@code-yeongyu/senpi 2026.8.13 → 2026.8.16

package/dist/core/extensions/builtin/reasoning/index.js

@@ -0,0 +1,211 @@
+import { SettingsManager } from "../../../settings-manager.js";
+import { classifyReasoningCapability } from "../../../thinking-levels.js";
+/**
+ * Capability-aware `/reasoning` (the on/off axis) and `/efforts` (the effort ladder).
+ *
+ * Both commands classify the CURRENT model on every invocation through
+ * `classifyReasoningCapability`, so a mid-session model switch is honored immediately and no
+ * capability is ever cached. Neither command opens a selector: status is notified as text so both
+ * work headless and over RPC.
+ */
+/** Effort ladder, ascending. `off` is deliberately absent: it is the /reasoning axis. */
+const EFFORT_LEVELS = ["minimal", "low", "medium", "high", "xhigh", "max"];
+/** Tokens `/efforts` recognizes grammatically; support is then checked per model. */
+const EFFORT_TOKENS = ["off", ...EFFORT_LEVELS];
+const REASONING_ARGUMENTS = ["on", "off"];
+const REASONING_USAGE = "Usage: /reasoning [on|off]";
+const EFFORTS_USAGE = "Usage: /efforts [minimal|low|medium|high|xhigh|max]";
+const DEFAULT_ON_LEVEL = "medium";
+function modelKey(model) {
+    return `${model.provider}/${model.id}`;
+}
+function noReasoningMessage(model) {
+    return `Model ${modelKey(model)} does not support reasoning.`;
+}
+function onOffOnlyMessage(model) {
+    return `Reasoning effort is not configurable for ${modelKey(model)}; this model supports on/off only. Use /reasoning on or /reasoning off.`;
+}
+function reasoningStatus(level) {
+    return level === "off" ? "Reasoning: off." : `Reasoning: on (${level}).`;
+}
+function effortsStatus(level, capability) {
+    return `Reasoning effort: ${level}. Available: ${capability.nonOffLevels.join(", ")}.`;
+}
+/** The single argument token, `""` for the no-arg form, or `null` when more than one token was given. */
+function parseSingleArgument(args) {
+    const trimmed = args.trim();
+    if (trimmed.length === 0)
+        return "";
+    const tokens = trimmed.split(/\s+/);
+    return tokens.length === 1 ? (tokens[0] ?? "") : null;
+}
+/**
+ * Clamp to the nearest supported non-off level, searching upward first and then downward — the
+ * same direction policy the session clamp uses, so a remembered level below a model's floor lands
+ * on that floor instead of silently disabling reasoning.
+ */
+function clampToNonOff(level, capability) {
+    const supported = new Set(capability.nonOffLevels);
+    if (supported.has(level))
+        return level;
+    const index = EFFORT_LEVELS.indexOf(level);
+    if (index === -1)
+        return capability.nonOffLevels[0];
+    for (let i = index; i < EFFORT_LEVELS.length; i++) {
+        const candidate = EFFORT_LEVELS[i];
+        if (candidate && supported.has(candidate))
+            return candidate;
+    }
+    for (let i = index - 1; i >= 0; i--) {
+        const candidate = EFFORT_LEVELS[i];
+        if (candidate && supported.has(candidate))
+            return candidate;
+    }
+    return capability.nonOffLevels[0];
+}
+/**
+ * Level to restore when reasoning is switched back on: this model's durable last non-off level,
+ * else a legacy non-off effective memory, else the global default when that is not "off", else
+ * medium — always clamped to what the model supports.
+ */
+function resolvePreferredOnLevel(ctx, model, capability) {
+    const settingsManager = SettingsManager.create(ctx.cwd, ctx.agentDir, { projectTrusted: ctx.isProjectTrusted() });
+    const lastOnLevel = settingsManager.getModelLastOnThinkingLevel(model.provider, model.id);
+    const remembered = settingsManager.getModelThinkingLevel(model.provider, model.id);
+    const globalDefault = settingsManager.getDefaultThinkingLevel();
+    const preferred = lastOnLevel ?? nonOff(remembered) ?? nonOff(globalDefault) ?? DEFAULT_ON_LEVEL;
+    return clampToNonOff(preferred, capability);
+}
+function createSettingsManager(ctx) {
+    return SettingsManager.create(ctx.cwd, ctx.agentDir, { projectTrusted: ctx.isProjectTrusted() });
+}
+function nonOff(level) {
+    return level !== undefined && level !== "off" ? level : undefined;
+}
+function toCompletions(values, prefix) {
+    const normalizedPrefix = prefix.trim();
+    const matches = values.filter((value) => value.startsWith(normalizedPrefix));
+    return matches.length > 0 ? matches.map((value) => ({ value, label: value })) : null;
+}
+export default function reasoningExtension(pi) {
+    // Completions run outside a command context, which carries no model, so the live model is
+    // tracked from the events that change it. Handlers always read ctx.model instead.
+    let currentModel;
+    pi.on("session_start", (_event, ctx) => {
+        currentModel = ctx.model;
+    });
+    pi.on("model_select", (event) => {
+        currentModel = event.model;
+    });
+    pi.registerCommand("reasoning", {
+        description: "Show or toggle reasoning for the current model",
+        argumentHint: "[on|off]",
+        getArgumentCompletions: (prefix) => toCompletions(REASONING_ARGUMENTS, prefix),
+        handler: async (args, ctx) => {
+            const model = ctx.model;
+            if (!model) {
+                ctx.ui.notify("No model is active.", "error");
+                return;
+            }
+            currentModel = model;
+            const argument = parseSingleArgument(args);
+            if (argument === null || (argument !== "" && !REASONING_ARGUMENTS.includes(argument))) {
+                ctx.ui.notify(REASONING_USAGE, "error");
+                return;
+            }
+            const capability = classifyReasoningCapability(model);
+            const currentLevel = pi.getThinkingLevel();
+            if (argument === "") {
+                ctx.ui.notify(reasoningStatus(currentLevel), "info");
+                return;
+            }
+            if (argument === "off") {
+                if (capability.kind === "none") {
+                    // Nothing to disable; the model is already, permanently, off.
+                    ctx.ui.notify(reasoningStatus("off"), "info");
+                    return;
+                }
+                if (capability.kind === "always-on") {
+                    ctx.ui.notify(`Reasoning cannot be disabled for ${modelKey(model)}.`, "error");
+                    return;
+                }
+                if (currentLevel !== "off") {
+                    const settingsManager = createSettingsManager(ctx);
+                    settingsManager.setModelLastOnThinkingLevel(model.provider, model.id, currentLevel);
+                    await settingsManager.flush();
+                }
+                pi.setThinkingLevel("off");
+                ctx.ui.notify(reasoningStatus(pi.getThinkingLevel()), "info");
+                return;
+            }
+            if (capability.kind === "none") {
+                ctx.ui.notify(noReasoningMessage(model), "error");
+                return;
+            }
+            // Already reasoning: re-persist the current level instead of overriding the user's choice.
+            if (currentLevel !== "off") {
+                pi.setThinkingLevel(currentLevel);
+                ctx.ui.notify(reasoningStatus(pi.getThinkingLevel()), "info");
+                return;
+            }
+            const target = resolvePreferredOnLevel(ctx, model, capability);
+            if (target === undefined) {
+                ctx.ui.notify(noReasoningMessage(model), "error");
+                return;
+            }
+            // Apply the clamped level to the session, but persist it without rewriting a stale raw
+            // last-on preference. If catalog support returns later, the original preference still wins.
+            pi.setSessionThinkingLevel(target);
+            const settingsManager = createSettingsManager(ctx);
+            settingsManager.setModelThinkingLevel(model.provider, model.id, target, { preserveLastOn: true });
+            settingsManager.setDefaultThinkingLevel(target);
+            await settingsManager.flush();
+            ctx.ui.notify(reasoningStatus(pi.getThinkingLevel()), "info");
+        },
+    });
+    pi.registerCommand("efforts", {
+        description: "Show or set the reasoning effort for the current model",
+        argumentHint: "[minimal|low|medium|high|xhigh|max]",
+        getArgumentCompletions: (prefix) => {
+            // Only a graded model has an effort ladder worth completing.
+            const capability = currentModel ? classifyReasoningCapability(currentModel) : undefined;
+            if (capability?.kind !== "graded")
+                return null;
+            return toCompletions(capability.nonOffLevels, prefix);
+        },
+        handler: async (args, ctx) => {
+            const model = ctx.model;
+            if (!model) {
+                ctx.ui.notify("No model is active.", "error");
+                return;
+            }
+            currentModel = model;
+            const argument = parseSingleArgument(args);
+            if (argument === null || (argument !== "" && !EFFORT_TOKENS.includes(argument))) {
+                ctx.ui.notify(EFFORTS_USAGE, "error");
+                return;
+            }
+            const capability = classifyReasoningCapability(model);
+            if (capability.kind === "none") {
+                ctx.ui.notify(noReasoningMessage(model), "error");
+                return;
+            }
+            if (capability.kind === "on-off") {
+                ctx.ui.notify(onOffOnlyMessage(model), "error");
+                return;
+            }
+            if (argument === "") {
+                ctx.ui.notify(effortsStatus(pi.getThinkingLevel(), capability), "info");
+                return;
+            }
+            const level = argument;
+            if (!capability.nonOffLevels.includes(level)) {
+                ctx.ui.notify(`Reasoning effort "${level}" is not supported by ${modelKey(model)}. Available: ${capability.nonOffLevels.join(", ")}.`, "error");
+                return;
+            }
+            pi.setThinkingLevel(level);
+            ctx.ui.notify(effortsStatus(pi.getThinkingLevel(), capability), "info");
+        },
+    });
+}
+//# sourceMappingURL=index.js.map

