@aibyzero/byz 0.0.0 → 0.1.1

package/dist/runtime/extensions/llama/huggingface.js

@@ -0,0 +1,141 @@
+import { readFile } from "node:fs/promises";
+import { homedir } from "node:os";
+import { join } from "node:path";
+const DEFAULT_HUGGING_FACE_URL = "https://huggingface.co";
+const QUANTIZATION_PATTERN = /(?:^|[-_.])((?:UD-)?(?:IQ\d(?:_[A-Z0-9]+)+|Q\d(?:_[A-Z0-9]+)+|BF16|F16|F32|MXFP\d(?:_[A-Z0-9]+)*))$/iu;
+const SHARD_SUFFIX_PATTERN = /-\d{5}-of-\d{5}$/u;
+function payloadError(payload, fallback) {
+    if (typeof payload !== "object" || payload === null)
+        return fallback;
+    const error = payload.error;
+    return typeof error === "string" && error ? error : fallback;
+}
+function parseRateLimitDelay(value) {
+    const match = value?.match(/(?:^|;)t=(\d+)/u);
+    return match ? Number(match[1]) : undefined;
+}
+async function readToken(path) {
+    try {
+        const token = (await readFile(path, "utf8")).trim();
+        return token || undefined;
+    }
+    catch {
+        return undefined;
+    }
+}
+export async function findHuggingFaceToken(env = process.env) {
+    const fromEnvironment = env.HF_TOKEN?.trim();
+    if (fromEnvironment)
+        return fromEnvironment;
+    const paths = [
+        env.HF_TOKEN_PATH,
+        env.HF_HOME ? join(env.HF_HOME, "token") : undefined,
+        env.XDG_CACHE_HOME ? join(env.XDG_CACHE_HOME, "huggingface", "token") : undefined,
+        join(homedir(), ".cache", "huggingface", "token"),
+    ].filter((path) => Boolean(path));
+    for (const path of new Set(paths)) {
+        const token = await readToken(path);
+        if (token)
+            return token;
+    }
+    return undefined;
+}
+export class HuggingFaceClient {
+    token;
+    baseUrl;
+    constructor(token, baseUrl = DEFAULT_HUGGING_FACE_URL) {
+        this.token = token;
+        this.baseUrl = baseUrl.replace(/\/+$/u, "");
+    }
+    async request(path, signal) {
+        const headers = new Headers();
+        if (this.token)
+            headers.set("Authorization", `Bearer ${this.token}`);
+        const timeout = AbortSignal.timeout(15_000);
+        const response = await fetch(`${this.baseUrl}${path}`, {
+            headers,
+            signal: signal ? AbortSignal.any([signal, timeout]) : timeout,
+        });
+        let payload;
+        try {
+            payload = await response.json();
+        }
+        catch {
+            payload = undefined;
+        }
+        if (!response.ok) {
+            const fallback = `Hugging Face returned HTTP ${response.status}`;
+            if (response.status === 429) {
+                const delay = Number(response.headers.get("retry-after")) || parseRateLimitDelay(response.headers.get("ratelimit"));
+                throw new Error(delay ? `Hugging Face rate limit reached; retry in ${delay}s` : "Hugging Face rate limit reached");
+            }
+            throw new Error(payloadError(payload, fallback));
+        }
+        return payload;
+    }
+    async search(query, signal) {
+        const params = new URLSearchParams({
+            search: query,
+            filter: "gguf",
+            sort: "downloads",
+            direction: "-1",
+            limit: "20",
+        });
+        const payload = await this.request(`/api/models?${params}`, signal);
+        if (!Array.isArray(payload))
+            throw new Error("Hugging Face returned invalid search results");
