@astralyn/pi 0.87.6 → 0.99.0

package/dist/extensions/llama/provider.js

@@ -1,3 +1,5 @@
+import { isModelType, } from "@earendil-works/pi-ai";
+import { llamaCppClassifyApi } from "@earendil-works/pi-ai/api/llama-cpp-classify.lazy";
 import { stream, streamSimple } from "@earendil-works/pi-ai/compat";
 import { LlamaClient, llamaInferenceUrl, normalizeLlamaServerUrl, } from "./client.js";
 export const LLAMA_PROVIDER_ID = "llama.cpp";
@@ -29,9 +31,46 @@ async function routerAutoloadEnabled(client, catalog, signal) {
         return false;
     }
 }
-function toPiModel(model, serverUrl, props) {
-    const reportedContextWindow = model.meta?.n_ctx ?? model.meta?.n_ctx_train;
-    const contextWindow = reportedContextWindow && reportedContextWindow > 0 ? reportedContextWindow : 128000;
+function configuredContextWindow(model) {
+    const args = model.status.args ?? [];
+    for (let index = 0; index < args.length - 1; index++) {
+        const flag = args[index];
+        if (flag !== "--ctx-size" && flag !== "-c" && flag !== "-ctx")
+            continue;
+        const contextWindow = Number(args[index + 1]);
+        if (Number.isSafeInteger(contextWindow) && contextWindow > 0)
+            return contextWindow;
+    }
+    return undefined;
+}
+function contextWindowOf(model, cachedContextWindow) {
+    const runtimeContextWindow = model.meta?.n_ctx;
+    if (runtimeContextWindow && runtimeContextWindow > 0)
+        return runtimeContextWindow;
+    const configuredContext = configuredContextWindow(model);
+    if (configuredContext)
+        return configuredContext;
+    if (cachedContextWindow && cachedContextWindow > 0)
+        return cachedContextWindow;
+    const trainingContextWindow = model.meta?.n_ctx_train;
+    return trainingContextWindow && trainingContextWindow > 0 ? trainingContextWindow : 128000;
+}
+/** The same llama.cpp model used as a classifier: answers are read from next-token label probabilities. */
+function toPiClassifierModel(model, serverUrl, cachedContextWindow) {
+    return {
+        type: "classifier",
+        id: model.id,
+        name: model.id,
+        api: "llama-cpp-classify",
+        provider: LLAMA_PROVIDER_ID,
+        baseUrl: serverUrl,
+        input: ["text"],
+        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+        contextWindow: contextWindowOf(model, cachedContextWindow),
+    };
+}
+function toPiModel(model, serverUrl, props, cachedContextWindow) {
+    const contextWindow = contextWindowOf(model, cachedContextWindow);
     const reasoning = props?.chat_template?.includes("enable_thinking") === true;
     return {
         id: model.id,
@@ -60,10 +99,12 @@ function toPiModel(model, serverUrl, props) {
 }
 export function createLlamaProvider() {
     let models = [];
+    let classifiers = [];
+    const classifier = llamaCppClassifyApi();
     const setCatalog = (catalog, serverUrl, options = {}) => {
-        models = catalog
-            .filter((model) => modelIsSelectable(model, options.routerAutoload === true))
-            .map((model) => toPiModel(model, serverUrl));
+        const selectable = catalog.filter((model) => modelIsSelectable(model, options.routerAutoload === true));
+        models = selectable.map((model) => toPiModel(model, serverUrl));
+        classifiers = selectable.map((model) => toPiClassifierModel(model, serverUrl));
     };
     const provider = {
         id: LLAMA_PROVIDER_ID,
@@ -110,12 +151,20 @@ export function createLlamaProvider() {
             },
         },
         getModels: () => models,
+        getAllModels: () => [...models, ...classifiers],
         refreshModels: async (context) => {
+            const cachedContextWindows = new Map();
             if (context.stored) {
-                const restored = context.stored.models.filter((model) => model.provider === LLAMA_PROVIDER_ID && model.api === "openai-completions");
+                const stored = context.stored.models.filter((model) => model.provider === LLAMA_PROVIDER_ID);
+                const restored = stored.filter((model) => isModelType(model, "chat") && model.api === "openai-completions");
+                const restoredClassifiers = stored.filter((model) => isModelType(model, "classifier") && model.api === "llama-cpp-classify");
+                for (const model of [...restored, ...restoredClassifiers]) {
+                    cachedContextWindows.set(model.id, model.contextWindow);
+                }
                 if (!(await context.publish({
                     update: () => {
                         models = restored;
+                        classifiers = restoredClassifiers;
                     },
                 }))) {
                     return;
@@ -133,28 +182,31 @@ export function createLlamaProvider() {
             const routerAutoload = await routerAutoloadEnabled(client, catalog, context.signal);
             if (context.signal.aborted)
                 return;
-            const refreshed = await Promise.all(catalog
-                .filter((model) => modelIsSelectable(model, routerAutoload))
-                .map(async (model) => {
+            const selectable = catalog.filter((model) => modelIsSelectable(model, routerAutoload));
+            const refreshed = await Promise.all(selectable.map(async (model) => {
+                const cachedContextWindow = cachedContextWindows.get(model.id);
                 // Only loaded models expose their template without side effects. Unloaded autoload presets
                 // would need to be loaded, while querying sleeping models may wake them. Those models remain
                 // unclassified until they are loaded or woken and a later catalog refresh discovers them.