package/dist/core/extensions/builtin/reasoning/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../../src/core/extensions/builtin/reasoning/index.ts"],"names":[],"mappings":"AAGA,OAAO,EAAE,eAAe,EAAE,MAAM,8BAA8B,CAAC;AAC/D,OAAO,EAAE,2BAA2B,EAA4B,MAAM,6BAA6B,CAAC;AAGpG;;;;;;;GAOG;AAEH,yFAAyF;AACzF,MAAM,aAAa,GAA6B,CAAC,SAAS,EAAE,KAAK,EAAE,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC;AACrG,qFAAqF;AACrF,MAAM,aAAa,GAAsB,CAAC,KAAK,EAAE,GAAG,aAAa,CAAC,CAAC;AACnE,MAAM,mBAAmB,GAAG,CAAC,IAAI,EAAE,KAAK,CAAU,CAAC;AACnD,MAAM,eAAe,GAAG,4BAA4B,CAAC;AACrD,MAAM,aAAa,GAAG,qDAAqD,CAAC;AAC5E,MAAM,gBAAgB,GAAkB,QAAQ,CAAC;AAIjD,SAAS,QAAQ,CAAC,KAAiB;IAClC,OAAO,GAAG,KAAK,CAAC,QAAQ,IAAI,KAAK,CAAC,EAAE,EAAE,CAAC;AACxC,CAAC;AAED,SAAS,kBAAkB,CAAC,KAAiB;IAC5C,OAAO,SAAS,QAAQ,CAAC,KAAK,CAAC,8BAA8B,CAAC;AAC/D,CAAC;AAED,SAAS,gBAAgB,CAAC,KAAiB;IAC1C,OAAO,4CAA4C,QAAQ,CAAC,KAAK,CAAC,yEAAyE,CAAC;AAC7I,CAAC;AAED,SAAS,eAAe,CAAC,KAAoB;IAC5C,OAAO,KAAK,KAAK,KAAK,CAAC,CAAC,CAAC,iBAAiB,CAAC,CAAC,CAAC,kBAAkB,KAAK,IAAI,CAAC;AAC1E,CAAC;AAED,SAAS,aAAa,CAAC,KAAoB,EAAE,UAA+B;IAC3E,OAAO,qBAAqB,KAAK,gBAAgB,UAAU,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC;AACxF,CAAC;AAED,yGAAyG;AACzG,SAAS,mBAAmB,CAAC,IAAY;IACxC,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IAC5B,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IACpC,MAAM,MAAM,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;IACpC,OAAO,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;AACvD,CAAC;AAED;;;;GAIG;AACH,SAAS,aAAa,CAAC,KAAoB,EAAE,UAA+B;IAC3E,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC;IACnD,IAAI,SAAS,CAAC,GAAG,CAAC,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IACvC,MAAM,KAAK,GAAG,aAAa,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;IAC3C,IAAI,KAAK,KAAK,CAAC,CAAC;QAAE,OAAO,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;IACpD,KAAK,IAAI,CAAC,GAAG,KAAK,EAAE,CAAC,GAAG,aAAa,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACnD,MAAM,SAAS,GAAG,aAAa,CAAC,CAAC,CAAC,CAAC;QACnC,IAAI,SAAS,IAAI,SAAS,CAAC,GAAG,CAAC,SAAS,CAAC;YAAE,OAAO,SAAS,CAAC;IAC7D,CAAC;IACD,KAAK,IAAI,CAAC,GAAG,KAAK,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QACrC,MAAM,SAAS,GAAG,aAAa,CAAC,CAAC,CAAC,CAAC;QACnC,IAAI,SAAS,IAAI,SAAS,CAAC,GAAG,CAAC,SAAS,CAAC;YAAE,OAAO,SAAS,CAAC;IAC7D,CAAC;IACD,OAAO,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;AACnC,CAAC;AAED;;;;GAIG;AACH,SAAS,uBAAuB,CAC/B,GAA4B,EAC5B,KAAiB,EACjB,UAA+B;IAE/B,MAAM,eAAe,GAAG,eAAe,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,GAAG,CAAC,QAAQ,EAAE,EAAE,cAAc,EAAE,GAAG,CAAC,gBAAgB,EAAE,EAAE,CAAC,CAAC;IAClH,MAAM,WAAW,GAAG,eAAe,CAAC,2BAA2B,CAAC,KAAK,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,CAAC,CAAC;IAC1F,MAAM,UAAU,GAAG,eAAe,CAAC,qBAAqB,CAAC,KAAK,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,CAAC,CAAC;IACnF,MAAM,aAAa,GAAG,eAAe,CAAC,uBAAuB,EAAE,CAAC;IAChE,MAAM,SAAS,GAAG,WAAW,IAAI,MAAM,CAAC,UAAU,CAAC,IAAI,MAAM,CAAC,aAAa,CAAC,IAAI,gBAAgB,CAAC;IACjG,OAAO,aAAa,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC;AAC7C,CAAC;AAED,SAAS,qBAAqB,CAAC,GAA4B;IAC1D,OAAO,eAAe,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,GAAG,CAAC,QAAQ,EAAE,EAAE,cAAc,EAAE,GAAG,CAAC,gBAAgB,EAAE,EAAE,CAAC,CAAC;AAClG,CAAC;AAED,SAAS,MAAM,CAAC,KAAgC;IAC/C,OAAO,KAAK,KAAK,SAAS,IAAI,KAAK,KAAK,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;AACnE,CAAC;AAED,SAAS,aAAa,CAAC,MAAyB,EAAE,MAAc;IAC/D,MAAM,gBAAgB,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC;IACvC,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,UAAU,CAAC,gBAAgB,CAAC,CAAC,CAAC;IAC7E,OAAO,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;AACtF,CAAC;AAED,MAAM,CAAC,OAAO,UAAU,kBAAkB,CAAC,EAAgB;IAC1D,0FAA0F;IAC1F,kFAAkF;IAClF,IAAI,YAAoC,CAAC;IAEzC,EAAE,CAAC,EAAE,CAAC,eAAe,EAAE,CAAC,MAAM,EAAE,GAAG,EAAE,EAAE;QACtC,YAAY,GAAG,GAAG,CAAC,KAAK,CAAC;IAC1B,CAAC,CAAC,CAAC;IACH,EAAE,CAAC,EAAE,CAAC,cAAc,EAAE,CAAC,KAAK,EAAE,EAAE;QAC/B,YAAY,GAAG,KAAK,CAAC,KAAK,CAAC;IAC5B,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,eAAe,CAAC,WAAW,EAAE;QAC/B,WAAW,EAAE,gDAAgD;QAC7D,YAAY,EAAE,UAAU;QACxB,sBAAsB,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,aAAa,CAAC,mBAAmB,EAAE,MAAM,CAAC;QAC9E,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;YAC5B,MAAM,KAAK,GAAG,GAAG,CAAC,KAAK,CAAC;YACxB,IAAI,CAAC,KAAK,EAAE,CAAC;gBACZ,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,qBAAqB,EAAE,OAAO,CAAC,CAAC;gBAC9C,OAAO;YACR,CAAC;YACD,YAAY,GAAG,KAAK,CAAC;YAErB,MAAM,QAAQ,GAAG,mBAAmB,CAAC,IAAI,CAAC,CAAC;YAC3C,IAAI,QAAQ,KAAK,IAAI,IAAI,CAAC,QAAQ,KAAK,EAAE,IAAI,CAAC,mBAAmB,CAAC,QAAQ,CAAC,QAA6B,CAAC,CAAC,EAAE,CAAC;gBAC5G,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,eAAe,EAAE,OAAO,CAAC,CAAC;gBACxC,OAAO;YACR,CAAC;YAED,MAAM,UAAU,GAAG,2BAA2B,CAAC,KAAK,CAAC,CAAC;YACtD,MAAM,YAAY,GAAG,EAAE,CAAC,gBAAgB,EAAE,CAAC;YAE3C,IAAI,QAAQ,KAAK,EAAE,EAAE,CAAC;gBACrB,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,eAAe,CAAC,YAAY,CAAC,EAAE,MAAM,CAAC,CAAC;gBACrD,OAAO;YACR,CAAC;YAED,IAAI,QAAQ,KAAK,KAAK,EAAE,CAAC;gBACxB,IAAI,UAAU,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;oBAChC,8DAA8D;oBAC9D,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,eAAe,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC,CAAC;oBAC9C,OAAO;gBACR,CAAC;gBACD,IAAI,UAAU,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;oBACrC,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,oCAAoC,QAAQ,CAAC,KAAK,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;oBAC/E,OAAO;gBACR,CAAC;gBACD,IAAI,YAAY,KAAK,KAAK,EAAE,CAAC;oBAC5B,MAAM,eAAe,GAAG,qBAAqB,CAAC,GAAG,CAAC,CAAC;oBACnD,eAAe,CAAC,2BAA2B,CAAC,KAAK,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,EAAE,YAAY,CAAC,CAAC;oBACpF,MAAM,eAAe,CAAC,KAAK,EAAE,CAAC;gBAC/B,CAAC;gBACD,EAAE,CAAC,gBAAgB,CAAC,KAAK,CAAC,CAAC;gBAC3B,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,eAAe,CAAC,EAAE,CAAC,gBAAgB,EAAE,CAAC,EAAE,MAAM,CAAC,CAAC;gBAC9D,OAAO;YACR,CAAC;YAED,IAAI,UAAU,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;gBAChC,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,kBAAkB,CAAC,KAAK,CAAC,EAAE,OAAO,CAAC,CAAC;gBAClD,OAAO;YACR,CAAC;YAED,2FAA2F;YAC3F,IAAI,YAAY,KAAK,KAAK,EAAE,CAAC;gBAC5B,EAAE,CAAC,gBAAgB,CAAC,YAAY,CAAC,CAAC;gBAClC,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,eAAe,CAAC,EAAE,CAAC,gBAAgB,EAAE,CAAC,EAAE,MAAM,CAAC,CAAC;gBAC9D,OAAO;YACR,CAAC;YAED,MAAM,MAAM,GAAG,uBAAuB,CAAC,GAAG,EAAE,KAAK,EAAE,UAAU,CAAC,CAAC;YAC/D,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;gBAC1B,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,kBAAkB,CAAC,KAAK,CAAC,EAAE,OAAO,CAAC,CAAC;gBAClD,OAAO;YACR,CAAC;YAED,uFAAuF;YACvF,4FAA4F;YAC5F,EAAE,CAAC,uBAAuB,CAAC,MAAM,CAAC,CAAC;YACnC,MAAM,eAAe,GAAG,qBAAqB,CAAC,GAAG,CAAC,CAAC;YACnD,eAAe,CAAC,qBAAqB,CAAC,KAAK,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,EAAE,MAAM,EAAE,EAAE,cAAc,EAAE,IAAI,EAAE,CAAC,CAAC;YAClG,eAAe,CAAC,uBAAuB,CAAC,MAAM,CAAC,CAAC;YAChD,MAAM,eAAe,CAAC,KAAK,EAAE,CAAC;YAC9B,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,eAAe,CAAC,EAAE,CAAC,gBAAgB,EAAE,CAAC,EAAE,MAAM,CAAC,CAAC;QAC/D,CAAC;KACD,CAAC,CAAC;IAEH,EAAE,CAAC,eAAe,CAAC,SAAS,EAAE;QAC7B,WAAW,EAAE,wDAAwD;QACrE,YAAY,EAAE,qCAAqC;QACnD,sBAAsB,EAAE,CAAC,MAAM,EAAE,EAAE;YAClC,6DAA6D;YAC7D,MAAM,UAAU,GAAG,YAAY,CAAC,CAAC,CAAC,2BAA2B,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;YACxF,IAAI,UAAU,EAAE,IAAI,KAAK,QAAQ;gBAAE,OAAO,IAAI,CAAC;YAC/C,OAAO,aAAa,CAAC,UAAU,CAAC,YAAY,EAAE,MAAM,CAAC,CAAC;QACvD,CAAC;QACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;YAC5B,MAAM,KAAK,GAAG,GAAG,CAAC,KAAK,CAAC;YACxB,IAAI,CAAC,KAAK,EAAE,CAAC;gBACZ,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,qBAAqB,EAAE,OAAO,CAAC,CAAC;gBAC9C,OAAO;YACR,CAAC;YACD,YAAY,GAAG,KAAK,CAAC;YAErB,MAAM,QAAQ,GAAG,mBAAmB,CAAC,IAAI,CAAC,CAAC;YAC3C,IAAI,QAAQ,KAAK,IAAI,IAAI,CAAC,QAAQ,KAAK,EAAE,IAAI,CAAC,aAAa,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,EAAE,CAAC;gBACjF,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,aAAa,EAAE,OAAO,CAAC,CAAC;gBACtC,OAAO;YACR,CAAC;YAED,MAAM,UAAU,GAAG,2BAA2B,CAAC,KAAK,CAAC,CAAC;YACtD,IAAI,UAAU,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;gBAChC,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,kBAAkB,CAAC,KAAK,CAAC,EAAE,OAAO,CAAC,CAAC;gBAClD,OAAO;YACR,CAAC;YACD,IAAI,UAAU,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;gBAClC,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,gBAAgB,CAAC,KAAK,CAAC,EAAE,OAAO,CAAC,CAAC;gBAChD,OAAO;YACR,CAAC;YAED,IAAI,QAAQ,KAAK,EAAE,EAAE,CAAC;gBACrB,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,aAAa,CAAC,EAAE,CAAC,gBAAgB,EAAE,EAAE,UAAU,CAAC,EAAE,MAAM,CAAC,CAAC;gBACxE,OAAO;YACR,CAAC;YAED,MAAM,KAAK,GAAG,QAAyB,CAAC;YACxC,IAAI,CAAC,UAAU,CAAC,YAAY,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;gBAC9C,GAAG,CAAC,EAAE,CAAC,MAAM,CACZ,qBAAqB,KAAK,yBAAyB,QAAQ,CAAC,KAAK,CAAC,gBAAgB,UAAU,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,EACvH,OAAO,CACP,CAAC;gBACF,OAAO;YACR,CAAC;YAED,EAAE,CAAC,gBAAgB,CAAC,KAAK,CAAC,CAAC;YAC3B,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,aAAa,CAAC,EAAE,CAAC,gBAAgB,EAAE,EAAE,UAAU,CAAC,EAAE,MAAM,CAAC,CAAC;QACzE,CAAC;KACD,CAAC,CAAC;AACJ,CAAC","sourcesContent":["import type { ThinkingLevel } from \"@earendil-works/pi-agent-core\";\nimport type { Api, Model } from \"@earendil-works/pi-ai\";\nimport type { AutocompleteItem } from \"@earendil-works/pi-tui\";\nimport { SettingsManager } from \"../../../settings-manager.ts\";\nimport { classifyReasoningCapability, type ReasoningCapability } from \"../../../thinking-levels.ts\";\nimport type { ExtensionAPI, ExtensionCommandContext } from \"../../types.ts\";\n\n/**\n * Capability-aware `/reasoning` (the on/off axis) and `/efforts` (the effort ladder).\n *\n * Both commands classify the CURRENT model on every invocation through\n * `classifyReasoningCapability`, so a mid-session model switch is honored immediately and no\n * capability is ever cached. Neither command opens a selector: status is notified as text so both\n * work headless and over RPC.\n */\n\n/** Effort ladder, ascending. `off` is deliberately absent: it is the /reasoning axis. */\nconst EFFORT_LEVELS: readonly ThinkingLevel[] = [\"minimal\", \"low\", \"medium\", \"high\", \"xhigh\", \"max\"];\n/** Tokens `/efforts` recognizes grammatically; support is then checked per model. */\nconst EFFORT_TOKENS: readonly string[] = [\"off\", ...EFFORT_LEVELS];\nconst REASONING_ARGUMENTS = [\"on\", \"off\"] as const;\nconst REASONING_USAGE = \"Usage: /reasoning [on|off]\";\nconst EFFORTS_USAGE = \"Usage: /efforts [minimal|low|medium|high|xhigh|max]\";\nconst DEFAULT_ON_LEVEL: ThinkingLevel = \"medium\";\n\ntype ReasoningArgument = (typeof REASONING_ARGUMENTS)[number];\n\nfunction modelKey(model: Model<Api>): string {\n\treturn `${model.provider}/${model.id}`;\n}\n\nfunction noReasoningMessage(model: Model<Api>): string {\n\treturn `Model ${modelKey(model)} does not support reasoning.`;\n}\n\nfunction onOffOnlyMessage(model: Model<Api>): string {\n\treturn `Reasoning effort is not configurable for ${modelKey(model)}; this model supports on/off only. Use /reasoning on or /reasoning off.`;\n}\n\nfunction reasoningStatus(level: ThinkingLevel): string {\n\treturn level === \"off\" ? \"Reasoning: off.\" : `Reasoning: on (${level}).`;\n}\n\nfunction effortsStatus(level: ThinkingLevel, capability: ReasoningCapability): string {\n\treturn `Reasoning effort: ${level}. Available: ${capability.nonOffLevels.join(\", \")}.