+        return payload.flatMap((value) => {
+            if (typeof value !== "object" || value === null || typeof value.id !== "string")
+                return [];
+            const model = value;
+            return [{ id: model.id, downloads: typeof model.downloads === "number" ? model.downloads : 0 }];
+        });
+    }
+    async details(id, signal) {
+        const encodedId = id.split("/").map(encodeURIComponent).join("/");
+        const payload = await this.request(`/api/models/${encodedId}?blobs=true`, signal);
+        if (typeof payload !== "object" || payload === null) {
+            throw new Error("Hugging Face returned invalid model details");
+        }
+        const model = payload;
+        const sizes = new Map();
+        if (Array.isArray(model.siblings)) {
+            for (const value of model.siblings) {
+                if (typeof value !== "object" || value === null)
+                    continue;
+                const file = value;
+                if (typeof file.rfilename !== "string" || !file.rfilename.toLowerCase().endsWith(".gguf"))
+                    continue;
+                const filename = file.rfilename.split("/").at(-1);
+                if (filename.toLowerCase().startsWith("mmproj"))
+                    continue;
+                const stem = filename.slice(0, -5).replace(SHARD_SUFFIX_PATTERN, "");
+                const quantization = stem.match(QUANTIZATION_PATTERN)?.[1]?.toUpperCase();
+                if (!quantization)
+                    continue;
+                const current = sizes.get(quantization) ?? { total: 0, complete: true };
+                if (typeof file.size === "number")
+                    current.total += file.size;
+                else
+                    current.complete = false;
+                sizes.set(quantization, current);
+            }
+        }
+        const quantizations = [...sizes]
+            .map(([name, size]) => ({ name, size: size.complete ? size.total : undefined }))
+            .sort((left, right) => {
+            if (left.name === "Q4_K_M")
+                return -1;
+            if (right.name === "Q4_K_M")
+                return 1;
+            return ((left.size ?? Number.MAX_SAFE_INTEGER) - (right.size ?? Number.MAX_SAFE_INTEGER) ||
+                left.name.localeCompare(right.name));
+        });
+        return {
+            id: typeof model.id === "string" ? model.id : id,
+            gated: model.gated === "auto" || model.gated === "manual" ? model.gated : false,
+            quantizations,
+        };
+    }
+}
+//# sourceMappingURL=huggingface.js.map

package/dist/runtime/extensions/llama/huggingface.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"huggingface.js","sourceRoot":"","sources":["../../../src/extensions/llama/huggingface.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC,MAAM,wBAAwB,GAAG,wBAAwB,CAAC;AAC1D,MAAM,oBAAoB,GACzB,uGAAuG,CAAC;AACzG,MAAM,oBAAoB,GAAG,mBAAmB,CAAC;AAkBjD,SAAS,YAAY,CAAC,OAAgB,EAAE,QAAgB,EAAU;IACjE,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI;QAAE,OAAO,QAAQ,CAAC;IACrE,MAAM,KAAK,GAAI,OAA+B,CAAC,KAAK,CAAC;IACrD,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC;AAAA,CAC7D;AAED,SAAS,mBAAmB,CAAC,KAAoB,EAAsB;IACtE,MAAM,KAAK,GAAG,KAAK,EAAE,KAAK,CAAC,iBAAiB,CAAC,CAAC;IAC9C,OAAO,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAC5C;AAED,KAAK,UAAU,SAAS,CAAC,IAAY,EAA+B;IACnE,IAAI,CAAC;QACJ,MAAM,KAAK,