.com/v1/messages");
let nested = let nested = AiConfig {
AiConfig { endpoint: "https://gateway.example.com/anthropic/v1/messages".to_string(), ..config.clone() }; max_output_tokens: None,
endpoint: "https://gateway.example.com/anthropic/v1/messages".to_string(),
..config.clone()
};
assert_eq!(resolve_endpoint(&nested), "https://gateway.example.com/anthropic/v1/messages"); assert_eq!(resolve_endpoint(&nested), "https://gateway.example.com/anthropic/v1/messages");
assert_eq!(resolve_model_list_endpoint(&nested).unwrap(), "https://gateway.example.com/anthropic/v1/models"); assert_eq!(resolve_model_list_endpoint(&nested).unwrap(), "https://gateway.example.com/anthropic/v1/models");
let dashscope = let dashscope = AiConfig {
AiConfig { endpoint: "https://dashscope-intl.aliyuncs.com/apps/anthropic".to_string(), ..config.clone() }; max_output_tokens: None,
endpoint: "https://dashscope-intl.aliyuncs.com/apps/anthropic".to_string(),
..config.clone()
};
assert_eq!(resolve_endpoint(&dashscope), "https://dashscope-intl.aliyuncs.com/apps/anthropic/v1/messages"); assert_eq!(resolve_endpoint(&dashscope), "https://dashscope-intl.aliyuncs.com/apps/anthropic/v1/messages");
assert_eq!( assert_eq!(
resolve_model_list_endpoint(&dashscope).unwrap(), resolve_model_list_endpoint(&dashscope).unwrap(),
@@ -5929,6 +5983,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -5955,14 +6010,15 @@ mod tests {
assert_eq!(resolve_model_list_endpoint(&config).unwrap(), "https://api.minimax.io/v1/models"); assert_eq!(resolve_model_list_endpoint(&config).unwrap(), "https://api.minimax.io/v1/models");
assert!(provider_requires_api_key(&config.provider)); assert!(provider_requires_api_key(&config.provider));
let china = AiConfig { endpoint: "https://api.minimaxi.com/v1".to_string(), ..config.clone() }; let china =
AiConfig { max_output_tokens: None, endpoint: "https://api.minimaxi.com/v1".to_string(), ..config.clone() };
assert_eq!(resolve_endpoint(&china), "https://api.minimaxi.com/v1/chat/completions"); assert_eq!(resolve_endpoint(&china), "https://api.minimaxi.com/v1/chat/completions");
assert_eq!(resolve_model_list_endpoint(&china).unwrap(), "https://api.minimaxi.com/v1/models"); assert_eq!(resolve_model_list_endpoint(&china).unwrap(), "https://api.minimaxi.com/v1/models");
let no_key = AiConfig { api_key: String::new(), ..config.clone() }; let no_key = AiConfig { max_output_tokens: None, api_key: String::new(), ..config.clone() };
assert_eq!(validate_config(&no_key).unwrap_err(), "API key is required"); assert_eq!(validate_config(&no_key).unwrap_err(), "API key is required");
let responses = AiConfig { api_style: AiApiStyle::Responses, ..config.clone() }; let responses = AiConfig { max_output_tokens: None, api_style: AiApiStyle::Responses, ..config.clone() };
assert_eq!( assert_eq!(
validate_config(&responses).unwrap_err(), validate_config(&responses).unwrap_err(),
"MiniMax currently supports the Chat Completions API style in DBX; select Completions and retry" "MiniMax currently supports the Chat Completions API style in DBX; select Completions and retry"
@@ -5976,6 +6032,7 @@ mod tests {
#[test] #[test]
fn kimi_provider_uses_openai_compatible_endpoints_and_requires_an_api_key() { fn kimi_provider_uses_openai_compatible_endpoints_and_requires_an_api_key() {
let config = AiConfig { let config = AiConfig {
max_output_tokens: None,
provider: AiProvider::Kimi, provider: AiProvider::Kimi,
api_key: "key".to_string(), api_key: "key".to_string(),
auth_method: AiAuthMethod::Bearer, auth_method: AiAuthMethod::Bearer,
@@ -6013,6 +6070,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -6037,25 +6095,31 @@ mod tests {
assert_eq!(resolve_model_list_endpoint(&config).unwrap(), "https://api.example.com/v1/models"); assert_eq!(resolve_model_list_endpoint(&config).unwrap(), "https://api.example.com/v1/models");
// Endpoint with /v1 already present — no change // Endpoint with /v1 already present — no change
let config_v1 = AiConfig { endpoint: "https://api.example.com/v1".to_string(), ..config.clone() }; let config_v1 =
AiConfig { max_output_tokens: None, endpoint: "https://api.example.com/v1".to_string(), ..config.clone() };