`;\n}\n\n/** The single argument token, `\"\"` for the no-arg form, or `null` when more than one token was given. */\nfunction parseSingleArgument(args: string): string | null {\n\tconst trimmed = args.trim();\n\tif (trimmed.length === 0) return \"\";\n\tconst tokens = trimmed.split(/\\s+/);\n\treturn tokens.length === 1 ? (tokens[0] ?? \"\") : null;\n}\n\n/**\n * Clamp to the nearest supported non-off level, searching upward first and then downward — the\n * same direction policy the session clamp uses, so a remembered level below a model's floor lands\n * on that floor instead of silently disabling reasoning.\n */\nfunction clampToNonOff(level: ThinkingLevel, capability: ReasoningCapability): ThinkingLevel | undefined {\n\tconst supported = new Set(capability.nonOffLevels);\n\tif (supported.has(level)) return level;\n\tconst index = EFFORT_LEVELS.indexOf(level);\n\tif (index === -1) return capability.nonOffLevels[0];\n\tfor (let i = index; i < EFFORT_LEVELS.length; i++) {\n\t\tconst candidate = EFFORT_LEVELS[i];\n\t\tif (candidate && supported.has(candidate)) return candidate;\n\t}\n\tfor (let i = index - 1; i >= 0; i--) {\n\t\tconst candidate = EFFORT_LEVELS[i];\n\t\tif (candidate && supported.has(candidate)) return candidate;\n\t}\n\treturn capability.nonOffLevels[0];\n}\n\n/**\n * Level to restore when reasoning is switched back on: this model's durable last non-off level,\n * else a legacy non-off effective memory, else the global default when that is not \"off\", else\n * medium — always clamped to what the model supports.\n */\nfunction resolvePreferredOnLevel(\n\tctx: ExtensionCommandContext,\n\tmodel: Model<Api>,\n\tcapability: ReasoningCapability,\n): ThinkingLevel | undefined {\n\tconst settingsManager = SettingsManager.create(ctx.cwd, ctx.agentDir, { projectTrusted: ctx.isProjectTrusted() });\n\tconst lastOnLevel = settingsManager.getModelLastOnThinkingLevel(model.provider, model.id);\n\tconst remembered = settingsManager.getModelThinkingLevel(model.provider, model.id);\n\tconst globalDefault = settingsManager.getDefaultThinkingLevel();\n\tconst preferred = lastOnLevel ?? nonOff(remembered) ?? nonOff(globalDefault) ?? DEFAULT_ON_LEVEL;\n\treturn clampToNonOff(preferred, capability);\n}\n\nfunction createSettingsManager(ctx: ExtensionCommandContext): SettingsManager {\n\treturn SettingsManager.create(ctx.cwd, ctx.agentDir, { projectTrusted: ctx.isProjectTrusted() });\n}\n\nfunction nonOff(level: ThinkingLevel | undefined): ThinkingLevel | undefined {\n\treturn level !== undefined && level !== \"off\" ? level : undefined;\n}\n\nfunction toCompletions(values: readonly string[], prefix: string): AutocompleteItem[] | null {\n\tconst normalizedPrefix = prefix.trim();\n\tconst matches = values.filter((value) => value.startsWith(normalizedPrefix));\n\treturn matches.length > 0 ? matches.map((value) => ({ value, label: value })) : null;\n}\n\nexport default function reasoningExtension(pi: ExtensionAPI): void {\n\t// Completions run outside a command context, which carries no model, so the live model is\n\t// tracked from the events that change it. Handlers always read ctx.model instead.\n\tlet currentModel: Model<Api> | undefined;\n\n\tpi.on(\"session_start\", (_event, ctx) => {\n\t\tcurrentModel = ctx.model;\n\t});\n\tpi.on(\"model_select\", (event) => {\n\t\tcurrentModel = event.model;\n\t});\n\n\tpi.registerCommand(\"reasoning\", {\n\t\tdescription: \"Show or toggle reasoning for the current model\",\n\t\targumentHint: \"[on|off]\",\n\t\tgetArgumentCompletions: (prefix) => toCompletions(REASONING_ARGUMENTS, prefix),\n\t\thandler: async (args, ctx) => {\n\t\t\tconst model = ctx.model;\n\t\t\tif (!model) {\n\t\t\t\tctx.ui.notify(\"No model is active.\", \"error\");\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tcurrentModel = model;\n\n\t\t\tconst argument = parseSingleArgument(args);\n\t\t\tif (argument === null || (argument !== \"\" && !REASONING_ARGUMENTS.includes(argument as ReasoningArgument))) {\n\t\t\t\tctx.ui.notify(REASONING_USAGE, \"error\");\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tconst capability = classifyReasoningCapability(model);\n\t\t\tconst currentLevel = pi.getThinkingLevel();\n\n\t\t\tif (argument === \"\") {\n\t\t\t\tctx.ui.notify(reasoningStatus(currentLevel), \"info\");\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tif (argument === \"off\") {\n\t\t\t\tif (capability.kind === \"none\") {\n\t\t\t\t\t// Nothing to disable; the model is already, permanently, off.\n\t\t\t\t\tctx.ui.notify(reasoningStatus(\"off\"), \"info\");\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tif (capability.kind === \"always-on\") {\n\t\t\t\t\tctx.ui.notify(`Reasoning cannot be disabled for ${modelKey(model)}.`, \"error\");\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tif (currentLevel !== \"off\") {\n\t\t\t\t\tconst settingsManager = createSettingsManager(ctx);\n\t\t\t\t\tsettingsManager.setModelLastOnThinkingLevel(model.provider, model.id, currentLevel);\n\t\t\t\t\tawait settingsManager.flush();\n\t\t\t\t}\n\t\t\t\tpi.setThinkingLevel(\"off\");\n\t\t\t\tctx.ui.notify(reasoningStatus(pi.getThinkingLevel()), \"info\");\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tif (capability.kind === \"none\") {\n\t\t\t\tctx.ui.notify(noReasoningMessage(model), \"error\");\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// Already reasoning: re-persist the current level instead of overriding the user's choice.\n\t\t\tif (currentLevel !== \"off\") {\n\t\t\t\tpi.setThinkingLevel(currentLevel);\n\t\t\t\tctx.ui.notify(reasoningStatus(pi.getThinkingLevel()), \"info\");\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tconst target = resolvePreferredOnLevel(ctx, model, capability);\n\t\t\tif (target === undefined) {\n\t\t\t\tctx.ui.notify(noReasoningMessage(model), \"error\");\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// Apply the clamped level to the session, but persist it without rewriting a stale raw\n\t\t\t// last-on preference. If catalog support returns later, the original preference still wins.\n\t\t\tpi.setSessionThinkingLevel(target);\n\t\t\tconst settingsManager = createSettingsManager(ctx);\n\t\t\tsettingsManager.setModelThinkingLevel(model.provider, model.id, target, { preserveLastOn: true });\n\t\t\tsettingsManager.setDefaultThinkingLevel(target);\n\t\t\tawait settingsManager.flush();\n\t\t\tctx.ui.notify(reasoningStatus(pi.getThinkingLevel()), \"info\");\n\t\t},\n\t});\n\n\tpi.registerCommand(\"efforts\", {\n\t\tdescription: \"Show or set the reasoning effort for the current model\",\n\t\targumentHint: \"[minimal|low|medium|high|xhigh|max]\",\n\t\tgetArgumentCompletions: (prefix) => {\n\t\t\t// Only a graded model has an effort ladder worth completing.\n\t\t\tconst capability = currentModel ? classifyReasoningCapability(currentModel) : undefined;\n\t\t\tif (capability?.kind !== \"graded\") return null;\n\t\t\treturn toCompletions(capability.nonOffLevels, prefix);\n\t\t},\n\t\thandler: async (args, ctx) => {\n\t\t\tconst model = ctx.model;\n\t\t\tif (!model) {\n\t\t\t\tctx.ui.notify(\"No model is active.\", \"error\");\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tcurrentModel = model;\n\n\t\t\tconst argument = parseSingleArgument(args);\n\t\t\tif (argument === null || (argument !== \"\" && !EFFORT_TOKENS.includes(argument))) {\n\t\t\t\tctx.ui.notify(EFFORTS_USAGE, \"error\");\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tconst capability = classifyReasoningCapability(model);\n\t\t\tif (capability.kind === \"none\") {\n\t\t\t\tctx.ui.notify(noReasoningMessage(model), \"error\");\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (capability.kind === \"on-off\") {\n\t\t\t\tctx.ui.notify(onOffOnlyMessage(model), \"error\");\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tif (argument === \"\") {\n\t\t\t\tctx.ui.notify(effortsStatus(pi.getThinkingLevel(), capability), \"info\");\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tconst level = argument as ThinkingLevel;\n\t\t\tif (!capability.nonOffLevels.includes(level)) {\n\t\t\t\tctx.ui.notify(\n\t\t\t\t\t`Reasoning effort \"${level}\" is not supported by ${modelKey(model)}. Available: ${capability.nonOffLevels.join(\", \")}.`,\n\t\t\t\t\t\"error\",\n\t\t\t\t);\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tpi.setThinkingLevel(level);\n\t\t\tctx.ui.notify(effortsStatus(pi.getThinkingLevel(), capability), \"info\");\n\t\t},\n\t});\n}\n"]}

package/dist/core/extensions/builtin/service-tier.d.ts

@@ -1,6 +1,49 @@
-import type { Api } from "@earendil-works/pi-ai";
+import type { Api, Model } from "@earendil-works/pi-ai";
+import type { ModelRegistry } from "../../model-registry.ts";
+import { SettingsManager } from "../../settings-manager.ts";
 import type { ExtensionAPI, ServiceTier } from "../types.ts";
 export type { ServiceTier };
+export declare const CODEX_RESPONSES_API = "openai-codex-responses";
 export declare function addServiceTierToPayload(api: Api | undefined, payload: unknown, serviceTier?: ServiceTier): unknown;
+/**
+ * The model that owns this model's persisted service-tier preference.
+ *
+ * A `-fast` catalog variant and its base model are one model to the user (`/fast` just swaps
+ * between them), so both must read and write ONE key. Anything else can hold contradictory
+ * preferences for the same choice.
+ */
+export declare function resolveServiceTierMemoryModel(modelRegistry: ModelRegistry, model: Model<Api>): Model<Api>;
+/** Remembered service tier for this model, normalized onto its base-model key. */
+export declare function getRememberedServiceTier(settingsManager: SettingsManager, modelRegistry: ModelRegistry, model: Model<Api>): ServiceTier | undefined;
+/** Host capabilities `applyFastMode` needs; satisfied by an extension context plus `pi`. */
+export interface FastModeContext {
+    cwd: string;
+    agentDir: string;
+    model: Model<Api> | undefined;
+    modelRegistry: ModelRegistry;
+    /** Tier the host resolved for the active model (scoped/favorite pin or catalog). */
+    serviceTier: ServiceTier | undefined;
+    isProjectTrusted(): boolean;
+    notify(message: string, level: "info" | "warning" | "error"): void;
+    setSessionModel(model: Model<Api>): Promise<boolean>;
+    setSessionFastMode(enabled: boolean): void;
+}
+export interface FastModeResult {
+    /** Fast mode state after the call; unchanged when the request was refused. */
+    enabled: boolean;
+    /** False when the request was refused (non-Codex model, active pin, failed model switch). */
+    applied: boolean;
+    /** The just-persisted tier (or the unchanged one when refused); `"priority"` on, `"auto"` off. */
+    recordedTier: ServiceTier;
+    message: string;
+}
+/**
+ * Single entry point for turning fast mode on/off: the `/fast` command and any other surface
+ * (RPC) must go through this so persistence and `-fast` key normalization exist exactly once.
+ *
+ * `off` persists an explicit `"auto"` rather than deleting the key: deletion re-inherits the
+ * catalog/global tier, which is what the user just asked to turn off.