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QACpD,OAAO,KAAK,IAAI,SAAS,CAAC;IAC3B,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,SAAS,CAAC;IAClB,CAAC;AAAA,CACD;AAED,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,GAAG,GAAsB,OAAO,CAAC,GAAG,EAA+B;IAC7G,MAAM,eAAe,GAAG,GAAG,CAAC,QAAQ,EAAE,IAAI,EAAE,CAAC;IAC7C,IAAI,eAAe;QAAE,OAAO,eAAe,CAAC;IAE5C,MAAM,KAAK,GAAG;QACb,GAAG,CAAC,aAAa;QACjB,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS;QACpD,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,cAAc,EAAE,aAAa,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS;QACjF,IAAI,CAAC,OAAO,EAAE,EAAE,QAAQ,EAAE,aAAa,EAAE,OAAO,CAAC;KACjD,CAAC,MAAM,CAAC,CAAC,IAAI,EAAkB,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC;IAClD,KAAK,MAAM,IAAI,IAAI,IAAI,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;QACnC,MAAM,KAAK,GAAG,MAAM,SAAS,CAAC,IAAI,CAAC,CAAC;QACpC,IAAI,KAAK;YAAE,OAAO,KAAK,CAAC;IACzB,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AAED,MAAM,OAAO,iBAAiB;IACZ,KAAK,CAAqB;IAC1B,OAAO,CAAS;IAEjC,YAAY,KAAc,EAAE,OAAO,GAAG,wBAAwB,EAAE;QAC/D,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;IAAA,CAC5C;IAEO,KAAK,CAAC,OAAO,CAAC,IAAY,EAAE,MAAoB,EAAoB;QAC3E,MAAM,OAAO,GAAG,IAAI,OAAO,EAAE,CAAC;QAC9B,IAAI,IAAI,CAAC,KAAK;YAAE,OAAO,CAAC,GAAG,CAAC,eAAe,EAAE,UAAU,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;QACrE,MAAM,OAAO,GAAG,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC5C,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,OAAO,GAAG,IAAI,EAAE,EAAE;YACtD,OAAO;YACP,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO;SAC7D,CAAC,CAAC;QACH,IAAI,OAAgB,CAAC;QACrB,IAAI,CAAC;YACJ,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QACjC,CAAC;QAAC,MAAM,CAAC;YACR,OAAO,GAAG,SAAS,CAAC;QACrB,CAAC;QACD,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;YAClB,MAAM,QAAQ,GAAG,8BAA8B,QAAQ,CAAC,MAAM,EAAE,CAAC;YACjE,IAAI,QAAQ,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;gBAC7B,MAAM,KAAK,GACV,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC,IAAI,mBAAmB,CAAC,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC;gBACvG,MAAM,IAAI,KAAK,CACd,KAAK,CAAC,CAAC,CAAC,6CAA6C,KAAK,GAAG,CAAC,CAAC,CAAC,iCAAiC,CACjG,CAAC;YACH,CAAC;YACD,MAAM,IAAI,KAAK,CAAC,YAAY,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC;QAClD,CAAC;QACD,OAAO,OAAO,CAAC;IAAA,CACf;IAED,KAAK,CAAC,MAAM,CAAC,KAAa,EAAE,MAAoB,EAA+B;QAC9E,MAAM,MAAM,GAAG,IAAI,eAAe,CAAC;YAClC,MAAM,EAAE,KAAK;YACb,MAAM,EAAE,MAAM;YACd,IAAI,EAAE,WAAW;YACjB,SAAS,EAAE,IAAI;YACf,KAAK,EAAE,IAAI;SACX,CAAC,CAAC;QACH,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,eAAe,MAAM,EAAE,EAAE,MAAM,CAAC,CAAC;QACpE,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,8CAA8C,CAAC,CAAC;QAC7F,OAAO,OAAO,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;YACjC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,OAAQ,KAA0B,CAAC,EAAE,KAAK,QAAQ;gBACpG,OAAO,EAAE,CAAC;YACX,MAAM,KAAK,GAAG,KAA4C,CAAC;YAC3D,OAAO,CAAC,EAAE,EAAE,EAAE,KAAK,CAAC,EAAE,EAAE,SAAS,EAAE,OAAO,KAAK,CAAC,SAAS,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QAAA,CAChG,CAAC,CAAC;IAAA,CACH;IAED,KAAK,CAAC,OAAO,CAAC,EAAU,EAAE,MAAoB,EAAoC;QACjF,MAAM,SAAS,GAAG,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,kBAAkB,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAClE,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,eAAe,SAAS,aAAa,EAAE,MAAM,CAAC,CAAC;QAClF,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI,EAAE,CAAC;YACrD,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