assert_eq!(resolve_endpoint(&config_v1), "https://api.example.com/v1/chat/completions"); assert_eq!(resolve_endpoint(&config_v1), "https://api.example.com/v1/chat/completions");
assert_eq!(resolve_model_list_endpoint(&config_v1).unwrap(), "https://api.example.com/v1/models"); assert_eq!(resolve_model_list_endpoint(&config_v1).unwrap(), "https://api.example.com/v1/models");
// Endpoint with /v2 — no change // Endpoint with /v2 — no change
let config_v2 = AiConfig { endpoint: "https://api.example.com/v2".to_string(), ..config.clone() }; let config_v2 =
AiConfig { max_output_tokens: None, endpoint: "https://api.example.com/v2".to_string(), ..config.clone() };
assert_eq!(resolve_endpoint(&config_v2), "https://api.example.com/v2/chat/completions"); assert_eq!(resolve_endpoint(&config_v2), "https://api.example.com/v2/chat/completions");
// Full path already specified — no change // Full path already specified — no change
let config_full = let config_full = AiConfig {
AiConfig { endpoint: "https://api.openai.com/v1/chat/completions".to_string(), ..config.clone() }; max_output_tokens: None,
endpoint: "https://api.openai.com/v1/chat/completions".to_string(),
..config.clone()
};
assert_eq!(resolve_endpoint(&config_full), "https://api.openai.com/v1/chat/completions"); assert_eq!(resolve_endpoint(&config_full), "https://api.openai.com/v1/chat/completions");
// Responses API style with /v1 missing // Responses API style with /v1 missing
let config_responses = AiConfig { api_style: AiApiStyle::Responses, ..config.clone() }; let config_responses = AiConfig { max_output_tokens: None, api_style: AiApiStyle::Responses, ..config.clone() };
assert_eq!(resolve_endpoint(&config_responses), "https://api.example.com/v1/responses"); assert_eq!(resolve_endpoint(&config_responses), "https://api.example.com/v1/responses");
// Ollama preset already has /v1 — no change // Ollama preset already has /v1 — no change
let ollama = AiConfig { let ollama = AiConfig {
max_output_tokens: None,
provider: AiProvider::Ollama, provider: AiProvider::Ollama,
endpoint: "http://localhost:11434/v1".to_string(), endpoint: "http://localhost:11434/v1".to_string(),
..config.clone() ..config.clone()
@@ -6063,12 +6127,14 @@ mod tests {
assert_eq!(resolve_endpoint(&ollama), "http://localhost:11434/v1/chat/completions"); assert_eq!(resolve_endpoint(&ollama), "http://localhost:11434/v1/chat/completions");
// Custom path without /v1 — left alone (CC-Switch strategy: only bare origin gets auto /v1) // Custom path without /v1 — left alone (CC-Switch strategy: only bare origin gets auto /v1)
let custom_path = AiConfig { endpoint: "https://my-gateway.com/api".to_string(), ..config.clone() }; let custom_path =
AiConfig { max_output_tokens: None, endpoint: "https://my-gateway.com/api".to_string(), ..config.clone() };
assert_eq!(resolve_endpoint(&custom_path), "https://my-gateway.com/api/chat/completions"); assert_eq!(resolve_endpoint(&custom_path), "https://my-gateway.com/api/chat/completions");
assert_eq!(resolve_model_list_endpoint(&custom_path).unwrap(), "https://my-gateway.com/api/models"); assert_eq!(resolve_model_list_endpoint(&custom_path).unwrap(), "https://my-gateway.com/api/models");
// Bare host with port — add /v1 // Bare host with port — add /v1
let bare_with_port = AiConfig { endpoint: "http://localhost:8080".to_string(), ..config.clone() }; let bare_with_port =
AiConfig { max_output_tokens: None, endpoint: "http://localhost:8080".to_string(), ..config.clone() };
assert_eq!(resolve_endpoint(&bare_with_port), "http://localhost:8080/v1/chat/completions"); assert_eq!(resolve_endpoint(&bare_with_port), "http://localhost:8080/v1/chat/completions");
} }
@@ -6087,6 +6153,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -6111,6 +6178,7 @@ mod tests {
assert_eq!(resolve_endpoint(&config), "https://api.openai.com/v1/responses"); assert_eq!(resolve_endpoint(&config), "https://api.openai.com/v1/responses");
let completions = AiConfig { let completions = AiConfig {
max_output_tokens: None,
endpoint: "https://api.openai.com/v1/responses".to_string(), endpoint: "https://api.openai.com/v1/responses".to_string(),
api_style: AiApiStyle::Completions, api_style: AiApiStyle::Completions,
custom_headers: Default::default(), custom_headers: Default::default(),