                 if (model.status.value !== "loaded")
-                    return toPiModel(model, serverUrl);
+                    return toPiModel(model, serverUrl, undefined, cachedContextWindow);
                 const props = await client.props({ model: model.id, signal: context.signal });
-                return toPiModel(model, serverUrl, props);
+                return toPiModel(model, serverUrl, props, cachedContextWindow);
             }));
+            const refreshedClassifiers = selectable.map((model) => toPiClassifierModel(model, serverUrl, cachedContextWindows.get(model.id)));
             if (context.signal.aborted)
                 return;
             await context.publish({
-                persist: { models: refreshed, checkedAt: Date.now() },
+                persist: { models: [...refreshed, ...refreshedClassifiers], checkedAt: Date.now() },
                 update: () => {
                     models = refreshed;
+                    classifiers = refreshedClassifiers;
                 },
             });
         },
         stream: (model, context, options) => stream(model, context, options),
         streamSimple: (model, context, options) => streamSimple(model, context, options),
+        classify: (model, context, options) => classifier.classify(model, context, options),
     };
     return { provider, setCatalog };
 }

package/dist/extensions/llama/provider.js.map

@@ -1 +1 @@
-{"version":3,"file":"provider.js","sourceRoot":"","sources":["../../../src/extensions/llama/provider.ts"],"names":[],"mappings":"AASA,OAAO,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,8BAA8B,CAAC;AACpE,OAAO,EACN,WAAW,EAGX,iBAAiB,EACjB,uBAAuB,GACvB,MAAM,aAAa,CAAC;AAErB,MAAM,CAAC,MAAM,iBAAiB,GAAG,WAAW,CAAC;AAC7C,MAAM,CAAC,MAAM,wBAAwB,GAAG,uBAAuB,CAAC;AAChE,SAAS,mBAAmB,CAAC,UAAwC;IACpE,MAAM,KAAK,GAAG,UAAU,EAAE,GAAG,EAAE,cAAc,CAAC;IAC9C,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,uBAAuB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAC/F,CAAC;AAED,KAAK,UAAU,gBAAgB,CAC9B,GAAgB,EAChB,UAAwC;IAExC,MAAM,UAAU,GAAG,mBAAmB,CAAC,UAAU,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC;IAChG,OAAO,UAAU,CAAC,CAAC,CAAC,uBAAuB,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AACrE,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAqB,EAAE,cAAuB;IACxE,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ;QAAE,OAAO,IAAI,CAAC;IACjD,uFAAuF;IACvF,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU;QAAE,OAAO,IAAI,CAAC;IACnD,gGAAgG;IAChG,OAAO,cAAc,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,IAAI,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC;AACjH,CAAC;AAED,KAAK,UAAU,qBAAqB,CACnC,MAAmB,EACnB,OAAkC,EAClC,MAAmB;IAEnB,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,IAAI,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC;QAAE,OAAO,KAAK,CAAC;IAC3G,IAAI,CAAC;QACJ,OAAO,CAAC,MAAM,MAAM,CAAC,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,eAAe,KAAK,IAAI,CAAC;IAClE,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,KAAK,CAAC;IACd,CAAC;AACF,CAAC;AAED,SAAS,SAAS,CAAC,KAAqB,EAAE,SAAiB,EAAE,KAAwB;IACpF,MAAM,qBAAqB,GAAG,KAAK,CAAC,IAAI,EAAE,KAAK,IAAI,KAAK,CAAC,IAAI,EAAE,WAAW,CAAC;IAC3E,MAAM,aAAa,GAAG,qBAAqB,IAAI,qBAAqB,GAAG,CAAC,CAAC,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC,MAAM,CAAC;IAC1G,MAAM,SAAS,GAAG,KAAK,EAAE,aAAa,EAAE,QAAQ,CAAC,iBAAiB,CAAC,KAAK,IAAI,CAAC;IAC7E,OAAO;QACN,EAAE,EAAE,KAAK,CAAC,EAAE;QACZ,IAAI,EAAE,KAAK,CAAC,EAAE;QACd,GAAG,EAAE,oBAAoB;QACzB,QAAQ,EAAE,iBAAiB;QAC3B,OAAO,EAAE,iBAAiB,CAAC,SAAS,CAAC;QACrC,SAAS;QACT,GAAG,CAAC,SAAS,IAAI;YAChB,gBAAgB,EAAE,EAAE,GAAG,EAAE,KAAK,EAAE,OAAO,EAAE,IAAI,EAAE,GAAG,EAAE,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE;SACrG,CAAC;QACF,KAAK,EAAE,KAAK,CAAC,YAAY,EAAE,gBAAgB,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC;QAC7F,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE;QAC1D,aAAa;QACb,SAAS,EAAE,aAAa;QACxB,MAAM,EAAE;YACP,aAAa,EAAE,KAAK;YACpB,qBAAqB,EAAE,KAAK;YAC5B,uBAAuB,EAAE,KAAK;YAC9B,wBAAwB,EAAE,IAAI;YAC9B,kBAAkB,EAAE,KAAK;YACzB,cAAc,EAAE,YAAY;YAC5B,GAAG,CAAC,SAAS,IAAI,EAAE,cAAc,EAAE,oBAAoB,EAAE,CAAC;SAC1D;KACD,CAAC;AACH,CAAC;AAOD,MAAM,UAAU,mBAAmB;IAClC,IAAI,MAAM,GAA2C,EAAE,CAAC;IAExD,MAAM,UAAU,GAAG,CAClB,OAAkC,EAClC,SAAiB,EACjB,UAAwC,EAAE,EACnC,EAAE;QACT,MAAM,GAAG,OAAO;aACd,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,iBAAiB,CAAC,KAAK,EAAE,OAAO,CAAC,cAAc,KAAK,IAAI,CAAC,CAAC;aAC5E,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,SAAS,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC,CAAC;IAC/C,CAAC,CAAC;IAEF,MAAM,QAAQ,GAAmC;QAChD,EAAE,EAAE,iBAAiB;QACrB,IAAI,EAAE,WAAW;QACjB,OAAO,EAAE,iBAAiB,CAAC,wBAAwB,CAAC;QACpD,IAAI,EAAE;YACL,MAAM,EAAE;gBACP,IAAI,EAAE,kBAAkB;gBACxB,KAAK,EAAE,KAAK,EAAE,WAAW,EAA6B,EAAE;oBACvD,MAAM,UAAU,GAAG,MAAM,WAAW,CAAC,MAAM,CAAC;wBAC3C,IAAI,EAAE,MAAM;wBACZ,OAAO,EAAE,sBAAsB;wBAC/B,WAAW,EAAE,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,wBAAwB;qBACnE,CAAC,CAAC;oBACH,MAAM,SAAS,GAAG,uBAAuB,CACxC,UAAU,CAAC,IAAI,EAAE,IAAI,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,wBAAwB,CAC3E,CAAC;oBACF,MAAM,MAAM,GAAG,CACd,MAAM,WAAW,CAAC,MAAM,CAAC;wBACxB,IAAI,EAAE,QAAQ;wBACd,OAAO,EAAE,oBAAoB;qBAC7B,CAAC,CACF,CAAC,IAAI,EAAE,CAAC;oBACT,MAAM,IAAI,WAAW,CAAC,SAAS,EAAE,MAAM,IAAI,SAAS,CAAC,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,WAAW,CAAC,MAAM,EAAE,CAAC,CAAC;oBAC3F,OAAO;wBACN,IAAI,EAAE,SAAS;wBACf,GAAG,EAAE,MAAM,IAAI,SAAS;wBACxB,GAAG,EAAE,EAAE,cAAc,EAAE,SAAS,EAAE;qBAClC,CAAC;gBACH,CAAC;gBACD,KAAK,EAAE,KAAK,EAAE,EAAE,GAAG,EAAE,UAAU,EAAE,EAAE,EAAE;oBACpC,MAAM,SAAS,GAAG,MAAM,gBAAgB,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;oBAC1D,OAAO,SAAS;wBACf,CAAC,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,MAAM,EAAE,UAAU,CAAC,CAAC,CAAC,mBAAmB,CAAC,CAAC,CAAC,gBAAgB,EAAE;wBAClF,CAAC,CAAC,SAAS,CAAC;gBACd,CAAC;gBACD,OAAO,EAAE,KAAK,EAAE,EAAE,GAAG,EAAE,UAAU,EAAE,EAAmC,EAAE;oBACvE,MAAM,SAAS,GAAG,MAAM,gBAAgB,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;oBAC1D,IAAI,CAAC,SAAS;wBAAE,OAAO,SAAS,CAAC;oBACjC,MAAM,MAAM,GAAG,UAAU,EAAE,GAAG,IAAI,CAAC,MAAM,GAAG,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC,IAAI,OAAO,CAAC;oBAC9E,OAAO;wBACN,IAAI,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,iBAAiB,CAAC,SAAS,CAAC,EAAE;wBACvD,GAAG,EAAE,EAAE,GAAG,UAAU,EAAE,GAAG,EAAE,cAAc,EAAE,SAAS,EAAE;wBACtD,MAAM,EAAE,UAAU,CAAC,CAAC,CAAC,mBAAmB,CAAC,CAAC,CAAC,gBAAgB;qBAC3D,CAAC;gBACH,CAAC;aACD;SACD;QACD,SAAS,EAAE,GAAG,EAAE,CAAC,MAAM;QACvB,aAAa,EAAE,KAAK,EAAE,OAA6B,EAAiB,EAAE;YACrE,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;gBACpB,MAAM,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAC5C,CAAC,KAAK,EAAwC,EAAE,CAC/C,KAAK,CAAC,QAAQ,KAAK,iBAAiB,IAAI,KAAK,CAAC,GAAG,KAAK,oBAAoB,CAC3E,CAAC;gBACF,IACC,CAAC,CAAC,MAAM,OAAO,CAAC,OAAO,CAAC;oBACvB,MAAM,EAAE,GAAG,EAAE;wBACZ,MAAM,GAAG,QAAQ,CAAC;oBACnB,CAAC;iBACD,CAAC,CAAC,EACF,CAAC;oBACF,OAAO;gBACR,CAAC;YACF,CAAC;YAED,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,IAAI,OAAO,CAAC,UAAU,EAAE,IAAI,KAAK,SAAS;gBAAE,OAAO;YACtG,MAAM,SAAS,GAAG,mBAAmB,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;YAC1D,IAAI,CAAC,SAAS;gBAAE,OAAO;YACvB,MAAM,MAAM,GAAG,IAAI,WAAW,CAAC,SAAS,EAAE,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC;YAClE,MAAM,OAAO,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;YAC9D,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,cAAc,GAAG,MAAM,qBAAqB,CAAC,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;YACpF,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC,GAAG,CAClC,OAAO;iBACL,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,iBAAiB,CAAC,KAAK,EAAE,cAAc,CAAC,CAAC;iBAC3D,GAAG,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE;gBACpB,2FAA2F;gBAC3F,6FAA6F;gBAC7F,0FAA0F;gBAC1F,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ;oBAAE,OAAO,SAAS,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;gBACxE,MAAM,KAAK,GAAG,MAAM,MAAM,CAAC,KAAK,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;gBAC9E,OAAO,SAAS,CAAC,KAAK,EAAE,SAAS,EAAE,KAAK,CAAC,CAAC;YAC3C,CAAC,CAAC,CACH,CAAC;YACF,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,OAAO,CAAC,OAAO,CAAC;gBACrB,OAAO,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE;gBACrD,MAAM,EAAE,GAAG,EAAE;oBACZ,MAAM,GAAG,SAAS,CAAC;gBACpB,CAAC;aACD,CAAC,CAAC;QACJ,CAAC;QACD,MAAM,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,OAA4C,CAAC;QACzG,YAAY,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC;KAChF,CAAC;IAEF,OAAO,EAAE,QAAQ,EAAE,UAAU,EAAE,CAAC;AACjC,CAAC","sourcesContent":["import type {\n\tApiKeyCredential,\n\tAuthContext,\n\tAuthResult,\n\tModel,\n\tProvider,\n\tProviderStreamOptions,\n\tRefreshModelsContext,\n} from \"@earendil-works/pi-ai\";\nimport { stream, streamSimple } from \"@earendil-works/pi-ai/compat\";\nimport {\n\tLlamaClient,\n\ttype LlamaModelInfo,\n\ttype LlamaServerProps,\n\tllamaInferenceUrl,\n\tnormalizeLlamaServerUrl,\n} from \"./client.ts\";\n\nexport const LLAMA_PROVIDER_ID = \"llama.cpp\";\nexport const DEFAULT_LLAMA_SERVER_URL = \"http://127.0.0.1:8080\";\nfunction credentialServerUrl(credential: ApiKeyCredential | undefined): string | undefined {\n\tconst value = credential?.env?.LLAMA_BASE_URL;\n\treturn typeof value === \"string\" && value.trim() ? normalizeLlamaServerUrl(value) : undefined;\n}\n\nasync function resolveServerUrl(\n\tctx: AuthContext,\n\tcredential: ApiKeyCredential | undefined,\n): Promise<string | undefined> {\n\tconst configured = credentialServerUrl(credential) ?? (await ctx.env(\"LLAMA_BASE_URL\"))?.trim();\n\treturn configured ? normalizeLlamaServerUrl(configured) : undefined;\n}\n\nfunction modelIsSelectable(model: LlamaModelInfo, routerAutoload: boolean): boolean {\n\tif (model.status.value === \"loaded\") return true;\n\t// llama.cpp reports idle-slept models as \"sleeping\"; requests wake them automatically.\n\tif (model.status.value === \"sleeping\") return true;\n\t// Unloaded presets are routable only when llama.cpp router autoload can load them on first use.\n\treturn routerAutoload && model.status.value === \"unloaded\" && !model.status.failed && model.source === \"preset\";\n}\n\nasync function routerAutoloadEnabled(\n\tclient: LlamaClient,\n\tcatalog: readonly LlamaModelInfo[],\n\tsignal: AbortSignal,\n): Promise<boolean> {\n\tif (!catalog.some((model) => model.status.value === \"unloaded\" && model.source === \"preset\")) return false;\n\ttry {\n\t\treturn (await client.props({ signal })).models_autoload === true;\n\t} catch {\n\t\treturn false;\n\t}\n}\n\nfunction toPiModel(model: LlamaModelInfo, serverUrl: string, props?: LlamaServerProps): Model<\"openai-completions\"> {\n\tconst reportedContextWindow = model.meta?.n_ctx ?? model.meta?.n_ctx_train;\n\tconst contextWindow = reportedContextWindow && reportedContextWindow > 0 ? reportedContextWindow : 128000;\n\tconst reasoning = props?.chat_template?.includes(\"enable_thinking\") === true;\n\treturn {\n\t\tid: model.id,\n\t\tname: model.id,\n\t\tapi: \"openai-completions\",\n\t\tprovider: LLAMA_PROVIDER_ID,\n\t\tbaseUrl: llamaInferenceUrl(serverUrl),\n\t\treasoning,\n\t\t...