;QAChE,CAAC;QACD,MAAM,KAAK,GAAG,OAAgE,CAAC;QAC/E,MAAM,KAAK,GAAG,IAAI,GAAG,EAAgD,CAAC;QACtE,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC;YACnC,KAAK,MAAM,KAAK,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;gBACpC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI;oBAAE,SAAS;gBAC1D,MAAM,IAAI,GAAG,KAAgD,CAAC;gBAC9D,IAAI,OAAO,IAAI,CAAC,SAAS,KAAK,QAAQ,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC;oBAAE,SAAS;gBACpG,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAE,CAAC;gBACnD,IAAI,QAAQ,CAAC,WAAW,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC;oBAAE,SAAS;gBAC1D,MAAM,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,oBAAoB,EAAE,EAAE,CAAC,CAAC;gBACrE,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,oBAAoB,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC;gBAC1E,IAAI,CAAC,YAAY;oBAAE,SAAS;gBAC5B,MAAM,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,YAAY,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;gBACxE,IAAI,OAAO,IAAI,CAAC,IAAI,KAAK,QAAQ;oBAAE,OAAO,CAAC,KAAK,IAAI,IAAI,CAAC,IAAI,CAAC;;oBACzD,OAAO,CAAC,QAAQ,GAAG,KAAK,CAAC;gBAC9B,KAAK,CAAC,GAAG,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC;YAClC,CAAC;QACF,CAAC;QACD,MAAM,aAAa,GAAG,CAAC,GAAG,KAAK,CAAC;aAC9B,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC;aAC/E,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;YACtB,IAAI,IAAI,CAAC,IAAI,KAAK,QAAQ;gBAAE,OAAO,CAAC,CAAC,CAAC;YACtC,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ;gBAAE,OAAO,CAAC,CAAC;YACtC,OAAO,CACN,CAAC,IAAI,CAAC,IAAI,IAAI,MAAM,CAAC,gBAAgB,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,IAAI,MAAM,CAAC,gBAAgB,CAAC;gBAChF,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CACnC,CAAC;QAAA,CACF,CAAC,CAAC;QACJ,OAAO;YACN,EAAE,EAAE,OAAO,KAAK,CAAC,EAAE,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE;YAChD,KAAK,EAAE,KAAK,CAAC,KAAK,KAAK,MAAM,IAAI,KAAK,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK;YAC/E,aAAa;SACb,CAAC;IAAA,CACF;CACD","sourcesContent":["import { readFile } from \"node:fs/promises\";\nimport { homedir } from \"node:os\";\nimport { join } from \"node:path\";\n\nconst DEFAULT_HUGGING_FACE_URL = \"https://huggingface.co\";\nconst QUANTIZATION_PATTERN =\n\t/(?:^|[-_.])((?:UD-)?(?:IQ\\d(?:_[A-Z0-9]+)+|Q\\d(?:_[A-Z0-9]+)+|BF16|F16|F32|MXFP\\d(?:_[A-Z0-9]+)*))$/iu;\nconst SHARD_SUFFIX_PATTERN = /-\\d{5}-of-\\d{5}$/u;\n\nexport interface HuggingFaceModel {\n\tid: string;\n\tdownloads: number;\n}\n\nexport interface HuggingFaceQuantization {\n\tname: string;\n\tsize?: number;\n}\n\nexport interface HuggingFaceModelDetails {\n\tid: string;\n\tgated: false | \"auto\" | \"manual\";\n\tquantizations: HuggingFaceQuantization[];\n}\n\nfunction payloadError(payload: unknown, fallback: string): string {\n\tif (typeof payload !== \"object\" || payload === null) return fallback;\n\tconst error = (payload as { error?: unknown }).error;\n\treturn typeof error === \"string\" && error ? error : fallback;\n}\n\nfunction parseRateLimitDelay(value: string | null): number | undefined {\n\tconst match = value?.match(/(?:^|;)t=(\\d+)/u);\n\treturn match ? Number(match[1]) : undefined;\n}\n\nasync function readToken(path: string): Promise<string | undefined> {\n\ttry {\n\t\tconst token = (await readFile(path, \"utf8\")).trim();\n\t\treturn token || undefined;\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nexport async function findHuggingFaceToken(env: NodeJS.ProcessEnv = process.env): Promise<string | undefined> {\n\tconst fromEnvironment = env.HF_TOKEN?.trim();\n\tif (fromEnvironment) return fromEnvironment;\n\n\tconst paths = [\n\t\tenv.HF_TOKEN_PATH,\n\t\tenv.HF_HOME ? join(env.HF_HOME, \"token\") : undefined,\n\t\tenv.XDG_CACHE_HOME ? join(env.XDG_CACHE_HOME, \"huggingface\", \"token\") : undefined,\n\t\tjoin(homedir(), \".cache\", \"huggingface\", \"token\"),\n\t].filter((path): path is string => Boolean(path));\n\tfor (const path of new Set(paths)) {\n\t\tconst token = await readToken(path);\n\t\tif (token) return token;\n\t}\n\treturn undefined;\n}\n\nexport class HuggingFaceClient {\n\tprivate readonly token: string | undefined;\n\tprivate readonly baseUrl: string;\n\n\tconstructor(token?: string, baseUrl = DEFAULT_HUGGING_FACE_URL) {\n\t\tthis.token = token;\n\t\tthis.baseUrl = baseUrl.replace(/\\/+$/u, \"\");\n\t}\n\n\tprivate async request(path: string, signal?: AbortSignal): Promise<unknown> {\n\t\tconst headers = new Headers();\n\t\tif (this.token) headers.set(\"Authorization\", `Bearer ${this.token}`);\n\t\tconst timeout = AbortSignal.timeout(15_000);\n\t\tconst response = await fetch(`${this.baseUrl}${path}`, {\n\t\t\theaders,\n\t\t\tsignal: signal ? AbortSignal.any([signal, timeout]) : timeout,\n\t\t});\n\t\tlet payload: unknown;\n\t\ttry {\n\t\t\tpayload = await response.json();\n\t\t} catch {\n\t\t\tpayload = undefined;\n\t\t}\n\t\tif (!response.ok) {\n\t\t\tconst fallback = `Hugging Face returned HTTP ${response.status}`;\n\t\t\tif (response.status === 429) {\n\t\t\t\tconst delay =\n\t\t\t\t\tNumber(response.headers.get(\"retry-after\")) || parseRateLimitDelay(response.headers.get(\"ratelimit\"));\n\t\t\t\tthrow new Error(\n\t\t\t\t\tdelay ? `Hugging Face rate limit reached; retry in ${delay}s` : \"Hugging Face rate limit reached\",\n\t\t\t\t);\n\t\t\t}\n\t\t\tthrow new Error(payloadError(payload, fallback));\n\t\t}\n\t\treturn payload;\n\t}\n\n\tasync search(query: string, signal?: AbortSignal): Promise<HuggingFaceModel[]> {\n\t\tconst params = new URLSearchParams({\n\t\t\tsearch: query,\n\t\t\tfilter: \"gguf\",\n\t\t\tsort: \"downloads\",\n\t\t\tdirection: \"-1\",\n\t\t\tlimit: \"20\",\n\t\t});\n\t\tconst payload = await this.request(`/api/models?${params}`, signal);\n\t\tif (!Array.isArray(payload)) throw new Error(\"Hugging Face returned invalid search results\");\n\t\treturn payload.flatMap((value) => {\n\t\t\tif (typeof value !== \"object\" || value === null || typeof (value as { id?: unknown }).id !== \"string\")\n\t\t\t\treturn [];\n\t\t\tconst model = value as { id: string; downloads?: unknown };\n\t\t\treturn [{ id: model.id, downloads: typeof model.downloads === \"number\" ? model.downloads : 0 }];\n\t\t});\n\t}\n\n\tasync details(id: string, signal?: AbortSignal): Promise<HuggingFaceModelDetails> {\n\t\tconst encodedId = id.split(\"/\").map(encodeURIComponent).join(\"/\");\n\t\tconst payload = await this.request(`/api/models/${encodedId}?blobs=true`, signal);\n\t\tif (typeof payload !== \"object\" || payload === null) {\n\t\t\tthrow new Error(\"Hugging Face returned invalid model details\");\n\t\t}\n\t\tconst model = payload as { id?: unknown; gated?: unknown; siblings?: unknown };\n\t\tconst sizes = new Map<string, { total: number; complete: boolean }>();\n\t\tif (Array.isArray(model.siblings)) {\n\t\t\tfor (const value of model.siblings) {\n\t\t\t\tif (typeof value !