+ */
+export declare function applyFastMode(ctx: FastModeContext, enabled: boolean): Promise<FastModeResult>;
 export default function serviceTierExtension(pi: ExtensionAPI): void;
 //# sourceMappingURL=service-tier.d.ts.map

package/dist/core/extensions/builtin/service-tier.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"service-tier.d.ts","sourceRoot":"","sources":["../../../../src/core/extensions/builtin/service-tier.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,GAAG,EAAS,MAAM,uBAAuB,CAAC;AAGxD,OAAO,KAAK,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAE7D,YAAY,EAAE,WAAW,EAAE,CAAC;AAa5B,wBAAgB,uBAAuB,CAAC,GAAG,EAAE,GAAG,GAAG,SAAS,EAAE,OAAO,EAAE,OAAO,EAAE,WAAW,CAAC,EAAE,WAAW,GAAG,OAAO,CAalH;AAkCD,MAAM,CAAC,OAAO,UAAU,oBAAoB,CAAC,EAAE,EAAE,YAAY,GAAG,IAAI,CAqEnE"}
+{"version":3,"file":"service-tier.d.ts","sourceRoot":"","sources":["../../../../src/core/extensions/builtin/service-tier.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,uBAAuB,CAAC;AAExD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,yBAAyB,CAAC;AAC7D,OAAO,EAAE,eAAe,EAAE,MAAM,2BAA2B,CAAC;AAC5D,OAAO,KAAK,EAAE,YAAY,EAA2B,WAAW,EAAE,MAAM,aAAa,CAAC;AAEtF,YAAY,EAAE,WAAW,EAAE,CAAC;AAK5B,eAAO,MAAM,mBAAmB,2BAA6B,CAAC;AAa9D,wBAAgB,uBAAuB,CAAC,GAAG,EAAE,GAAG,GAAG,SAAS,EAAE,OAAO,EAAE,OAAO,EAAE,WAAW,CAAC,EAAE,WAAW,GAAG,OAAO,CAalH;AAkCD;;;;;;GAMG;AACH,wBAAgB,6BAA6B,CAAC,aAAa,EAAE,aAAa,EAAE,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,GAAG,KAAK,CAAC,GAAG,CAAC,CAEzG;AAED,kFAAkF;AAClF,wBAAgB,wBAAwB,CACvC,eAAe,EAAE,eAAe,EAChC,aAAa,EAAE,aAAa,EAC5B,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,GACf,WAAW,GAAG,SAAS,CAGzB;AAED,4FAA4F;AAC5F,MAAM,WAAW,eAAe;IAC/B,GAAG,EAAE,MAAM,CAAC;IACZ,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,GAAG,SAAS,CAAC;IAC9B,aAAa,EAAE,aAAa,CAAC;IAC7B,oFAAoF;IACpF,WAAW,EAAE,WAAW,GAAG,SAAS,CAAC;IACrC,gBAAgB,IAAI,OAAO,CAAC;IAC5B,MAAM,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,SAAS,GAAG,OAAO,GAAG,IAAI,CAAC;IACnE,eAAe,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;IACrD,kBAAkB,CAAC,OAAO,EAAE,OAAO,GAAG,IAAI,CAAC;CAC3C;AAED,MAAM,WAAW,cAAc;IAC9B,8EAA8E;IAC9E,OAAO,EAAE,OAAO,CAAC;IACjB,6FAA6F;IAC7F,OAAO,EAAE,OAAO,CAAC;IACjB,kGAAkG;IAClG,YAAY,EAAE,WAAW,CAAC;IAC1B,OAAO,EAAE,MAAM,CAAC;CAChB;AAeD;;;;;;GAMG;AACH,wBAAsB,aAAa,CAAC,GAAG,EAAE,eAAe,EAAE,OAAO,EAAE,OAAO,GAAG,OAAO,CAAC,cAAc,CAAC,CAsCnG;AAqBD,MAAM,CAAC,OAAO,UAAU,oBAAoB,CAAC,EAAE,EAAE,YAAY,GAAG,IAAI,CAmInE"}

package/dist/core/extensions/builtin/service-tier.js

@@ -1,8 +1,11 @@
 import { SettingsManager } from "../../settings-manager.js";
 const OPENAI_CODEX_RESPONSES_API = "openai-codex-responses";
+export const CODEX_RESPONSES_API = OPENAI_CODEX_RESPONSES_API;
 const FAST_MODEL_SUFFIX = "-fast";
 const PRIORITY_TIER = "priority";
 const SERVICE_TIER_APIS = new Set(["openai-responses", OPENAI_CODEX_RESPONSES_API]);
+const FAST_ARGUMENTS = ["on", "off"];
+const FAST_USAGE = "Usage: /fast [on|off]";
 function isRecord(value) {
     return typeof value === "object" && value !== null;
 }
@@ -42,6 +45,93 @@ function findFastModel(modelRegistry, baseModel) {
     const fastModel = modelRegistry.find(baseModel.provider, `${baseModel.id}${FAST_MODEL_SUFFIX}`);
     return fastModel && isCompatibleFastVariant(modelRegistry, baseModel, fastModel) ? fastModel : undefined;
 }
+/**
+ * The model that owns this model's persisted service-tier preference.
+ *
+ * A `-fast` catalog variant and its base model are one model to the user (`/fast` just swaps
+ * between them), so both must read and write ONE key. Anything else can hold contradictory
+ * preferences for the same choice.
+ */
+export function resolveServiceTierMemoryModel(modelRegistry, model) {
+    return findBaseModel(modelRegistry, model) ?? model;
+}
+/** Remembered service tier for this model, normalized onto its base-model key. */
+export function getRememberedServiceTier(settingsManager, modelRegistry, model) {
+    const memoryModel = resolveServiceTierMemoryModel(modelRegistry, model);
+    return settingsManager.getModelServiceTier(memoryModel.provider, memoryModel.id);
+}
+/**
+ * Whether the active model's priority tier is PINNED by the user's model selection (a scoped
+ * or favorite `provider/id:priority` pattern) rather than inherited from the catalog.
+ *
+ * `ctx.serviceTier` merges both sources, so a priority the catalog does not explain must have
+ * come from a `:priority` decorator; checking `scopedModels` alone would miss favorite pins,
+ * which the session resolves the same way.
+ */
+function hasPriorityPin(ctx, model) {
+    if (ctx.serviceTier !== PRIORITY_TIER)
+        return false;
+    return ctx.modelRegistry.getServiceTier(model) !== PRIORITY_TIER;
+}
+/**
+ * Single entry point for turning fast mode on/off: the `/fast` command and any other surface
+ * (RPC) must go through this so persistence and `-fast` key normalization exist exactly once.
+ *
+ * `off` persists an explicit `"auto"` rather than deleting the key: deletion re-inherits the
+ * catalog/global tier, which is what the user just asked to turn off.
+ */
+export async function applyFastMode(ctx, enabled) {
+    const model = ctx.model;
+    if (model?.api !== OPENAI_CODEX_RESPONSES_API) {
+        const message = "Fast mode is only available for OpenAI Codex models.";
+        ctx.notify(message, "warning");
+        return { enabled: false, applied: false, recordedTier: "auto", message };
+    }
+    if (!enabled && hasPriorityPin(ctx, model)) {
+        // The pin outranks the memory, so writing "auto" here would be a silent no-op on the wire.
+        const message = "Fast mode is fixed by the active model selection's priority tier.";
+        ctx.notify(message, "info");
+        // The pin keeps the model on priority no matter what the memory says, so that is the honest
+        // tier to report; nothing was written, and no caller reads this field when `applied` is false.
+        return { enabled: true, applied: false, recordedTier: PRIORITY_TIER, message };
+    }
+    const settingsManager = SettingsManager.create(ctx.cwd, ctx.agentDir, { projectTrusted: ctx.isProjectTrusted() });
+    const memoryModel = resolveServiceTierMemoryModel(ctx.modelRegistry, model);
+    const tier = enabled ? PRIORITY_TIER : "auto";
+    settingsManager.setModelServiceTier(memoryModel.provider, memoryModel.id, tier);
+    await settingsManager.flush();
+    const baseModel = findBaseModel(ctx.modelRegistry, model);
+    const targetModel = enabled ? findFastModel(ctx.modelRegistry, model) : baseModel;
+    if (targetModel) {
+        if (!(await ctx.setSessionModel(targetModel))) {
+            const message = `Could not switch to ${targetModel.provider}/${targetModel.id}.`;
+            ctx.notify(message, "error");
+            return { enabled: baseModel === undefined, applied: false, recordedTier: "auto", message };
+        }
+    }
+    ctx.setSessionFastMode(enabled);
+    const labelModel = targetModel ?? model;
+    const message = `Fast mode ${enabled ? "enabled" : "disabled"}: ${labelModel.id}`;
+    ctx.notify(message, "info");
+    return { enabled, applied: true, recordedTier: tier, message };
+}
+function toCompletions(values, prefix) {
+    const matches = values.filter((value) => value.startsWith(prefix.trim()));
+    return matches.length > 0 ? matches.map((value) => ({ value, label: value })) : null;
+}
+function toFastModeContext(pi, ctx) {
+    return {
+        cwd: ctx.cwd,
+        agentDir: ctx.agentDir,
+        model: ctx.model,
+        modelRegistry: ctx.modelRegistry,
+        serviceTier: ctx.serviceTier,
+        isProjectTrusted: () => ctx.isProjectTrusted(),
+        notify: (message, level) => ctx.ui.notify(message, level),
+        setSessionModel: (model) => pi.setSessionModel(model),
+        setSessionFastMode: (enabled) => pi.setSessionFastMode(enabled),
+    };
+}
 export default function serviceTierExtension(pi) {
     let settingsServiceTier;
     /**
@@ -58,48 +148,110 @@ export default function serviceTierExtension(pi) {
      * sent — so this toggle drives the request field directly. See issue #545.
      */
     let sessionFastMode = false;
+    /**
+     * The per-model memory tier `/fast` last wrote or read for the active base model this
+     * session. Lives only to suppress an inherited priority when the user turns fast off in
+     * the SAME session: the host's resolved `ctx.serviceTier` is recomputed on model switch,
+     * not on a memory write, so without this an in-session `/fast off` (no model swap) would
+     * keep sending `priority` until a restart. Scoped to the active base key (re-derived on
+     * model_select for the incoming model) so it never leaks across models.
+     */
+    let liveMemoryTier;
+    let liveMemoryKey;
     pi.on("session_start", async (_event, ctx) => {
-        const settingsManager = SettingsManager.create(ctx.cwd);
+        const settingsManager = SettingsManager.create(ctx.cwd, ctx.agentDir, { projectTrusted: ctx.isProjectTrusted() });
         settingsServiceTier = settingsManager.getOpenAIServiceTier();
-        sessionFastMode = false;
-        pi.setSessionFastMode(false);
         const model = ctx.model;
         if (model?.api !== OPENAI_CODEX_RESPONSES_API) {
+            sessionFastMode = false;
+            pi.setSessionFastMode(false);
             return;
         }
+        // Fast mode is remembered per model, so a new session inherits the user's last choice for
+        // THIS model instead of silently starting off. A model whose catalog/scoped tier is already
+        // `priority` (a models.json priority entry, or a `-fast`-less codex model) is treated as
+        // fast-active when nothing is remembered; an explicit remembered `"auto"` (`/fast off`)
+        // wins over that inheritance. Malformed values read back as undefined.
+        //
+        // The flag is derived from the POST-swap model: a `-fast` catalog variant swaps down to
+        // its base before this reads `ctx.serviceTier`, so inheritance is judged on the model the
+        // user actually ends up on.
+        const remembered = getRememberedServiceTier(settingsManager, ctx.modelRegistry, model);
         const baseModel = findBaseModel(ctx.modelRegistry, model);
         if (baseModel) {
             await pi.setSessionModel(baseModel);
         }
+        sessionFastMode = remembered === PRIORITY_TIER || (remembered === undefined && ctx.serviceTier === PRIORITY_TIER);
+        pi.setSessionFastMode(sessionFastMode);
+        const memoryModel = resolveServiceTierMemoryModel(ctx.modelRegistry, model);
+        liveMemoryKey = `${memoryModel.provider}/${memoryModel.id}`;
+        liveMemoryTier = remembered;
+    });
+    pi.on("model_select", (event, ctx) => {
+        // A model switch changes the base key the memory lives under, so the live tier is RE-DERIVED
+        // for the incoming model instead of merely dropped: dropping it would leave a remembered
+        // "auto" unable to suppress a catalog-inherited priority after switching away and back in one
+        // session, silently re-sending the tier `/fast off` turned off. Per-key scoping is preserved —
+        // each model reads ITS OWN memory, never the previous model's.
+        const settingsManager = SettingsManager.create(ctx.cwd, ctx.agentDir, { projectTrusted: ctx.isProjectTrusted() });
+        const memoryModel = resolveServiceTierMemoryModel(ctx.modelRegistry, event.model);
+        liveMemoryKey = `${memoryModel.provider}/${memoryModel.id}`;
+        liveMemoryTier = getRememberedServiceTier(settingsManager, ctx.modelRegistry, event.model);
     });
     pi.registerCommand("fast", {
-        description: "Toggle OpenAI Codex fast mode for this session",
-        handler: async (_args, ctx) => {
-            const model = ctx.model;
-            if (model?.api !== OPENAI_CODEX_RESPONSES_API) {
-                ctx.ui.notify("Fast mode is only available for OpenAI Codex models.", "warning");
+        description: "Turn OpenAI Codex fast mode on or off for the current model",
+        argumentHint: "[on|off]",
+        getArgumentCompletions: (prefix) => toCompletions(FAST_ARGUMENTS, prefix),
+        handler: async (args, ctx) => {
+            const argument = args.trim().toLowerCase();
+            if (argument !== "" && !FAST_ARGUMENTS.includes(argument)) {
+                ctx.ui.notify(FAST_USAGE, "error");
                 return;
             }
-            const baseModel = findBaseModel(ctx.modelRegistry, model);
-            const targetModel = baseModel ?? findFastModel(ctx.modelRegistry, model);
-            if (!targetModel) {
-                sessionFastMode = !sessionFastMode;
-                pi.setSessionFastMode(sessionFastMode);
-                ctx.ui.notify(`Fast mode ${sessionFastMode ? "enabled" : "disabled"}: ${model.id}`, "info");
-                return;
+            // No argument keeps the established toggle UX.