(reasoning && {\n\t\t\tthinkingLevelMap: { off: \"off\", minimal: null, low: null, medium: \"medium\", high: null, xhigh: null },\n\t\t}),\n\t\tinput: model.architecture?.input_modalities?.includes(\"image\") ? [\"text\", \"image\"] : [\"text\"],\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },\n\t\tcontextWindow,\n\t\tmaxTokens: contextWindow,\n\t\tcompat: {\n\t\t\tsupportsStore: false,\n\t\t\tsupportsDeveloperRole: false,\n\t\t\tsupportsReasoningEffort: false,\n\t\t\tsupportsUsageInStreaming: true,\n\t\t\tsupportsStrictMode: false,\n\t\t\tmaxTokensField: \"max_tokens\",\n\t\t\t...(reasoning && { thinkingFormat: \"qwen-chat-template\" }),\n\t\t},\n\t};\n}\n\nexport interface LlamaProviderController {\n\tprovider: Provider<\"openai-completions\">;\n\tsetCatalog(models: readonly LlamaModelInfo[], serverUrl: string, options?: { routerAutoload?: boolean }): void;\n}\n\nexport function createLlamaProvider(): LlamaProviderController {\n\tlet models: readonly Model<\"openai-completions\">[] = [];\n\n\tconst setCatalog = (\n\t\tcatalog: readonly LlamaModelInfo[],\n\t\tserverUrl: string,\n\t\toptions: { routerAutoload?: boolean } = {},\n\t): void => {\n\t\tmodels = catalog\n\t\t\t.filter((model) => modelIsSelectable(model, options.routerAutoload === true))\n\t\t\t.map((model) => toPiModel(model, serverUrl));\n\t};\n\n\tconst provider: Provider<\"openai-completions\"> = {\n\t\tid: LLAMA_PROVIDER_ID,\n\t\tname: \"llama.cpp\",\n\t\tbaseUrl: llamaInferenceUrl(DEFAULT_LLAMA_SERVER_URL),\n\t\tauth: {\n\t\t\tapiKey: {\n\t\t\t\tname: \"llama.cpp server\",\n\t\t\t\tlogin: async (interaction): Promise<ApiKeyCredential> => {\n\t\t\t\t\tconst enteredUrl = await interaction.prompt({\n\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\tmessage: \"llama.cpp server URL\",\n\t\t\t\t\t\tplaceholder: process.env.LLAMA_BASE_URL ?? DEFAULT_LLAMA_SERVER_URL,\n\t\t\t\t\t});\n\t\t\t\t\tconst serverUrl = normalizeLlamaServerUrl(\n\t\t\t\t\t\tenteredUrl.trim() || process.env.LLAMA_BASE_URL || DEFAULT_LLAMA_SERVER_URL,\n\t\t\t\t\t);\n\t\t\t\t\tconst apiKey = (\n\t\t\t\t\t\tawait interaction.prompt({\n\t\t\t\t\t\t\ttype: \"secret\",\n\t\t\t\t\t\t\tmessage: \"API key (optional)\",\n\t\t\t\t\t\t})\n\t\t\t\t\t).trim();\n\t\t\t\t\tawait new LlamaClient(serverUrl, apiKey || undefined).list({ signal: interaction.signal });\n\t\t\t\t\treturn {\n\t\t\t\t\t\ttype: \"api_key\",\n\t\t\t\t\t\tkey: apiKey || undefined,\n\t\t\t\t\t\tenv: { LLAMA_BASE_URL: serverUrl },\n\t\t\t\t\t};\n\t\t\t\t},\n\t\t\t\tcheck: async ({ ctx, credential }) => {\n\t\t\t\t\tconst serverUrl = await resolveServerUrl(ctx, credential);\n\t\t\t\t\treturn serverUrl\n\t\t\t\t\t\t? { type: \"api_key\", source: credential ? \"stored credential\" : \"LLAMA_BASE_URL\" }\n\t\t\t\t\t\t: undefined;\n\t\t\t\t},\n\t\t\t\tresolve: async ({ ctx, credential }): Promise<AuthResult | undefined> => {\n\t\t\t\t\tconst serverUrl = await resolveServerUrl(ctx, credential);\n\t\t\t\t\tif (!serverUrl) return undefined;\n\t\t\t\t\tconst apiKey = credential?.key ?? (await ctx.env(\"LLAMA_API_KEY\")) ?? \"local\";\n\t\t\t\t\treturn {\n\t\t\t\t\t\tauth: { apiKey, baseUrl: llamaInferenceUrl(serverUrl) },\n\t\t\t\t\t\tenv: { ...credential?.env, LLAMA_BASE_URL: serverUrl },\n\t\t\t\t\t\tsource: credential ? \"stored credential\" : \"LLAMA_BASE_URL\",\n\t\t\t\t\t};\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tgetModels: () => models,\n\t\trefreshModels: async (context: RefreshModelsContext): Promise<void> => {\n\t\t\tif (context.stored) {\n\t\t\t\tconst restored = context.stored.models.filter(\n\t\t\t\t\t(model): model is Model<\"openai-completions\"> =>\n\t\t\t\t\t\tmodel.provider === LLAMA_PROVIDER_ID && model.api === \"openai-completions\",\n\t\t\t\t);\n\t\t\t\tif (\n\t\t\t\t\t!(await context.publish({\n\t\t\t\t\t\tupdate: () => {\n\t\t\t\t\t\t\tmodels = restored;\n\t\t\t\t\t\t},\n\t\t\t\t\t}))\n\t\t\t\t) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (!context.allowNetwork || context.signal.aborted || context.credential?.type !== \"api_key\") return;\n\t\t\tconst serverUrl = credentialServerUrl(context.credential);\n\t\t\tif (!serverUrl) return;\n\t\t\tconst client = new LlamaClient(serverUrl, context.credential.key);\n\t\t\tconst catalog = await client.list({ signal: context.signal });\n\t\t\tif (context.signal.aborted) return;\n\t\t\tconst routerAutoload = await routerAutoloadEnabled(client, catalog, context.signal);\n\t\t\tif (context.signal.aborted) return;\n\t\t\tconst refreshed = await Promise.all(\n\t\t\t\tcatalog\n\t\t\t\t\t.filter((model) => modelIsSelectable(model, routerAutoload))\n\t\t\t\t\t.map(async (model) => {\n\t\t\t\t\t\t// Only loaded models expose their template without side effects. Unloaded autoload presets\n\t\t\t\t\t\t// would need to be loaded, while querying sleeping models may wake them. Those models remain\n\t\t\t\t\t\t// unclassified until they are loaded or woken and a later catalog refresh discovers them.\n\t\t\t\t\t\tif (model.status.value !== \"loaded\") return toPiModel(model, serverUrl);\n\t\t\t\t\t\tconst props = await client.props({ model: model.id, signal: context.signal });\n\t\t\t\t\t\treturn toPiModel(model, serverUrl, props);\n\t\t\t\t\t}),\n\t\t\t);\n\t\t\tif (context.signal.aborted) return;\n\t\t\tawait context.publish({\n\t\t\t\tpersist: { models: refreshed, checkedAt: Date.now() },\n\t\t\t\tupdate: () => {\n\t\t\t\t\tmodels = refreshed;\n\t\t\t\t},\n\t\t\t});\n\t\t},\n\t\tstream: (model, context, options) => stream(model, context, options as ProviderStreamOptions | undefined),\n\t\tstreamSimple: (model, context, options) => streamSimple(model, context, options),\n\t};\n\n\treturn { provider, setCatalog };\n}\n"]}