@@ -6134,6 +6202,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -6304,6 +6373,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: false, enable_thinking: false,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -6326,7 +6396,8 @@ mod tests {
}; };
assert_eq!(resolve_ollama_show_endpoint(&config).unwrap(), "http://localhost:11434/api/show"); assert_eq!(resolve_ollama_show_endpoint(&config).unwrap(), "http://localhost:11434/api/show");
let prefixed = AiConfig { endpoint: "https://example.com/ollama/v1".to_string(), ..config }; let prefixed =
AiConfig { max_output_tokens: None, endpoint: "https://example.com/ollama/v1".to_string(), ..config };
assert_eq!(resolve_ollama_show_endpoint(&prefixed).unwrap(), "https://example.com/ollama/api/show"); assert_eq!(resolve_ollama_show_endpoint(&prefixed).unwrap(), "https://example.com/ollama/api/show");
} }
@@ -6347,6 +6418,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: false, enable_thinking: false,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -6439,6 +6511,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: false, enable_thinking: false,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -6473,10 +6546,15 @@ mod tests {
#[test] #[test]
fn responses_api_clamps_tiny_output_token_requests() { fn responses_api_clamps_tiny_output_token_requests() {
assert_eq!(responses_max_output_tokens(Some(1)), 16); let mut config = minimax_test_config("https://api.minimaxi.com/v1");
assert_eq!(responses_max_output_tokens(Some(16)), 16); assert_eq!(responses_max_output_tokens(Some(1), &config), 16);
assert_eq!(responses_max_output_tokens(Some(2400)), 2400); assert_eq!(responses_max_output_tokens(Some(16), &config), 16);
assert_eq!(responses_max_output_tokens(None), 2048); assert_eq!(responses_max_output_tokens(Some(2400), &config), 2400);
assert_eq!(responses_max_output_tokens(None, &config), 2048);
config.max_output_tokens = Some(32_768);
assert_eq!(responses_max_output_tokens(None, &config), 32_768);
assert_eq!(responses_max_output_tokens(Some(2400), &config), 2400);
} }
#[test] #[test]
@@ -6717,6 +6795,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -6743,7 +6822,7 @@ mod tests {
"messages": [{ "role": "user", "content": TEST_PROMPT }], "messages": [{ "role": "user", "content": TEST_PROMPT }],
"stream": true, "stream": true,
}); });
set_chat_completion_token_limit(&mut body, &config, 1024); set_chat_completion_token_limit(&mut body, &config, Some(1024));
assert_eq!(body.get("max_completion_tokens"), Some(&serde_json::json!(1024))); assert_eq!(body.get("max_completion_tokens"), Some(&serde_json::json!(1024)));
assert!(body.get("max_tokens").is_none()); assert!(body.get("max_tokens").is_none());
@@ -6754,7 +6833,7 @@ mod tests {
"messages": [{ "role": "user", "content": TEST_PROMPT }], "messages": [{ "role": "user", "content": TEST_PROMPT }],
"stream": true, "stream": true,
}); });
set_chat_completion_token_limit(&mut body, &config, 1024); set_chat_completion_token_limit(&mut body, &config, Some(1024));
assert_eq!(body.get("max_tokens"), Some(&serde_json::json!(1024))); assert_eq!(body.get("max_tokens"), Some(&serde_json::json!(1024)));
assert!(body.get("max_completion_tokens").is_none()); assert!(body.get("max_completion_tokens").is_none());
@@ -6766,7 +6845,7 @@ mod tests {
"messages": [{ "role": "user", "content": TEST_PROMPT }], "messages": [{ "role": "user", "content": TEST_PROMPT }],
"stream": true, "stream": true,
}); });
set_chat_completion_token_limit(&mut body, &config, 1024); set_chat_completion_token_limit(&mut body, &config, Some(1024));
assert_eq!(body.get("max_completion_tokens"), Some(&serde_json::json!(1024))); assert_eq!(body.get("max_completion_tokens"), Some(&serde_json::json!(1024)));
assert!(body.get("max_tokens").is_none()); assert!(body.get("max_tokens").is_none());
@@ -6779,10 +6858,19 @@ mod tests {
"messages": [{ "role": "user", "content": TEST_PROMPT }], "messages": [{ "role": "user", "content": TEST_PROMPT }],
"stream": true, "stream": true,
}); });
set_chat_completion_token_limit(&mut body, &config, 1024); set_chat_completion_token_limit(&mut body, &config, Some(1024));
assert_eq!(body.get("max_tokens"), Some(&serde_json::json!(1024))); assert_eq!(body.get("max_tokens"), Some(&serde_json::json!(1024)));