+            const active = sessionFastMode || ctx.serviceTier === PRIORITY_TIER;
+            const enabled = argument === "" ? !active : argument === "on";
+            const result = await applyFastMode(toFastModeContext(pi, ctx), enabled);
+            sessionFastMode = result.enabled;
+            // Track this session's own write AFTER applyFastMode's internal model switch (which fires
+            // model_select and clears the live tier). Keyed to the active base model so a later switch
+            // cannot carry it across models.
+            if (result.applied && ctx.model) {
+                const memoryModel = resolveServiceTierMemoryModel(ctx.modelRegistry, ctx.model);
+                liveMemoryKey = `${memoryModel.provider}/${memoryModel.id}`;
+                liveMemoryTier = result.recordedTier;
             }
-            if (!(await pi.setSessionModel(targetModel))) {
-                ctx.ui.notify(`Could not switch to ${targetModel.provider}/${targetModel.id}.`, "error");
-                return;
-            }
-            const enabled = baseModel === undefined;
-            ctx.ui.notify(`Fast mode ${enabled ? "enabled" : "disabled"}: ${targetModel.id}`, "info");
         },
     });
     pi.on("before_provider_request", (event, ctx) => {
-        const effectiveServiceTier = ctx.model?.api === OPENAI_CODEX_RESPONSES_API
-            ? (ctx.serviceTier ?? (sessionFastMode ? PRIORITY_TIER : undefined))
-            : (ctx.serviceTier ?? settingsServiceTier);
+        let effectiveServiceTier;
+        if (ctx.model?.api === OPENAI_CODEX_RESPONSES_API) {
+            if (sessionFastMode) {
+                effectiveServiceTier = PRIORITY_TIER;
+            }
+            else {
+                const memoryModel = ctx.model ? resolveServiceTierMemoryModel(ctx.modelRegistry, ctx.model) : undefined;
+                const activeBaseKey = memoryModel ? `${memoryModel.provider}/${memoryModel.id}` : undefined;
+                // Same-session /fast off: the host's resolved ctx.serviceTier is only recomputed on a
+                // model switch, so an inherited priority would otherwise keep emitting until a restart.
+                // Only a CATALOG-explained priority may be suppressed: a priority the catalog does not
+                // explain is an explicit scoped/favorite `:priority` pin, which outranks the memory
+                // (same discriminator `applyFastMode` uses to refuse `/fast off` under a pin).
+                const catalogExplainsPriority = ctx.model !== undefined && ctx.modelRegistry.getServiceTier(ctx.model) === PRIORITY_TIER;
+                if (liveMemoryTier === "auto" &&
+                    activeBaseKey === liveMemoryKey &&
+                    ctx.serviceTier === PRIORITY_TIER &&
+                    catalogExplainsPriority) {
+                    effectiveServiceTier = undefined;
+                }
+                else {
+                    effectiveServiceTier = ctx.serviceTier;
+                }
+            }
+        }
+        else {
+            effectiveServiceTier = ctx.serviceTier ?? settingsServiceTier;
+        }
         return addServiceTierToPayload(ctx.model?.api, event.payload, effectiveServiceTier);
     });
 }

package/dist/core/extensions/builtin/service-tier.js.map

@@ -1 +1 @@
-{"version":3,"file":"service-tier.js","sourceRoot":"","sources":["../../../../src/core/extensions/builtin/service-tier.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,eAAe,EAAE,MAAM,2BAA2B,CAAC;AAO5D,MAAM,0BAA0B,GAAG,wBAAwB,CAAC;AAC5D,MAAM,iBAAiB,GAAG,OAAO,CAAC;AAClC,MAAM,aAAa,GAAgB,UAAU,CAAC;AAC9C,MAAM,iBAAiB,GAAqB,IAAI,GAAG,CAAC,CAAC,kBAAkB,EAAE,0BAA0B,CAAC,CAAC,CAAC;AAEtG,SAAS,QAAQ,CAAC,KAAc;IAC/B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,CAAC;AACpD,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,GAAoB,EAAE,OAAgB,EAAE,WAAyB;IACxG,IAAI,CAAC,GAAG,IAAI,CAAC,iBAAiB,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC;QACzD,OAAO,OAAO,CAAC;IAChB,CAAC;IAED,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,OAAO,CAAC,YAAY,KAAK,SAAS,EAAE,CAAC;QAC9D,OAAO,OAAO,CAAC;IAChB,CAAC;IAED,OAAO;QACN,GAAG,OAAO;QACV,YAAY,EAAE,WAAW;KACzB,CAAC;AACH,CAAC;AAED,SAAS,iBAAiB,CAAC,aAA4B,EAAE,KAAiB;IACzE,OAAO,aAAa,CAAC,kBAAkB,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC;AAC5D,CAAC;AAED,SAAS,uBAAuB,CAAC,aAA4B,EAAE,SAAqB,EAAE,SAAqB;IAC1G,OAAO,CACN,SAAS,CAAC,QAAQ,KAAK,SAAS,CAAC,QAAQ;QACzC,SAAS,CAAC,GAAG,KAAK,SAAS,CAAC,GAAG;QAC/B,aAAa,CAAC,cAAc,CAAC,SAAS,CAAC,KAAK,UAAU;QACtD,iBAAiB,CAAC,aAAa,EAAE,SAAS,CAAC,KAAK,iBAAiB,CAAC,aAAa,EAAE,SAAS,CAAC,CAC3F,CAAC;AACH,CAAC;AAED,SAAS,aAAa,CAAC,aAA4B,EAAE,SAAqB;IACzE,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC,EAAE,CAAC;QAC/C,OAAO,SAAS,CAAC;IAClB,CAAC;IAED,MAAM,WAAW,GAAG,SAAS,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,iBAAiB,CAAC,MAAM,CAAC,CAAC;IACrE,MAAM,SAAS,GAAG,aAAa,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,EAAE,WAAW,CAAC,CAAC;IACtE,OAAO,SAAS,IAAI,uBAAuB,CAAC,aAAa,EAAE,SAAS,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC;AAC1G,CAAC;AAED,SAAS,aAAa,CAAC,aAA4B,EAAE,SAAqB;IACzE,IAAI,SAAS,CAAC,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC,EAAE,CAAC;QAC9C,OAAO,SAAS,CAAC;IAClB,CAAC;IAED,MAAM,SAAS,GAAG,aAAa,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,EAAE,GAAG,SAAS,CAAC,EAAE,GAAG,iBAAiB,EAAE,CAAC,CAAC;IAChG,OAAO,SAAS,IAAI,uBAAuB,CAAC,aAAa,EAAE,SAAS,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC;AAC1G,CAAC;AAED,MAAM,CAAC,OAAO,UAAU,oBAAoB,CAAC,EAAgB;IAC5D,IAAI,mBAA4C,CAAC;IACjD;;;;;;;;;;;;OAYG;IACH,IAAI,eAAe,GAAG,KAAK,CAAC;IAE5B,EAAE,CAAC,EAAE,CAAC,eAAe,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,EAAE;QAC5C,MAAM,eAAe,GAAG,eAAe,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACxD,mBAAmB,GAAG,eAAe,CAAC,oBAAoB,EAAE,CAAC;QAC7D,eAAe,GAAG,KAAK,CAAC;QACxB,EAAE,CAAC,kBAAkB,CAAC,KAAK,CAAC,CAAC;QAE7B,MAAM,KAAK,GAAG,GAAG,CAAC,KAAK,CAAC;QACxB,IAAI,KAAK,EAAE,GAAG,KAAK,0BAA0B,EAAE,CAAC;YAC/C,OAAO;QACR,CAAC;QAED,MAAM,SAAS,GAAG,aAAa,CAAC,GAAG,CAAC,aAAa,EAAE,KAAK,CAAC,CAAC;QAC1D,IAAI,SAAS,EAAE,CAAC;YACf,MAAM,EAAE,CAAC,eAAe,CAAC,SAAS,CAAC,CAAC;QACrC,CAAC;IACF,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,eAAe,CAAC,MAAM,EAAE;QAC1B,WAAW,EAAE,gDAAgD;QAC7D,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,GAAG,EAAE,EAAE;YAC7B,MAAM,KAAK,GAAG,GAAG,CAAC,KAAK,CAAC;YACxB,IAAI,KAAK,EAAE,GAAG,KAAK,0BAA0B,EAAE,CAAC;gBAC/C,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,sDAAsD,EAAE,SAAS,CAAC,CAAC;gBACjF,OAAO;YACR,CAAC;YAED,MAAM,SAAS,GAAG,aAAa,CAAC,GAAG,CAAC,aAAa,EAAE,KAAK,CAAC,CAAC;YAC1D,MAAM,WAAW,GAAG,SAAS,IAAI,aAAa,CAAC,GAAG,CAAC,aAAa,EAAE,KAAK,CAAC,CAAC;YACzE,IAAI,CAAC,WAAW,EAAE,CAAC;gBAClB,eAAe,GAAG,CAAC,eAAe,CAAC;gBACnC,EAAE,CAAC,kBAAkB,CAAC,eAAe,CAAC,CAAC;gBACvC,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,aAAa,eAAe,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,KAAK,KAAK,CAAC,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC;gBAC5F,OAAO;YACR,CAAC;YAED,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,eAAe,CAAC,WAAW,CAAC,CAAC,EAAE,CAAC;gBAC9C,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,uBAAuB,WAAW,CAAC,QAAQ,IAAI,WAAW,CAAC,EAAE,GAAG,EAAE,OAAO,CAAC,CAAC;gBACzF,OAAO;YACR,CAAC;YAED,MAAM,OAAO,GAAG,SAAS,KAAK,SAAS,CAAC;YACxC,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,aAAa,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,KAAK,WAAW,CAAC,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC;QAC3F,CAAC;KACD,CAAC,CAAC;IAEH,EAAE,CAAC,EAAE,CAAC,yBAAyB,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE,EAAE;QAC/C,MAAM,oBAAoB,GACzB,GAAG,CAAC,KAAK,EAAE,GAAG,KAAK,0BAA0B;YAC5C,CAAC,CAAC,CAAC,GAAG,CAAC,WAAW,IAAI,CAAC,eAAe,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;YACpE,CAAC,CAAC,CAAC,GAAG,CAAC,WAAW,IAAI,mBAAmB,CAAC,CAAC;QAC7C,OAAO,uBAAuB,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,EAAE,KAAK,CAAC,OAAO,EAAE,oBAAoB,CAAC,CAAC;IACrF,CAAC,CAAC,CAAC;AACJ,CAAC","sourcesContent":["import type { Api, Model } from \"@earendil-works/pi-ai\";\nimport type { ModelRegistry } from \"../../model-registry.ts\";\nimport { SettingsManager } from \"../../settings-manager.ts\";\nimport type { ExtensionAPI, ServiceTier } from \"../types.ts\";\n\nexport type { ServiceTier };\n\ntype ProviderPayload = Record<string, unknown>;\n\nconst OPENAI_CODEX_RESPONSES_API = \"openai-codex-responses\";\nconst FAST_MODEL_SUFFIX = \"-fast\";\nconst PRIORITY_TIER: ServiceTier = \"priority\";\nconst SERVICE_TIER_APIS: ReadonlySet<Api> = new Set([\"openai-responses\", OPENAI_CODEX_RESPONSES_API]);\n\nfunction isRecord(value: unknown): value is ProviderPayload {\n\treturn typeof value === \"object\" && value !== null;\n}\n\nexport function addServiceTierToPayload(api: Api | undefined, payload: unknown, serviceTier?: ServiceTier): unknown {\n\tif (!api || !SERVICE_TIER_APIS.has(api) || !serviceTier) {\n\t\treturn payload;\n\t}\n\n\tif (!isRecord(payload) || payload.service_tier !