+{"version":3,"file":"provider.js","sourceRoot":"","sources":["../../../src/extensions/llama/provider.ts"],"names":[],"mappings":"AAAA,OAAO,EAKN,WAAW,GAKX,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAAE,mBAAmB,EAAE,MAAM,mDAAmD,CAAC;AACxF,OAAO,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,8BAA8B,CAAC;AACpE,OAAO,EACN,WAAW,EAGX,iBAAiB,EACjB,uBAAuB,GACvB,MAAM,aAAa,CAAC;AAErB,MAAM,CAAC,MAAM,iBAAiB,GAAG,WAAW,CAAC;AAC7C,MAAM,CAAC,MAAM,wBAAwB,GAAG,uBAAuB,CAAC;AAChE,SAAS,mBAAmB,CAAC,UAAwC;IACpE,MAAM,KAAK,GAAG,UAAU,EAAE,GAAG,EAAE,cAAc,CAAC;IAC9C,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,uBAAuB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAC/F,CAAC;AAED,KAAK,UAAU,gBAAgB,CAC9B,GAAgB,EAChB,UAAwC;IAExC,MAAM,UAAU,GAAG,mBAAmB,CAAC,UAAU,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC;IAChG,OAAO,UAAU,CAAC,CAAC,CAAC,uBAAuB,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AACrE,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAqB,EAAE,cAAuB;IACxE,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ;QAAE,OAAO,IAAI,CAAC;IACjD,uFAAuF;IACvF,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU;QAAE,OAAO,IAAI,CAAC;IACnD,gGAAgG;IAChG,OAAO,cAAc,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,IAAI,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC;AACjH,CAAC;AAED,KAAK,UAAU,qBAAqB,CACnC,MAAmB,EACnB,OAAkC,EAClC,MAAmB;IAEnB,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,IAAI,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC;QAAE,OAAO,KAAK,CAAC;IAC3G,IAAI,CAAC;QACJ,OAAO,CAAC,MAAM,MAAM,CAAC,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,eAAe,KAAK,IAAI,CAAC;IAClE,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,KAAK,CAAC;IACd,CAAC;AACF,CAAC;AAED,SAAS,uBAAuB,CAAC,KAAqB;IACrD,MAAM,IAAI,GAAG,KAAK,CAAC,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC;IACrC,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,KAAK,EAAE,EAAE,CAAC;QACtD,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC;QACzB,IAAI,IAAI,KAAK,YAAY,IAAI,IAAI,KAAK,IAAI,IAAI,IAAI,KAAK,MAAM;YAAE,SAAS;QACxE,MAAM,aAAa,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC;QAC9C,IAAI,MAAM,CAAC,aAAa,CAAC,aAAa,CAAC,IAAI,aAAa,GAAG,CAAC;YAAE,OAAO,aAAa,CAAC;IACpF,CAAC;IACD,OAAO,SAAS,CAAC;AAClB,CAAC;AAED,SAAS,eAAe,CAAC,KAAqB,EAAE,mBAA4B;IAC3E,MAAM,oBAAoB,GAAG,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC;IAC/C,IAAI,oBAAoB,IAAI,oBAAoB,GAAG,CAAC;QAAE,OAAO,oBAAoB,CAAC;IAClF,MAAM,iBAAiB,GAAG,uBAAuB,CAAC,KAAK,CAAC,CAAC;IACzD,IAAI,iBAAiB;QAAE,OAAO,iBAAiB,CAAC;IAChD,IAAI,mBAAmB,IAAI,mBAAmB,GAAG,CAAC;QAAE,OAAO,mBAAmB,CAAC;IAC/E,MAAM,qBAAqB,GAAG,KAAK,CAAC,IAAI,EAAE,WAAW,CAAC;IACtD,OAAO,qBAAqB,IAAI,qBAAqB,GAAG,CAAC,CAAC,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC,MAAM,CAAC;AAC5F,CAAC;AAED,2GAA2G;AAC3G,SAAS,mBAAmB,CAC3B,KAAqB,EACrB,SAAiB,EACjB,mBAA4B;IAE5B,OAAO;QACN,IAAI,EAAE,YAAY;QAClB,EAAE,EAAE,KAAK,CAAC,EAAE;QACZ,IAAI,EAAE,KAAK,CAAC,EAAE;QACd,GAAG,EAAE,oBAAoB;QACzB,QAAQ,EAAE,iBAAiB;QAC3B,OAAO,EAAE,SAAS;QAClB,KAAK,EAAE,CAAC,MAAM,CAAC;QACf,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE;QAC1D,aAAa,EAAE,eAAe,CAAC,KAAK,EAAE,mBAAmB,CAAC;KAC1D,CAAC;AACH,CAAC;AAED,SAAS,SAAS,CACjB,KAAqB,EACrB,SAAiB,EACjB,KAAwB,EACxB,mBAA4B;IAE5B,MAAM,aAAa,GAAG,eAAe,CAAC,KAAK,EAAE,mBAAmB,CAAC,CAAC;IAClE,MAAM,SAAS,GAAG,KAAK,EAAE,aAAa,EAAE,QAAQ,CAAC,iBAAiB,CAAC,KAAK,IAAI,CAAC;IAC7E,OAAO;QACN,EAAE,EAAE,KAAK,CAAC,EAAE;QACZ,IAAI,EAAE,KAAK,CAAC,EAAE;QACd,GAAG,EAAE,oBAAoB;QACzB,QAAQ,EAAE,iBAAiB;QAC3B,OAAO,EAAE,iBAAiB,CAAC,SAAS,CAAC;QACrC,SAAS;QACT,GAAG,CAAC,SAAS,IAAI;YAChB,gBAAgB,EAAE,EAAE,GAAG,EAAE,KAAK,EAAE,OAAO,EAAE,IAAI,EAAE,GAAG,EAAE,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE;SACrG,CAAC;QACF,KAAK,EAAE,KAAK,CAAC,YAAY,EAAE,gBAAgB,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC;QAC7F,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE;QAC1D,aAAa;QACb,SAAS,EAAE,aAAa;QACxB,MAAM,EAAE;YACP,aAAa,EAAE,KAAK;YACpB,qBAAqB,EAAE,KAAK;YAC5B,uBAAuB,EAAE,KAAK;YAC9B,wBAAwB,EAAE,IAAI;YAC9B,kBAAkB,EAAE,KAAK;YACzB,cAAc,EAAE,YAAY;YAC5B,GAAG,CAAC,SAAS,IAAI,EAAE,cAAc,EAAE,oBAAoB,EAAE,CAAC;SAC1D;KACD,CAAC;AACH,CAAC;AAOD,MAAM,UAAU,mBAAmB;IAClC,IAAI,MAAM,GAA2C,EAAE,CAAC;IACxD,IAAI,WAAW,GAAqD,EAAE,CAAC;IACvE,MAAM,UAAU,GAAG,mBAAmB,EAAE,CAAC;IAEzC,MAAM,UAAU,GAAG,CAClB,OAAkC,EAClC,SAAiB,EACjB,UAAwC,EAAE,EACnC,EAAE;QACT,MAAM,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,iBAAiB,CAAC,KAAK,EAAE,OAAO,CAAC,cAAc,KAAK,IAAI,CAAC,CAAC,CAAC;QACxG,MAAM,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,SAAS,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC,CAAC;QAChE,WAAW,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,mBAAmB,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC,CAAC;IAChF,CAAC,CAAC;IAEF,MAAM,QAAQ,GAAmC;QAChD,EAAE,EAAE,iBAAiB;QACrB,IAAI,EAAE,WAAW;QACjB,OAAO,EAAE,iBAAiB,CAAC,wBAAwB,CAAC;QACpD,IAAI,EAAE;YACL,MAAM,EAAE;gBACP,IAAI,EAAE,kBAAkB;gBACxB,KAAK,EAAE,KAAK,EAAE,WAAW,EAA6B,EAAE;oBACvD,MAAM,UAAU,GAAG,MAAM,WAAW,CAAC,MAAM,CAAC;wBAC3C,IAAI,EAAE,MAAM;wBACZ,OAAO,EAAE,sBAAsB;wBAC/B,WAAW,EAAE,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,wBAAwB;qBACnE,CAAC,CAAC;oBACH,MAAM,SAAS,GAAG,uBAAuB,CACxC,UAAU,CAAC,IAAI,EAAE,IAAI,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,wBAAwB,CAC3E,CAAC;oBACF,MAAM,MAAM,GAAG,CACd,MAAM,WAAW,CAAC,MAAM,CAAC;wBACxB,IAAI,EAAE,QAAQ;wBACd,OAAO,EAAE,oBAAoB;qBAC7B,CAAC,CACF,CAAC,IAAI,EAAE,CAAC;oBACT,MAAM,IAAI,WAAW,CAAC,SAAS,EAAE,MAAM,IAAI,SAAS,CAAC,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,WAAW,CAAC,MAAM,EAAE,CAAC,CAAC;oBAC3F,OAAO;wBACN,IAAI,EAAE,SAAS;wBACf,GAAG,EAAE,MAAM,IAAI,SAAS;wBACxB,GAAG,EAAE,EAAE,cAAc,EAAE,SAAS,EAAE;qBAClC,CAAC;gBACH,CAAC;gBACD,KAAK,EAAE,KAAK,EAAE,EAAE,GAAG,EAAE,UAAU,EAAE,EAAE,EAAE;oBACpC,MAAM,SAAS,GAAG,MAAM,gBAAgB,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;oBAC1D,OAAO,SAAS;wBACf,CAAC,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,MAAM,EAAE,UAAU,CAAC,CAAC,CAAC,mBAAmB,CAAC,CAAC,CAAC,gBAAgB,EAAE;wBAClF,CAAC,CAAC,SAAS,CAAC;gBACd,CAAC;gBACD,OAAO,EAAE,KAAK,EAAE,EAAE,GAAG,EAAE,UAAU,EAAE,EAAmC,EAAE;oBACvE,MAAM,SAAS,GAAG,MAAM,gBAAgB,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;oBAC1D,IAAI,CAAC,SAAS;wBAAE,OAAO,SAAS,CAAC;oBACjC,MAAM,MAAM,GAAG,UAAU,EAAE,GAAG,IAAI,CAAC,MAAM,GAAG,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC,IAAI,OAAO,CAAC;oBAC9E,OAAO;wBACN,IAAI,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,iBAAiB,CAAC,SAAS,CAAC,EAAE;wBACvD,GAAG,EAAE,EAAE,GAAG,UAAU,EAAE,GAAG,EAAE,cAAc,EAAE,SAAS,EAAE;wBACtD,MAAM,EAAE,UAAU,CAAC,CAAC,CAAC,mBAAmB,CAAC,CAAC,CAAC,gBAAgB;qBAC3D,CAAC;gBACH,CAAC;aACD;SACD;QACD,SAAS,EAAE,GAAG,EAAE,CAAC,MAAM;QACvB,YAAY,EAAE,GAAG,EAAE,CAAC,CAAC,GAAG,MAAM,EAAE,GAAG,WAAW,CAAC;QAC/C,aAAa,EAAE,KAAK,EAAE,OAA6B,EAAiB,EAAE;YACrE,MAAM,oBAAoB,GAAG,IAAI,GAAG,EAAkB,CAAC;YACvD,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;gBACpB,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,iBAAiB,CAAC,CAAC;gBAC7F,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAC7B,CAAC,KAAK,EAAwC,EAAE,CAC/C,WAAW,CAAC,KAAK,EAAE,MAAM,CAAC,IAAI,KAAK,CAAC,GAAG,KAAK,oBAAoB,CACjE,CAAC;gBACF,MAAM,mBAAmB,GAAG,MAAM,CAAC,MAAM,CACxC,CAAC,KAAK,EAAkD,EAAE,CACzD,WAAW,CAAC,KAAK,EAAE,YAAY,CAAC,IAAI,KAAK,CAAC,GAAG,KAAK,oBAAoB,CACvE,CAAC;gBACF,KAAK,MAAM,KAAK,IAAI,CAAC,GAAG,QAAQ,EAAE,GAAG,mBAAmB,CAAC,EAAE,CAAC;oBAC3D,oBAAoB,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,aAAa,CAAC,CAAC;gBACzD,CAAC;gBACD,IACC,CAAC,CAAC,MAAM,OAAO,CAAC,OAAO,CAAC;oBACvB,MAAM,EAAE,GAAG,EAAE;wBACZ,MAAM,GAAG,QAAQ,CAAC;wBAClB,WAAW,GAAG,mBAAmB,CAAC;oBACnC,CAAC;iBACD,CAAC,CAAC,EACF,CAAC;oBACF,OAAO;gBACR,CAAC;YACF,CAAC;YAED,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,IAAI,OAAO,CAAC,UAAU,EAAE,IAAI,KAAK,SAAS;gBAAE,OAAO;YACtG,MAAM,SAAS,GAAG,mBAAmB,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;YAC1D,IAAI,CAAC,SAAS;gBAAE,OAAO;YACvB,MAAM,MAAM,GAAG,IAAI,WAAW,CAAC,SAAS,EAAE,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC;YAClE,MAAM,OAAO,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;YAC9D,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,cAAc,GAAG,MAAM,qBAAqB,CAAC,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;YACpF,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,iBAAiB,CAAC,KAAK,EAAE,cAAc,CAAC,CAAC,CAAC;YACvF,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC,GAAG,CAClC,UAAU,CAAC,GAAG,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE;gBAC9B,MAAM,mBAAmB,GAAG,oBAAoB,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;gBAC/D,2FAA2F;gBAC3F,6FAA6F;gBAC7F,0FAA0F;gBAC1F,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ;oBAAE,OAAO,SAAS,CAAC,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,mBAAmB,CAAC,CAAC;gBACxG,MAAM,KAAK,GAAG,MAAM,MAAM,CAAC,KAAK,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;gBAC9E,OAAO,SAAS,CAAC,KAAK,EAAE,SAAS,EAAE,KAAK,EAAE,mBAAmB,CAAC,CAAC;YAChE,CAAC,CAAC,CACF,CAAC;YACF,MAAM,oBAAoB,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CACrD,mBAAmB,CAAC,KAAK,EAAE,SAAS,EAAE,oBAAoB,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CACzE,CAAC;YACF,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,OAAO,CAAC,OAAO,CAAC;gBACrB,OAAO,EAAE,EAAE,MAAM,EAAE,CAAC,GAAG,SAAS,EAAE,GAAG,oBAAoB,CAAC,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE;gBACnF,MAAM,EAAE,GAAG,EAAE;oBACZ,MAAM,GAAG,SAAS,CAAC;oBACnB,WAAW,GAAG,oBAAoB,CAAC;gBACpC,CAAC;aACD,CAAC,CAAC;QACJ,CAAC;QACD,MAAM,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,OAA4C,CAAC;QACzG,YAAY,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC;QAChF,QAAQ,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC;KACnF,CAAC;IAEF,OAAO,EAAE,QAAQ,EAAE,UAAU,EAAE,CAAC;AACjC,CAAC","sourcesContent":["import {\n\ttype ApiKeyCredential,\n\ttype AuthContext,\n\ttype AuthResult,\n\ttype ClassifierModel,\n\tisModelType,\n\ttype Model,\n\ttype Provider,\n\ttype ProviderStreamOptions,\n\ttype RefreshModelsContext,\n} from \"@earendil-works/pi-ai\";\nimport { llamaCppClassifyApi } from \"@earendil-works/pi-ai/api/llama-cpp-classify.lazy\";\nimport { stream, streamSimple } from \"@earendil-works/pi-ai/compat\";\nimport {\n\tLlamaClient,\n\ttype LlamaModelInfo,\n\ttype LlamaServerProps,\n\tllamaInferenceUrl,\n\tnormalizeLlamaServerUrl,\n} from \"./client.ts\";\n\nexport const LLAMA_PROVIDER_ID = \"llama.cpp\";\nexport const DEFAULT_LLAMA_SERVER_URL = \"http://127.0.0.1:8080\";\nfunction credentialServerUrl(credential: ApiKeyCredential | undefined): string | undefined {\n\tconst value = credential?.env?.LLAMA_BASE_URL;\n\treturn typeof value === \"string\" && value.trim() ? normalizeLlamaServerUrl(value) : undefined;\n}\n\nasync function resolveServerUrl(\n\tctx: AuthContext,\n\tcredential: ApiKeyCredential | undefined,\n): Promise<string | undefined> {\n\tconst configured = credentialServerUrl(credential) ?? (await ctx.env(\"LLAMA_BASE_URL\"))?.trim();\n\treturn configured ? normalizeLlamaServerUrl(configured) : undefined;\n}\n\nfunction modelIsSelectable(model: LlamaModelInfo, routerAutoload: boolean): boolean {\n\tif (model.status.value === \"loaded\") return true;\n\t// llama.cpp reports idle-slept models as \"sleeping\"; requests wake them automatically.\n\tif (model.status.value === \"sleeping\") return true;\n\t// Unloaded presets are routable only when llama.cpp router autoload can load them on first use.