assert!(body.get("max_completion_tokens").is_none()); assert!(body.get("max_completion_tokens").is_none());
config.max_output_tokens = Some(32_768);
let mut body = serde_json::json!({
"model": &config.model,
"messages": [{ "role": "user", "content": TEST_PROMPT }],
"stream": true,
});
set_chat_completion_token_limit(&mut body, &config, None);
assert_eq!(body.get("max_tokens"), Some(&serde_json::json!(32_768)));
} }
#[test] #[test]
@@ -6939,6 +7027,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: false, enable_thinking: false,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -6990,6 +7079,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: false, enable_thinking: false,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -7044,6 +7134,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: false, enable_thinking: false,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -7092,6 +7183,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: false, enable_thinking: false,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -7127,7 +7219,7 @@ mod tests {
config.enable_thinking = false; config.enable_thinking = false;
let mut body = serde_json::json!({ "model": &config.model }); let mut body = serde_json::json!({ "model": &config.model });
decorate_chat_completion_body(&mut body, &config, 1024); decorate_chat_completion_body(&mut body, &config, Some(1024));
assert_eq!(body["reasoning_split"], true); assert_eq!(body["reasoning_split"], true);
assert_eq!(body["max_completion_tokens"], 1024); assert_eq!(body["max_completion_tokens"], 1024);
@@ -7484,6 +7576,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: false, enable_thinking: false,
reasoning_level: AiReasoningLevel::High, reasoning_level: AiReasoningLevel::High,
max_output_tokens: None,
runtime_effort: Some(AiEffortSelection::ProviderDefault), runtime_effort: Some(AiEffortSelection::ProviderDefault),
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -7559,6 +7652,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -7629,6 +7723,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: false, enable_thinking: false,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -573,6 +573,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
+1
View File
@@ -928,6 +928,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
+1
View File
@@ -1119,6 +1119,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
+1
View File
@@ -410,6 +410,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
+1
View File
@@ -469,6 +469,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
+1
View File
@@ -790,6 +790,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
+1
View File
@@ -456,6 +456,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
+1
View File
@@ -864,6 +864,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::High, reasoning_level: AiReasoningLevel::High,
max_output_tokens: None,
runtime_effort: Some(AiEffortSelection::Enum("high".to_string())), runtime_effort: Some(AiEffortSelection::Enum("high".to_string())),
context_window: None, context_window: None,
max_retries: None, max_retries: None,
+1
View File
@@ -532,6 +532,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
+1
View File
@@ -1639,6 +1639,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: crate::ai::AiReasoningLevel::Default, reasoning_level: crate::ai::AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
+1
View File
@@ -7582,6 +7582,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
+2
View File
@@ -526,6 +526,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: true, enable_thinking: true,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
@@ -633,6 +634,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: false, enable_thinking: false,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,
+1
View File
@@ -670,6 +670,7 @@ mod tests {
proxy_url: String::new(), proxy_url: String::new(),
enable_thinking: false, enable_thinking: false,
reasoning_level: AiReasoningLevel::Default, reasoning_level: AiReasoningLevel::Default,
max_output_tokens: None,
runtime_effort: None, runtime_effort: None,
context_window: None, context_window: None,
max_retries: None, max_retries: None,