== undefined) {\n\t\treturn payload;\n\t}\n\n\treturn {\n\t\t...payload,\n\t\tservice_tier: serviceTier,\n\t};\n}\n\nfunction getRequestModelId(modelRegistry: ModelRegistry, model: Model<Api>): string {\n\treturn modelRegistry.getUpstreamModelId(model) ?? model.id;\n}\n\nfunction isCompatibleFastVariant(modelRegistry: ModelRegistry, baseModel: Model<Api>, fastModel: Model<Api>): boolean {\n\treturn (\n\t\tfastModel.provider === baseModel.provider &&\n\t\tfastModel.api === baseModel.api &&\n\t\tmodelRegistry.getServiceTier(fastModel) === \"priority\" &&\n\t\tgetRequestModelId(modelRegistry, fastModel) === getRequestModelId(modelRegistry, baseModel)\n\t);\n}\n\nfunction findBaseModel(modelRegistry: ModelRegistry, fastModel: Model<Api>): Model<Api> | undefined {\n\tif (!fastModel.id.endsWith(FAST_MODEL_SUFFIX)) {\n\t\treturn undefined;\n\t}\n\n\tconst baseModelId = fastModel.id.slice(0, -FAST_MODEL_SUFFIX.length);\n\tconst baseModel = modelRegistry.find(fastModel.provider, baseModelId);\n\treturn baseModel && isCompatibleFastVariant(modelRegistry, baseModel, fastModel) ? baseModel : undefined;\n}\n\nfunction findFastModel(modelRegistry: ModelRegistry, baseModel: Model<Api>): Model<Api> | undefined {\n\tif (baseModel.id.endsWith(FAST_MODEL_SUFFIX)) {\n\t\treturn undefined;\n\t}\n\n\tconst fastModel = modelRegistry.find(baseModel.provider, `${baseModel.id}${FAST_MODEL_SUFFIX}`);\n\treturn fastModel && isCompatibleFastVariant(modelRegistry, baseModel, fastModel) ? fastModel : undefined;\n}\n\nexport default function serviceTierExtension(pi: ExtensionAPI): void {\n\tlet settingsServiceTier: ServiceTier | undefined;\n\t/**\n\t * Session-level fast mode for Codex models that have no `-fast` catalog sibling.\n\t *\n\t * The catalog generator emits `-fast` priority variants only for the direct\n\t * `openai` provider, so `openai-codex` models never have a switch target. The\n\t * ChatGPT backend still offers the tier to subscriptions:\n\t * `chatgpt.com/backend-api/codex/models` advertises a `priority` service tier\n\t * labelled \"Fast\" (\"1.5x speed, increased usage\")\n\t * for gpt-5.4/5.5/5.6-*, and the first-party Codex CLI sends\n\t * `service_tier: \"priority\"` on subscription OAuth. The response echo is not a\n\t * confirmation channel — it reads `auto`/`default` whether or not a tier was\n\t * sent — so this toggle drives the request field directly. See issue #545.\n\t */\n\tlet sessionFastMode = false;\n\n\tpi.on(\"session_start\", async (_event, ctx) => {\n\t\tconst settingsManager = SettingsManager.create(ctx.cwd);\n\t\tsettingsServiceTier = settingsManager.getOpenAIServiceTier();\n\t\tsessionFastMode = false;\n\t\tpi.setSessionFastMode(false);\n\n\t\tconst model = ctx.model;\n\t\tif (model?.api !== OPENAI_CODEX_RESPONSES_API) {\n\t\t\treturn;\n\t\t}\n\n\t\tconst baseModel = findBaseModel(ctx.modelRegistry, model);\n\t\tif (baseModel) {\n\t\t\tawait pi.setSessionModel(baseModel);\n\t\t}\n\t});\n\n\tpi.registerCommand(\"fast\", {\n\t\tdescription: \"Toggle OpenAI Codex fast mode for this session\",\n\t\thandler: async (_args, ctx) => {\n\t\t\tconst model = ctx.model;\n\t\t\tif (model?.api !== OPENAI_CODEX_RESPONSES_API) {\n\t\t\t\tctx.ui.notify(\"Fast mode is only available for OpenAI Codex models.\", \"warning\");\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tconst baseModel = findBaseModel(ctx.modelRegistry, model);\n\t\t\tconst targetModel = baseModel ?? findFastModel(ctx.modelRegistry, model);\n\t\t\tif (!targetModel) {\n\t\t\t\tsessionFastMode = !sessionFastMode;\n\t\t\t\tpi.setSessionFastMode(sessionFastMode);\n\t\t\t\tctx.ui.notify(`Fast mode ${sessionFastMode ? \"enabled\" : \"disabled\"}: ${model.id}`, \"info\");\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tif (!(await pi.setSessionModel(targetModel))) {\n\t\t\t\tctx.ui.notify(`Could not switch to ${targetModel.provider}/${targetModel.id}.`, \"error\");\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tconst enabled = baseModel === undefined;\n\t\t\tctx.ui.notify(`Fast mode ${enabled ? \"enabled\" : \"disabled\"}: ${targetModel.id}`, \"info\");\n\t\t},\n\t});\n\n\tpi.on(\"before_provider_request\", (event, ctx) => {\n\t\tconst effectiveServiceTier =\n\t\t\tctx.model?.api === OPENAI_CODEX_RESPONSES_API\n\t\t\t\t? (ctx.serviceTier ?? (sessionFastMode ? PRIORITY_TIER : undefined))\n\t\t\t\t: (ctx.serviceTier ?? settingsServiceTier);\n\t\treturn addServiceTierToPayload(ctx.model?.api, event.payload, effectiveServiceTier);\n\t});\n}\n"]}
+{"version":3,"file":"service-tier.js","sourceRoot":"","sources":["../../../../src/core/extensions/builtin/service-tier.ts"],"names":[],"mappings":"AAGA,OAAO,EAAE,eAAe,EAAE,MAAM,2BAA2B,CAAC;AAO5D,MAAM,0BAA0B,GAAG,wBAAwB,CAAC;AAC5D,MAAM,CAAC,MAAM,mBAAmB,GAAG,0BAA0B,CAAC;AAC9D,MAAM,iBAAiB,GAAG,OAAO,CAAC;AAClC,MAAM,aAAa,GAAgB,UAAU,CAAC;AAC9C,MAAM,iBAAiB,GAAqB,IAAI,GAAG,CAAC,CAAC,kBAAkB,EAAE,0BAA0B,CAAC,CAAC,CAAC;AACtG,MAAM,cAAc,GAAG,CAAC,IAAI,EAAE,KAAK,CAAU,CAAC;AAC9C,MAAM,UAAU,GAAG,uBAAuB,CAAC;AAI3C,SAAS,QAAQ,CAAC,KAAc;IAC/B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,CAAC;AACpD,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,GAAoB,EAAE,OAAgB,EAAE,WAAyB;IACxG,IAAI,CAAC,GAAG,IAAI,CAAC,iBAAiB,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC;QACzD,OAAO,OAAO,CAAC;IAChB,CAAC;IAED,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,OAAO,CAAC,YAAY,KAAK,SAAS,EAAE,CAAC;QAC9D,OAAO,OAAO,CAAC;IAChB,CAAC;IAED,OAAO;QACN,GAAG,OAAO;QACV,YAAY,EAAE,WAAW;KACzB,CAAC;AACH,CAAC;AAED,SAAS,iBAAiB,CAAC,aAA4B,EAAE,KAAiB;IACzE,OAAO,aAAa,CAAC,kBAAkB,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC;AAC5D,CAAC;AAED,SAAS,uBAAuB,CAAC,aAA4B,EAAE,SAAqB,EAAE,SAAqB;IAC1G,OAAO,CACN,SAAS,CAAC,QAAQ,KAAK,SAAS,CAAC,QAAQ;QACzC,SAAS,CAAC,GAAG,KAAK,SAAS,CAAC,GAAG;QAC/B,aAAa,CAAC,cAAc,CAAC,SAAS,CAAC,KAAK,UAAU;QACtD,iBAAiB,CAAC,aAAa,EAAE,SAAS,CAAC,KAAK,iBAAiB,CAAC,aAAa,EAAE,SAAS,CAAC,CAC3F,CAAC;AACH,CAAC;AAED,SAAS,aAAa,CAAC,aAA4B,EAAE,SAAqB;IACzE,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC,EAAE,CAAC;QAC/C,OAAO,SAAS,CAAC;IAClB,CAAC;IAED,MAAM,WAAW,GAAG,SAAS,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,iBAAiB,CAAC,MAAM,CAAC,CAAC;IACrE,MAAM,SAAS,GAAG,aAAa,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,EAAE,WAAW,CAAC,CAAC;IACtE,OAAO,SAAS,IAAI,uBAAuB,CAAC,aAAa,EAAE,SAAS,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC;AAC1G,CAAC;AAED,SAAS,aAAa,CAAC,aAA4B,EAAE,SAAqB;IACzE,IAAI,SAAS,CAAC,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC,EAAE,CAAC;QAC9C,OAAO,SAAS,CAAC;IAClB,CAAC;IAED,MAAM,SAAS,GAAG,aAAa,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,EAAE,GAAG,SAAS,CAAC,EAAE,GAAG,iBAAiB,EAAE,CAAC,CAAC;IAChG,OAAO,SAAS,IAAI,uBAAuB,CAAC,aAAa,EAAE,SAAS,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC;AAC1G,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,6BAA6B,CAAC,aAA4B,EAAE,KAAiB;IAC5F,OAAO,aAAa,CAAC,aAAa,EAAE,KAAK,CAAC,IAAI,KAAK,CAAC;AACrD,CAAC;AAED,kFAAkF;AAClF,MAAM,UAAU,wBAAwB,CACvC,eAAgC,EAChC,aAA4B,EAC5B,KAAiB;IAEjB,MAAM,WAAW,GAAG,6BAA6B,CAAC,aAAa,EAAE,KAAK,CAAC,CAAC;IACxE,OAAO,eAAe,CAAC,mBAAmB,CAAC,WAAW,CAAC,QAAQ,EAAE,WAAW,CAAC,EAAE,CAAC,CAAC;AAClF,CAAC;AA0BD;;;;;;;GAOG;AACH,SAAS,cAAc,CAAC,GAAoB,EAAE,KAAiB;IAC9D,IAAI,GAAG,CAAC,WAAW,KAAK,aAAa;QAAE,OAAO,KAAK,CAAC;IACpD,OAAO,GAAG,CAAC,aAAa,CAAC,cAAc,CAAC,KAAK,CAAC,KAAK,aAAa,CAAC;AAClE,CAAC;AAED;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,aAAa,CAAC,GAAoB,EAAE,OAAgB;IACzE,MAAM,KAAK,GAAG,GAAG,CAAC,KAAK,CAAC;IACxB,IAAI,KAAK,EAAE,GAAG,KAAK,0BAA0B,EAAE,CAAC;QAC/C,MAAM,OAAO,GAAG,sDAAsD,CAAC;QACvE,GAAG,CAAC,MAAM,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;QAC/B,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,YAAY,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;IAC1E,CAAC;IAED,IAAI,CAAC,OAAO,IAAI,cAAc,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,CAAC;QAC5C,2FAA2F;QAC3F,MAAM,OAAO,GAAG,mEAAmE,CAAC;QACpF,GAAG,CAAC,MAAM,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;QAC5B,4FAA4F;QAC5F,+FAA+F;QAC/F,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE,YAAY,EAAE,aAAa,EAAE,OAAO,EAAE,CAAC;IAChF,CAAC;IAED,MAAM,eAAe,GAAG,eAAe,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,GAAG,CAAC,QAAQ,EAAE,EAAE,cAAc,EAAE,GAAG,CAAC,gBAAgB,EAAE,EAAE,CAAC,CAAC;IAClH,MAAM,WAAW,GAAG,6BAA6B,CAAC,GAAG,CAAC,aAAa,EAAE,KAAK,CAAC,CAAC;IAC5E,MAAM,IAAI,GAAgB,OAAO,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,MAAM,CAAC;IAC3D,eAAe,CAAC,mBAAmB,CAAC,WAAW,CAAC,QAAQ,EAAE,WAAW,CAAC,EAAE,EAAE,IAAI,CAAC,CAAC;IAChF,MAAM,eAAe,CAAC,KAAK,EAAE,CAAC;IAE9B,MAAM,SAAS,GAAG,aAAa,CAAC,GAAG,CAAC,aAAa,EAAE,KAAK,CAAC,CAAC;IAC1D,MAAM,WAAW,GAAG,OAAO,CAAC,CAAC,CAAC,aAAa,CAAC,GAAG,CAAC,aAAa,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAClF,IAAI,WAAW,EAAE,CAAC;QACjB,IAAI,CAAC,CAAC,MAAM,GAAG,CAAC,eAAe,CAAC,WAAW,CAAC,CAAC,EAAE,CAAC;YAC/C,MAAM,OAAO,GAAG,uBAAuB,WAAW,CAAC,QAAQ,IAAI,WAAW,CAAC,EAAE,GAAG,CAAC;YACjF,GAAG,CAAC,MAAM,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YAC7B,OAAO,EAAE,OAAO,EAAE,SAAS,KAAK,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,YAAY,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;QAC5F,CAAC;IACF,CAAC;IAED,GAAG,CAAC,kBAAkB,CAAC,OAAO,CAAC,CAAC;IAChC,MAAM,UAAU,GAAG,WAAW,IAAI,KAAK,CAAC;IACxC,MAAM,OAAO,GAAG,aAAa,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,KAAK,UAAU,CAAC,EAAE,EAAE,CAAC;IAClF,GAAG,CAAC,MAAM,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;IAC5B,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC;AAChE,CAAC;AAED,SAAS,aAAa,CAAC,MAAyB,EAAE,MAAc;IAC/D,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,UAAU,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;IAC1E,OAAO,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;AACtF,CAAC;AAED,SAAS,iBAAiB,CAAC,EAAgB,EAAE,GAA4B;IACxE,OAAO;QACN,GAAG,EAAE,GAAG,CAAC,GAAG;QACZ,QAAQ,EAAE,GAAG,CAAC,QAAQ;QACtB,KAAK,EAAE,GAAG,CAAC,KAAK;QAChB,aAAa,EAAE,GAAG,CAAC,aAAa;QAChC,WAAW,EAAE,GAAG,CAAC,WAAW;QAC5B,gBAAgB,EAAE,GAAG,EAAE,CAAC,GAAG,CAAC,gBAAgB,EAAE;QAC9C,MAAM,EAAE,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC;QACzD,eAAe,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,EAAE,CAAC,eAAe,CAAC,KAAK,CAAC;QACrD,kBAAkB,EAAE,CAAC,OAAO,EAAE,EAAE,CAAC,EAAE,CAAC,kBAAkB,CAAC,OAAO,CAAC;KAC/D,CAAC;AACH,CAAC;AAED,MAAM,CAAC,OAAO,UAAU,oBAAoB,CAAC,EAAgB;IAC5D,IAAI,mBAA4C,CAAC;IACjD;;;;;;;;;;;;OAYG;IACH,IAAI,eAAe,GAAG,KAAK,CAAC;IAC5B;;;;;;;OAOG;IACH,IAAI,cAAuC,CAAC;IAC5C,IAAI,aAAiC,CAAC;IAEtC,EAAE,CAAC,EAAE,CAAC,eAAe,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,EAAE;QAC5C,MAAM,eAAe,GAAG,eAAe,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,GAAG,CAAC,QAAQ,EAAE,EAAE,cAAc,EAAE,GAAG,CAAC,gBAAgB,EAAE,EAAE,CAAC,CAAC;QAClH,mBAAmB,GAAG,eAAe,CAAC,oBAAoB,EAAE,CAAC;QAE7D,MAAM,KAAK,GAAG,GAAG,CAAC,KAAK,CAAC;QACxB,IAAI,KAAK,EAAE,GAAG,KAAK,0BAA0B,EAAE,CAAC;YAC/C,eAAe,GAAG,KAAK,CAAC;YACxB,EAAE,CAAC,kBAAkB,CAAC,KAAK,CAAC,CAAC;YAC7B,OAAO;QACR,CAAC;QAED,0FAA0F;QAC1F,4FAA4F;QAC5F,yFAAyF;QACzF,wFAAwF;QACxF,uEAAuE;QACvE,EAAE;QACF,wFAAwF;QACxF,0FAA0F;QAC1F,4BAA4B;QAC5B,MAAM,UAAU,GAAG,wBAAwB,CAAC,eAAe,EAAE,GAAG,CAAC,aAAa,EAAE,KAAK,CAAC,CAAC;QAEvF,MAAM,SAAS,GAAG,aAAa,CAAC,GAAG,CAAC,aAAa,EAAE,KAAK,CAAC,CAAC;QAC1D,IAAI,SAAS,EAAE,CAAC;YACf,MAAM,EAAE,CAAC,eAAe,CAAC,SAAS,CAAC,CAAC;QACrC,CAAC;QAED,eAAe,GAAG,UAAU,KAAK,aAAa,IAAI,CAAC,UAAU,KAAK,SAAS,IAAI,GAAG,CAAC,WAAW,KAAK,aAAa,CAAC,CAAC;QAClH,EAAE,CAAC,kBAAkB,CAAC,eAAe,CAAC,CAAC;QACvC,MAAM,WAAW,GAAG,6BAA6B,CAAC,GAAG,CAAC,aAAa,EAAE,KAAK,CAAC,CAAC;QAC5E,aAAa,GAAG,GAAG,WAAW,CAAC,QAAQ,IAAI,WAAW,CAAC,EAAE,EAAE,CAAC;QAC5D,cAAc,GAAG,UAAU,CAAC;IAC7B,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,EAAE,CAAC,cAAc,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE,EAAE;QACpC,6FAA6F;QAC7F,yFAAyF;QACzF,8FAA8F;QAC9F,+FAA+F;QAC/F,+DAA+D;QAC/D,MAAM,eAAe,GAAG,eAAe,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,GAAG,CAAC,QAAQ,EAAE,EAAE,cAAc,EAAE,GAAG,CAAC,gBAAgB,EAAE,EAAE,CAAC,CAAC;QAClH,MAAM,WAAW,GAAG,6BAA6B,CAAC,GAAG,CAAC,aAAa,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC;QAClF,aAAa,GAAG,GAAG,WAAW,CAAC,QAAQ,IAAI,WAAW,CAAC,EAAE,EAAE,CAAC;QAC5D,cAAc,GAAG,wBAAwB,CAAC,eAAe,EAAE,GAAG,CAAC,aAAa,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC;IAC5F,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,eAAe,CAAC,MAAM,EAAE;QAC1B,WAAW,EAAE,6DAA6D;QAC1E,YAAY,EAAE,UAAU;QACxB,sBAAsB,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,aAAa,CAAC,cAAc,EAAE,MAAM,CAAC;QACzE,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;YAC5B,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;YAC3C,IAAI,QAAQ,KAAK,EAAE,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,QAAwB,CAAC,EAAE,CAAC;gBAC3E,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC;gBACnC,OAAO;YACR,CAAC;YAED,+CAA+C;YAC/C,MAAM,MAAM,GAAG,eAAe,IAAI,GAAG,CAAC,WAAW,KAAK,aAAa,CAAC;YACpE,MAAM,OAAO,GAAG,QAAQ,KAAK,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,QAAQ,KAAK,IAAI,CAAC;YAC9D,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,iBAAiB,CAAC,EAAE,EAAE,GAAG,CAAC,EAAE,OAAO,CAAC,CAAC;YACxE,eAAe,GAAG,MAAM,CAAC,OAAO,CAAC;YACjC,0FAA0F;YAC1F,2FAA2F;YAC3F,iCAAiC;YACjC,IAAI,MAAM,CAAC,OAAO,IAAI,GAAG,CAAC,KAAK,EAAE,CAAC;gBACjC,MAAM,WAAW,GAAG,6BAA6B,CAAC,GAAG,CAAC,aAAa,EAAE,GAAG,CAAC,KAAK,CAAC,CAAC;gBAChF,aAAa,GAAG,GAAG,WAAW,CAAC,QAAQ,IAAI,WAAW,CAAC,EAAE,EAAE,CAAC;gBAC5D,cAAc,GAAG,MAAM,CAAC,YAAY,CAAC;YACtC,CAAC;QACF,CAAC;KACD,CAAC,CAAC;IAEH,EAAE,CAAC,EAAE,CAAC,yBAAyB,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE,EAAE;QAC/C,IAAI,oBAA6C,CAAC;QAClD,IAAI,GAAG,CAAC,KAAK,EAAE,GAAG,KAAK,0BAA0B,EAAE,CAAC;YACnD,IAAI,eAAe,EAAE,CAAC;gBACrB,oBAAoB,GAAG,aAAa,CAAC;YACtC,CAAC;iBAAM,CAAC;gBACP,MAAM,WAAW,GAAG,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,6BAA6B,CAAC,GAAG,CAAC,aAAa,EAAE,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;gBACxG,MAAM,aAAa,GAAG,WAAW,CAAC,CAAC,CAAC,GAAG,WAAW,CAAC,QAAQ,IAAI,WAAW,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;gBAC5F,sFAAsF;gBACtF,wFAAwF;gBACxF,uFAAuF;gBACvF,oFAAoF;gBACpF,+EAA+E;gBAC/E,MAAM,uBAAuB,GAC5B,GAAG,CAAC,KAAK,KAAK,SAAS,IAAI,GAAG,CAAC,aAAa,CAAC,cAAc,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,aAAa,CAAC;gBAC1F,IACC,cAAc,KAAK,MAAM;oBACzB,aAAa,KAAK,aAAa;oBAC/B,GAAG,CAAC,WAAW,KAAK,aAAa;oBACjC,uBAAuB,EACtB,CAAC;oBACF,oBAAoB,GAAG,SAAS,CAAC;gBAClC,CAAC;qBAAM,CAAC;oBACP,oBAAoB,GAAG,GAAG,CAAC,WAAW,CAAC;gBACxC,CAAC;YACF,CAAC;QACF,CAAC;aAAM,CAAC;YACP,oBAAoB,GAAG,GAAG,CAAC,WAAW,IAAI,mBAAmB,CAAC;QAC/D,CAAC;QACD,OAAO,uBAAuB,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,EAAE,KAAK,CAAC,OAAO,EAAE,oBAAoB,CAAC,CAAC;IACrF,CAAC,CAAC,CAAC;AACJ,CAAC","sourcesContent":["import type { Api, Model } from \"@earendil-works/pi-ai\";\nimport type { AutocompleteItem } from \"@earendil-works/pi-tui\";\nimport type { ModelRegistry } from \"../../model-registry.ts\";\nimport { SettingsManager } from \"../../settings-manager.ts\";\nimport type { ExtensionAPI, ExtensionCommandContext, ServiceTier } from \"../types.ts\";\n\nexport type { ServiceTier };\n\ntype ProviderPayload = Record<string, unknown>;\n\nconst OPENAI_CODEX_RESPONSES_API = \"openai-codex-responses\";\nexport const CODEX_RESPONSES_API = OPENAI_CODEX_RESPONSES_API;\nconst FAST_MODEL_SUFFIX = \"-fast\";\nconst PRIORITY_TIER: ServiceTier = \"priority\";\nconst SERVICE_TIER_APIS: ReadonlySet<Api> = new Set([\"openai-responses\", OPENAI_CODEX_RESPONSES_API]);\nconst FAST_ARGUMENTS = [\"on\", \"off\"] as const;\nconst FAST_USAGE = \"Usage: /fast [on|off]\";\n\ntype FastArgument = (typeof FAST_ARGUMENTS)[number];\n\nfunction isRecord(value: unknown): value is ProviderPayload {\n\treturn typeof value === \"object\" && value !== null;\n}\n\nexport function addServiceTierToPayload(api: Api | undefined, payload: unknown, serviceTier?: ServiceTier): unknown {\n\tif (!api || !SERVICE_TIER_APIS.has(api) || !serviceTier) {\n\t\treturn payload;\n\t}\n\n\tif (!isRecord(payload) || payload.service_tier !== undefined) {\n\t\treturn payload;\n\t}\n\n\treturn {\n\t\t...payload,\n\t\tservice_tier: serviceTier,\n\t};\n}\n\nfunction getRequestModelId(modelRegistry: ModelRegistry, model: Model<Api>): string {\n\treturn modelRegistry.getUpstreamModelId(model) ?? model.id;\n}\n\nfunction isCompatibleFastVariant(modelRegistry: ModelRegistry, baseModel: Model<Api>, fastModel: Model<Api>): boolean {\n\treturn (\n\t\tfastModel.provider === baseModel.provider &&\n\t\tfastModel.api === baseModel.api &&\n\t\tmodelRegistry.getServiceTier(fastModel) === \"priority\" &&\n\t\tgetRequestModelId(modelRegistry, fastModel) === getRequestModelId(modelRegistry, baseModel)\n\t);\n}\n\nfunction findBaseModel(modelRegistry: ModelRegistry, fastModel: Model<Api>): Model<Api> | undefined {\n\tif (!fastModel.id.endsWith(FAST_MODEL_SUFFIX)) {\n\t\treturn undefined;\n\t}\n\n\tconst baseModelId = fastModel.id.slice(0, -FAST_MODEL_SUFFIX.length);\n\tconst baseModel = modelRegistry.find(fastModel.provider, baseModelId);\n\treturn baseModel && isCompatibleFastVariant(modelRegistry, baseModel, fastModel) ? baseModel : undefined;\n}\n\nfunction findFastModel(modelRegistry: ModelRegistry, baseModel: Model<Api>): Model<Api> | undefined {\n\tif (baseModel.id.endsWith(FAST_MODEL_SUFFIX)) {\n\t\treturn undefined;\n\t}\n\n\tconst fastModel = modelRegistry.find(baseModel.provider, `${baseModel.id}${FAST_MODEL_SUFFIX}`);\n\treturn fastModel && isCompatibleFastVariant(modelRegistry, baseModel, fastModel) ? fastModel : undefined;\n}\n\n/**\n * The model that owns this model's persisted service-tier preference.\n *\n * A `-fast` catalog variant and its base model are one model to the user (`/fast` just swaps\n * between them), so both must read and write ONE key. Anything else can hold contradictory\n * preferences for the same choice.\n */\nexport function resolveServiceTierMemoryModel(modelRegistry: ModelRegistry, model: Model<Api>): Model<Api> {\n\treturn findBaseModel(modelRegistry, model) ?? model;\n}\n\n/** Remembered service tier for this model, normalized onto its base-model key. */\nexport function getRememberedServiceTier(\n\tsettingsManager: SettingsManager,\n\tmodelRegistry: ModelRegistry,\n\tmodel: Model<Api>,\n): ServiceTier | undefined {\n\tconst memoryModel = resolveServiceTierMemoryModel(modelRegistry, model);\n\treturn settingsManager.getModelServiceTier(memoryModel.provider, memoryModel.id);\n}\n\n/** Host capabilities `applyFastMode` needs; satisfied by an extension context plus `pi`. */\nexport interface FastModeContext {\n\tcwd: string;\n\tagentDir: string;\n\tmodel: Model<Api> | undefined;\n\tmodelRegistry: ModelRegistry;\n\t/** Tier the host resolved for the active model (scoped/favorite pin or catalog). */\n\tserviceTier: ServiceTier | undefined;\n\tisProjectTrusted(): boolean;\n\tnotify(message: string, level: \"info\" | \"warning\" | \"error\"): void;\n\tsetSessionModel(model: Model<Api>): Promise<boolean>;\n\tsetSessionFastMode(enabled: boolean): void;\n}\n\nexport interface FastModeResult {\n\t/** Fast mode state after the call; unchanged when the request was refused. */\n\tenabled: boolean;\n\t/** False when the request was refused (non-Codex model, active pin, failed model switch). */\n\tapplied: boolean;\n\t/** The just-persisted tier (or the unchanged one when refused); `\"priority\"` on, `\"auto\"` off. */\n\trecordedTier: ServiceTier;\n\tmessage: string;\n}\n\n/**\n * Whether the active model's priority tier is PINNED by the user's model selection (a scoped\n * or favorite `provider/id:priority` pattern) rather than inherited from the catalog.\n *\n * `ctx.serviceTier` merges both sources, so a priority the catalog does not explain must have\n * come from a `:priority` decorator; checking `scopedModels` alone would miss favorite pins,\n * which the session resolves the same way.\n */\nfunction hasPriorityPin(ctx: FastModeContext, model: Model<Api>): boolean {\n\tif (ctx.serviceTier !== PRIORITY_TIER) return false;\n\treturn ctx.modelRegistry.getServiceTier(model) !== PRIORITY_TIER;\n}\n\n/**\n * Single entry point for turning fast mode on/off: the `/fast` command and any other surface\n * (RPC) must go through this so persistence and `-fast` key normalization exist exactly once.\n *\n * `off` persists an explicit `\"auto\"` rather than deleting the key: deletion re-inherits the\n * catalog/global tier, which is what the user just asked to turn off.\n */\nexport async function applyFastMode(ctx: FastModeContext, enabled: boolean): Promise<FastModeResult> {\n\tconst model = ctx.model;\n\tif (model?.api !== OPENAI_CODEX_RESPONSES_API) {\n\t\tconst message = \"Fast mode is only available for OpenAI Codex models.