\n\treturn routerAutoload && model.status.value === \"unloaded\" && !model.status.failed && model.source === \"preset\";\n}\n\nasync function routerAutoloadEnabled(\n\tclient: LlamaClient,\n\tcatalog: readonly LlamaModelInfo[],\n\tsignal: AbortSignal,\n): Promise<boolean> {\n\tif (!catalog.some((model) => model.status.value === \"unloaded\" && model.source === \"preset\")) return false;\n\ttry {\n\t\treturn (await client.props({ signal })).models_autoload === true;\n\t} catch {\n\t\treturn false;\n\t}\n}\n\nfunction configuredContextWindow(model: LlamaModelInfo): number | undefined {\n\tconst args = model.status.args ?? [];\n\tfor (let index = 0; index < args.length - 1; index++) {\n\t\tconst flag = args[index];\n\t\tif (flag !== \"--ctx-size\" && flag !== \"-c\" && flag !== \"-ctx\") continue;\n\t\tconst contextWindow = Number(args[index + 1]);\n\t\tif (Number.isSafeInteger(contextWindow) && contextWindow > 0) return contextWindow;\n\t}\n\treturn undefined;\n}\n\nfunction contextWindowOf(model: LlamaModelInfo, cachedContextWindow?: number): number {\n\tconst runtimeContextWindow = model.meta?.n_ctx;\n\tif (runtimeContextWindow && runtimeContextWindow > 0) return runtimeContextWindow;\n\tconst configuredContext = configuredContextWindow(model);\n\tif (configuredContext) return configuredContext;\n\tif (cachedContextWindow && cachedContextWindow > 0) return cachedContextWindow;\n\tconst trainingContextWindow = model.meta?.n_ctx_train;\n\treturn trainingContextWindow && trainingContextWindow > 0 ? trainingContextWindow : 128000;\n}\n\n/** The same llama.cpp model used as a classifier: answers are read from next-token label probabilities. */\nfunction toPiClassifierModel(\n\tmodel: LlamaModelInfo,\n\tserverUrl: string,\n\tcachedContextWindow?: number,\n): ClassifierModel<\"llama-cpp-classify\"> {\n\treturn {\n\t\ttype: \"classifier\",\n\t\tid: model.id,\n\t\tname: model.id,\n\t\tapi: \"llama-cpp-classify\",\n\t\tprovider: LLAMA_PROVIDER_ID,\n\t\tbaseUrl: serverUrl,\n\t\tinput: [\"text\"],\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },\n\t\tcontextWindow: contextWindowOf(model, cachedContextWindow),\n\t};\n}\n\nfunction toPiModel(\n\tmodel: LlamaModelInfo,\n\tserverUrl: string,\n\tprops?: LlamaServerProps,\n\tcachedContextWindow?: number,\n): Model<\"openai-completions\"> {\n\tconst contextWindow = contextWindowOf(model, cachedContextWindow);\n\tconst reasoning = props?.chat_template?.includes(\"enable_thinking\") === true;\n\treturn {\n\t\tid: model.id,\n\t\tname: model.id,\n\t\tapi: \"openai-completions\",\n\t\tprovider: LLAMA_PROVIDER_ID,\n\t\tbaseUrl: llamaInferenceUrl(serverUrl),\n\t\treasoning,\n\t\t...(reasoning && {\n\t\t\tthinkingLevelMap: { off: \"off\", minimal: null, low: null, medium: \"medium\", high: null, xhigh: null },\n\t\t}),\n\t\tinput: model.architecture?.input_modalities?.includes(\"image\") ? [\"text\", \"image\"] : [\"text\"],\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },\n\t\tcontextWindow,\n\t\tmaxTokens: contextWindow,\n\t\tcompat: {\n\t\t\tsupportsStore: false,\n\t\t\tsupportsDeveloperRole: false,\n\t\t\tsupportsReasoningEffort: false,\n\t\t\tsupportsUsageInStreaming: true,\n\t\t\tsupportsStrictMode: false,\n\t\t\tmaxTokensField: \"max_tokens\",\n\t\t\t...(reasoning && { thinkingFormat: \"qwen-chat-template\" }),\n\t\t},\n\t};\n}\n\nexport interface LlamaProviderController {\n\tprovider: Provider<\"openai-completions\">;\n\tsetCatalog(models: readonly LlamaModelInfo[], serverUrl: string, options?: { routerAutoload?: boolean }): void;\n}\n\nexport function createLlamaProvider(): LlamaProviderController {\n\tlet models: readonly Model<\"openai-completions\">[] = [];\n\tlet classifiers: readonly ClassifierModel<\"llama-cpp-classify\">[] = [];\n\tconst classifier = llamaCppClassifyApi();\n\n\tconst setCatalog = (\n\t\tcatalog: readonly LlamaModelInfo[],\n\t\tserverUrl: string,\n\t\toptions: { routerAutoload?: boolean } = {},\n\t): void => {\n\t\tconst selectable = catalog.filter((model) => modelIsSelectable(model, options.routerAutoload === true));\n\t\tmodels = selectable.map((model) => toPiModel(model, serverUrl));\n\t\tclassifiers = selectable.map((model) => toPiClassifierModel(model, serverUrl));\n\t};\n\n\tconst provider: Provider<\"openai-completions\"> = {\n\t\tid: LLAMA_PROVIDER_ID,\n\t\tname: \"llama.cpp\",\n\t\tbaseUrl: llamaInferenceUrl(DEFAULT_LLAMA_SERVER_URL),\n\t\tauth: {\n\t\t\tapiKey: {\n\t\t\t\tname: \"llama.cpp server\",\n\t\t\t\tlogin: async (interaction): Promise<ApiKeyCredential> => {\n\t\t\t\t\tconst enteredUrl = await interaction.prompt({\n\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\tmessage: \"llama.cpp server URL\",\n\t\t\t\t\t\tplaceholder: process.env.LLAMA_BASE_URL ?? DEFAULT_LLAMA_SERVER_URL,\n\t\t\t\t\t});\n\t\t\t\t\tconst serverUrl = normalizeLlamaServerUrl(\n\t\t\t\t\t\tenteredUrl.trim() || process.env.LLAMA_BASE_URL || DEFAULT_LLAMA_SERVER_URL,\n\t\t\t\t\t);\n\t\t\t\t\tconst apiKey = (\n\t\t\t\t\t\tawait interaction.prompt({\n\t\t\t\t\t\t\ttype: \"secret\",\n\t\t\t\t\t\t\tmessage: \"API key (optional)\",\n\t\t\t\t\t\t})\n\t\t\t\t\t).trim();\n\t\t\t\t\tawait new LlamaClient(serverUrl, apiKey || undefined).list({ signal: interaction.signal });\n\t\t\t\t\treturn {\n\t\t\t\t\t\ttype: \"api_key\",\n\t\t\t\t\t\tkey: apiKey || undefined,\n\t\t\t\t\t\tenv: { LLAMA_BASE_URL: serverUrl },\n\t\t\t\t\t};\n\t\t\t\t},\n\t\t\t\tcheck: async ({ ctx, credential }) => {\n\t\t\t\t\tconst serverUrl = await resolveServerUrl(ctx, credential);\n\t\t\t\t\treturn serverUrl\n\t\t\t\t\t\t? { type: \"api_key\", source: credential ? \"stored credential\" : \"LLAMA_BASE_URL\" }\n\t\t\t\t\t\t: undefined;\n\t\t\t\t},\n\t\t\t\tresolve: async ({ ctx, credential }): Promise<AuthResult | undefined> => {\n\t\t\t\t\tconst serverUrl = await resolveServerUrl(ctx, credential);\n\t\t\t\t\tif (!serverUrl) return undefined;\n\t\t\t\t\tconst apiKey = credential?.key ?? (await ctx.env(\"LLAMA_API_KEY\")) ?? \"local\";\n\t\t\t\t\treturn {\n\t\t\t\t\t\tauth: { apiKey, baseUrl: llamaInferenceUrl(serverUrl) },\n\t\t\t\t\t\tenv: { ...credential?.env, LLAMA_BASE_URL: serverUrl },\n\t\t\t\t\t\tsource: credential ? \"stored credential\" : \"LLAMA_BASE_URL\",\n\t\t\t\t\t};\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tgetModels: () => models,\n\t\tgetAllModels: () => [...models, ...classifiers],\n\t\trefreshModels: async (context: RefreshModelsContext): Promise<void> => {\n\t\t\tconst cachedContextWindows = new Map<string, number>();\n\t\t\tif (context.stored) {\n\t\t\t\tconst stored = context.stored.models.filter((model) => model.provider === LLAMA_PROVIDER_ID);\n\t\t\t\tconst restored = stored.filter(\n\t\t\t\t\t(model): model is Model<\"openai-completions\"> =>\n\t\t\t\t\t\tisModelType(model, \"chat\") && model.api === \"openai-completions\",\n\t\t\t\t);\n\t\t\t\tconst restoredClassifiers = stored.filter(\n\t\t\t\t\t(model): model is ClassifierModel<\"llama-cpp-classify\"> =>\n\t\t\t\t\t\tisModelType(model, \"classifier\") && model.api === \"llama-cpp-classify\",\n\t\t\t\t);\n\t\t\t\tfor (const model of [...restored, ...restoredClassifiers]) {\n\t\t\t\t\tcachedContextWindows.set(model.id, model.contextWindow);\n\t\t\t\t}\n\t\t\t\tif (\n\t\t\t\t\t!(await context.publish({\n\t\t\t\t\t\tupdate: () => {\n\t\t\t\t\t\t\tmodels = restored;\n\t\t\t\t\t\t\tclassifiers = restoredClassifiers;\n\t\t\t\t\t\t},\n\t\t\t\t\t}))\n\t\t\t\t) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (!context.allowNetwork || context.signal.aborted || context.credential?.type !== \"api_key\") return;\n\t\t\tconst serverUrl = credentialServerUrl(context.credential);\n\t\t\tif (!serverUrl) return;\n\t\t\tconst client = new LlamaClient(serverUrl, context.credential.key);\n\t\t\tconst catalog = await client.list({ signal: context.signal });\n\t\t\tif (context.signal.aborted) return;\n\t\t\tconst routerAutoload = await routerAutoloadEnabled(client, catalog, context.signal);\n\t\t\tif (context.signal.aborted) return;\n\t\t\tconst selectable = catalog.filter((model) => modelIsSelectable(model, routerAutoload));\n\t\t\tconst refreshed = await Promise.all(\n\t\t\t\tselectable.map(async (model) => {\n\t\t\t\t\tconst cachedContextWindow = cachedContextWindows.get(model.id);\n\t\t\t\t\t// Only loaded models expose their template without side effects. Unloaded autoload presets\n\t\t\t\t\t// would need to be loaded, while querying sleeping models may wake them. Those models remain\n\t\t\t\t\t// unclassified until they are loaded or woken and a later catalog refresh discovers them.\n\t\t\t\t\tif (model.status.value !== \"loaded\") return toPiModel(model, serverUrl, undefined, cachedContextWindow);\n\t\t\t\t\tconst props = await client.props({ model: model.id, signal: context.signal });\n\t\t\t\t\treturn toPiModel(model, serverUrl, props, cachedContextWindow);\n\t\t\t\t}),\n\t\t\t);\n\t\t\tconst refreshedClassifiers = selectable.map((model) =>\n\t\t\t\ttoPiClassifierModel(model, serverUrl, cachedContextWindows.get(model.id)),\n\t\t\t);\n\t\t\tif (context.signal.aborted) return;\n\t\t\tawait context.publish({\n\t\t\t\tpersist: { models: [...refreshed, ...refreshedClassifiers], checkedAt: Date.now() },\n\t\t\t\tupdate: () => {\n\t\t\t\t\tmodels = refreshed;\n\t\t\t\t\tclassifiers = refreshedClassifiers;\n\t\t\t\t},\n\t\t\t});\n\t\t},\n\t\tstream: (model, context, options) => stream(model, context, options as ProviderStreamOptions | undefined),\n\t\tstreamSimple: (model, context, options) => streamSimple(model, context, options),\n\t\tclassify: (model, context, options) => classifier.classify(model, context, options),\n\t};\n\n\treturn { provider, setCatalog };\n}\n"]}

package/dist/extensions/mcp/cli.d.ts

@@ -0,0 +1,21 @@
+/**
+ * `pi mcp`: add, remove, and check MCP servers and sign in to them outside a session. Agents run it
+ * through bash to configure servers, verify an `mcp.json` they wrote, and start an OAuth sign-in;
+ * the user only approves access in the browser. Running sessions pick up new credentials on their
+ * next turn.
+ */
+import { McpOAuthCredentialStore } from "./runtime.ts";
+export interface McpCommandOptions {
+    cwd: string;
+    agentDir: string;
+    /** Defaults to `mcp-auth.json` in the agent directory. */
+    credentials?: McpOAuthCredentialStore;
+    /** Defaults to the platform browser. */
+    openUrl?: (url: string) => void;
+    /** Defaults to console output. */
+    log?: (line: string) => void;
+    error?: (line: string) => void;
+}
+/** Run `pi mcp <args>` and return the exit code. */
+export declare function runMcpCommand(args: string[], options: McpCommandOptions): Promise<number>;
+//# sourceMappingURL=cli.d.ts.map

package/dist/extensions/mcp/cli.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"cli.d.ts","sourceRoot":"","sources":["../../../src/extensions/mcp/cli.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAkBH,OAAO,EAEN,uBAAuB,EAKvB,MAAM,cAAc,CAAC;AA8CtB,MAAM,WAAW,iBAAiB;IACjC,GAAG,EAAE,MAAM,CAAC;IACZ,QAAQ,EAAE,MAAM,CAAC;IACjB,0DAA0D;IAC1D,WAAW,CAAC,EAAE,uBAAuB,CAAC;IACtC,wCAAwC;IACxC,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,MAAM,KAAK,IAAI,CAAC;IAChC,kCAAkC;IAClC,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAC;IAC7B,KAAK,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAC;CAC/B;AA6FD,oDAAoD;AACpD,wBAAsB,aAAa,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,OAAO,EAAE,iBAAiB,GAAG,OAAO,CAAC,MAAM,CAAC,CA2E/F"}