\";\n\t\tctx.notify(message, \"warning\");\n\t\treturn { enabled: false, applied: false, recordedTier: \"auto\", message };\n\t}\n\n\tif (!enabled && hasPriorityPin(ctx, model)) {\n\t\t// The pin outranks the memory, so writing \"auto\" here would be a silent no-op on the wire.\n\t\tconst message = \"Fast mode is fixed by the active model selection's priority tier.\";\n\t\tctx.notify(message, \"info\");\n\t\t// The pin keeps the model on priority no matter what the memory says, so that is the honest\n\t\t// tier to report; nothing was written, and no caller reads this field when `applied` is false.\n\t\treturn { enabled: true, applied: false, recordedTier: PRIORITY_TIER, message };\n\t}\n\n\tconst settingsManager = SettingsManager.create(ctx.cwd, ctx.agentDir, { projectTrusted: ctx.isProjectTrusted() });\n\tconst memoryModel = resolveServiceTierMemoryModel(ctx.modelRegistry, model);\n\tconst tier: ServiceTier = enabled ? PRIORITY_TIER : \"auto\";\n\tsettingsManager.setModelServiceTier(memoryModel.provider, memoryModel.id, tier);\n\tawait settingsManager.flush();\n\n\tconst baseModel = findBaseModel(ctx.modelRegistry, model);\n\tconst targetModel = enabled ? findFastModel(ctx.modelRegistry, model) : baseModel;\n\tif (targetModel) {\n\t\tif (!(await ctx.setSessionModel(targetModel))) {\n\t\t\tconst message = `Could not switch to ${targetModel.provider}/${targetModel.id}.`;\n\t\t\tctx.notify(message, \"error\");\n\t\t\treturn { enabled: baseModel === undefined, applied: false, recordedTier: \"auto\", message };\n\t\t}\n\t}\n\n\tctx.setSessionFastMode(enabled);\n\tconst labelModel = targetModel ?? model;\n\tconst message = `Fast mode ${enabled ? \"enabled\" : \"disabled\"}: ${labelModel.id}`;\n\tctx.notify(message, \"info\");\n\treturn { enabled, applied: true, recordedTier: tier, message };\n}\n\nfunction toCompletions(values: readonly string[], prefix: string): AutocompleteItem[] | null {\n\tconst matches = values.filter((value) => value.startsWith(prefix.trim()));\n\treturn matches.length > 0 ? matches.map((value) => ({ value, label: value })) : null;\n}\n\nfunction toFastModeContext(pi: ExtensionAPI, ctx: ExtensionCommandContext): FastModeContext {\n\treturn {\n\t\tcwd: ctx.cwd,\n\t\tagentDir: ctx.agentDir,\n\t\tmodel: ctx.model,\n\t\tmodelRegistry: ctx.modelRegistry,\n\t\tserviceTier: ctx.serviceTier,\n\t\tisProjectTrusted: () => ctx.isProjectTrusted(),\n\t\tnotify: (message, level) => ctx.ui.notify(message, level),\n\t\tsetSessionModel: (model) => pi.setSessionModel(model),\n\t\tsetSessionFastMode: (enabled) => pi.setSessionFastMode(enabled),\n\t};\n}\n\nexport default function serviceTierExtension(pi: ExtensionAPI): void {\n\tlet settingsServiceTier: ServiceTier | undefined;\n\t/**\n\t * Session-level fast mode for Codex models that have no `-fast` catalog sibling.\n\t *\n\t * The catalog generator emits `-fast` priority variants only for the direct\n\t * `openai` provider, so `openai-codex` models never have a switch target. The\n\t * ChatGPT backend still offers the tier to subscriptions:\n\t * `chatgpt.com/backend-api/codex/models` advertises a `priority` service tier\n\t * labelled \"Fast\" (\"1.5x speed, increased usage\")\n\t * for gpt-5.4/5.5/5.6-*, and the first-party Codex CLI sends\n\t * `service_tier: \"priority\"` on subscription OAuth. The response echo is not a\n\t * confirmation channel — it reads `auto`/`default` whether or not a tier was\n\t * sent — so this toggle drives the request field directly. See issue #545.\n\t */\n\tlet sessionFastMode = false;\n\t/**\n\t * The per-model memory tier `/fast` last wrote or read for the active base model this\n\t * session. Lives only to suppress an inherited priority when the user turns fast off in\n\t * the SAME session: the host's resolved `ctx.serviceTier` is recomputed on model switch,\n\t * not on a memory write, so without this an in-session `/fast off` (no model swap) would\n\t * keep sending `priority` until a restart. Scoped to the active base key (re-derived on\n\t * model_select for the incoming model) so it never leaks across models.\n\t */\n\tlet liveMemoryTier: ServiceTier | undefined;\n\tlet liveMemoryKey: string | undefined;\n\n\tpi.on(\"session_start\", async (_event, ctx) => {\n\t\tconst settingsManager = SettingsManager.create(ctx.cwd, ctx.agentDir, { projectTrusted: ctx.isProjectTrusted() });\n\t\tsettingsServiceTier = settingsManager.getOpenAIServiceTier();\n\n\t\tconst model = ctx.model;\n\t\tif (model?.api !== OPENAI_CODEX_RESPONSES_API) {\n\t\t\tsessionFastMode = false;\n\t\t\tpi.setSessionFastMode(false);\n\t\t\treturn;\n\t\t}\n\n\t\t// Fast mode is remembered per model, so a new session inherits the user's last choice for\n\t\t// THIS model instead of silently starting off. A model whose catalog/scoped tier is already\n\t\t// `priority` (a models.json priority entry, or a `-fast`-less codex model) is treated as\n\t\t// fast-active when nothing is remembered; an explicit remembered `\"auto\"` (`/fast off`)\n\t\t// wins over that inheritance. Malformed values read back as undefined.\n\t\t//\n\t\t// The flag is derived from the POST-swap model: a `-fast` catalog variant swaps down to\n\t\t// its base before this reads `ctx.serviceTier`, so inheritance is judged on the model the\n\t\t// user actually ends up on.\n\t\tconst remembered = getRememberedServiceTier(settingsManager, ctx.modelRegistry, model);\n\n\t\tconst baseModel = findBaseModel(ctx.modelRegistry, model);\n\t\tif (baseModel) {\n\t\t\tawait pi.setSessionModel(baseModel);\n\t\t}\n\n\t\tsessionFastMode = remembered === PRIORITY_TIER || (remembered === undefined && ctx.serviceTier === PRIORITY_TIER);\n\t\tpi.setSessionFastMode(sessionFastMode);\n\t\tconst memoryModel = resolveServiceTierMemoryModel(ctx.modelRegistry, model);\n\t\tliveMemoryKey = `${memoryModel.provider}/${memoryModel.id}`;\n\t\tliveMemoryTier = remembered;\n\t});\n\n\tpi.on(\"model_select\", (event, ctx) => {\n\t\t// A model switch changes the base key the memory lives under, so the live tier is RE-DERIVED\n\t\t// for the incoming model instead of merely dropped: dropping it would leave a remembered\n\t\t// \"auto\" unable to suppress a catalog-inherited priority after switching away and back in one\n\t\t// session, silently re-sending the tier `/fast off` turned off. Per-key scoping is preserved —\n\t\t// each model reads ITS OWN memory, never the previous model's.\n\t\tconst settingsManager = SettingsManager.create(ctx.cwd, ctx.agentDir, { projectTrusted: ctx.isProjectTrusted() });\n\t\tconst memoryModel = resolveServiceTierMemoryModel(ctx.modelRegistry, event.model);\n\t\tliveMemoryKey = `${memoryModel.provider}/${memoryModel.id}`;\n\t\tliveMemoryTier = getRememberedServiceTier(settingsManager, ctx.modelRegistry, event.model);\n\t});\n\n\tpi.registerCommand(\"fast\", {\n\t\tdescription: \"Turn OpenAI Codex fast mode on or off for the current model\",\n\t\targumentHint: \"[on|off]\",\n\t\tgetArgumentCompletions: (prefix) => toCompletions(FAST_ARGUMENTS, prefix),\n\t\thandler: async (args, ctx) => {\n\t\t\tconst argument = args.trim().toLowerCase();\n\t\t\tif (argument !== \"\" && !FAST_ARGUMENTS.includes(argument as FastArgument)) {\n\t\t\t\tctx.ui.notify(FAST_USAGE, \"error\");\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// No argument keeps the established toggle UX.\n\t\t\tconst active = sessionFastMode || ctx.serviceTier === PRIORITY_TIER;\n\t\t\tconst enabled = argument === \"\" ? !active : argument === \"on\";\n\t\t\tconst result = await applyFastMode(toFastModeContext(pi, ctx), enabled);\n\t\t\tsessionFastMode = result.enabled;\n\t\t\t// Track this session's own write AFTER applyFastMode's internal model switch (which fires\n\t\t\t// model_select and clears the live tier). Keyed to the active base model so a later switch\n\t\t\t// cannot carry it across models.\n\t\t\tif (result.applied && ctx.model) {\n\t\t\t\tconst memoryModel = resolveServiceTierMemoryModel(ctx.modelRegistry, ctx.model);\n\t\t\t\tliveMemoryKey = `${memoryModel.provider}/${memoryModel.id}`;\n\t\t\t\tliveMemoryTier = result.recordedTier;\n\t\t\t}\n\t\t},\n\t});\n\n\tpi.on(\"before_provider_request\", (event, ctx) => {\n\t\tlet effectiveServiceTier: ServiceTier | undefined;\n\t\tif (ctx.model?.api === OPENAI_CODEX_RESPONSES_API) {\n\t\t\tif (sessionFastMode) {\n\t\t\t\teffectiveServiceTier = PRIORITY_TIER;\n\t\t\t} else {\n\t\t\t\tconst memoryModel = ctx.model ? resolveServiceTierMemoryModel(ctx.modelRegistry, ctx.model) : undefined;\n\t\t\t\tconst activeBaseKey = memoryModel ? `${memoryModel.provider}/${memoryModel.id}` : undefined;\n\t\t\t\t// Same-session /fast off: the host's resolved ctx.serviceTier is only recomputed on a\n\t\t\t\t// model switch, so an inherited priority would otherwise keep emitting until a restart.\n\t\t\t\t// Only a CATALOG-explained priority may be suppressed: a priority the catalog does not\n\t\t\t\t// explain is an explicit scoped/favorite `:priority` pin, which outranks the memory\n\t\t\t\t// (same discriminator `applyFastMode` uses to refuse `/fast off` under a pin).\n\t\t\t\tconst catalogExplainsPriority =\n\t\t\t\t\tctx.model !== undefined && ctx.modelRegistry.getServiceTier(ctx.model) === PRIORITY_TIER;\n\t\t\t\tif (\n\t\t\t\t\tliveMemoryTier === \"auto\" &&\n\t\t\t\t\tactiveBaseKey === liveMemoryKey &&\n\t\t\t\t\tctx.serviceTier === PRIORITY_TIER &&\n\t\t\t\t\tcatalogExplainsPriority\n\t\t\t\t) {\n\t\t\t\t\teffectiveServiceTier = undefined;\n\t\t\t\t} else {\n\t\t\t\t\teffectiveServiceTier = ctx.serviceTier;\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\teffectiveServiceTier = ctx.serviceTier ?? settingsServiceTier;\n\t\t}\n\t\treturn addServiceTierToPayload(ctx.model?.api, event.payload, effectiveServiceTier);\n\t});\n}\n"]}