== \"object\" || value === null) continue;\n\t\t\t\tconst file = value as { rfilename?: unknown; size?: unknown };\n\t\t\t\tif (typeof file.rfilename !== \"string\" || !file.rfilename.toLowerCase().endsWith(\".gguf\")) continue;\n\t\t\t\tconst filename = file.rfilename.split(\"/\").at(-1)!;\n\t\t\t\tif (filename.toLowerCase().startsWith(\"mmproj\")) continue;\n\t\t\t\tconst stem = filename.slice(0, -5).replace(SHARD_SUFFIX_PATTERN, \"\");\n\t\t\t\tconst quantization = stem.match(QUANTIZATION_PATTERN)?.[1]?.toUpperCase();\n\t\t\t\tif (!quantization) continue;\n\t\t\t\tconst current = sizes.get(quantization) ?? { total: 0, complete: true };\n\t\t\t\tif (typeof file.size === \"number\") current.total += file.size;\n\t\t\t\telse current.complete = false;\n\t\t\t\tsizes.set(quantization, current);\n\t\t\t}\n\t\t}\n\t\tconst quantizations = [...sizes]\n\t\t\t.map(([name, size]) => ({ name, size: size.complete ? size.total : undefined }))\n\t\t\t.sort((left, right) => {\n\t\t\t\tif (left.name === \"Q4_K_M\") return -1;\n\t\t\t\tif (right.name === \"Q4_K_M\") return 1;\n\t\t\t\treturn (\n\t\t\t\t\t(left.size ?? Number.MAX_SAFE_INTEGER) - (right.size ?? Number.MAX_SAFE_INTEGER) ||\n\t\t\t\t\tleft.name.localeCompare(right.name)\n\t\t\t\t);\n\t\t\t});\n\t\treturn {\n\t\t\tid: typeof model.id === \"string\" ? model.id : id,\n\t\t\tgated: model.gated === \"auto\" || model.gated === \"manual\" ? model.gated : false,\n\t\t\tquantizations,\n\t\t};\n\t}\n}\n"]}

package/dist/runtime/extensions/llama/index.d.ts

@@ -0,0 +1,3 @@
+import type { ExtensionAPI } from "../../core/extensions/types.ts";
+export default function llamaExtension(pi: ExtensionAPI): void;
+//# sourceMappingURL=index.d.ts.map

package/dist/runtime/extensions/llama/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAA2B,MAAM,gCAAgC,CAAC;AAyC5F,MAAM,CAAC,OAAO,UAAU,cAAc,CAAC,EAAE,EAAE,YAAY,GAAG,IAAI,CA4L7D","sourcesContent":["import type { ExtensionAPI, ExtensionCommandContext } from \"../../core/extensions/types.ts\";\nimport { formatBytes, LlamaClient, type LlamaModelInfo, normalizeLlamaServerUrl } from \"./client.ts\";\nimport { findHuggingFaceToken, HuggingFaceClient } from \"./huggingface.ts\";\nimport { createLlamaProvider, LLAMA_PROVIDER_ID } from \"./provider.ts\";\nimport { type LlamaUi, runWithProgress, showLlamaUi } from \"./ui.ts\";\n\nfunction modelIsLoaded(model: LlamaModelInfo): boolean {\n\treturn model.status.value === \"loaded\" || model.status.value === \"sleeping\";\n}\n\nfunction isConnectionError(error: unknown): boolean {\n\tif (!(error instanceof Error)) return false;\n\tconst message = `${error.name} ${error.message}`.toLowerCase();\n\treturn message.includes(\"fetch failed\") || message.includes(\"timeout\") || message.includes(\"network\");\n}\n\nfunction connectionErrorMessage(error: unknown): string {\n\tif (isConnectionError(error)) return \"Could not connect to the server.\";\n\treturn error instanceof Error ? error.message : String(error);\n}\n\nfunction parseHuggingFaceModel(value: string): { repository: string; quantization?: string } {\n\tconst colon = value.indexOf(\":\", value.indexOf(\"/\") + 1);\n\treturn colon < 0\n\t\t? { repository: value }\n\t\t: { repository: value.slice(0, colon), quantization: value.slice(colon + 1) };\n}\n\nasync function configuredClient(ctx: ExtensionCommandContext): Promise<LlamaClient | undefined> {\n\tconst result = await ctx.modelRegistry.getProviderAuth(LLAMA_PROVIDER_ID);\n\tif (!result) {\n\t\tctx.ui.notify(`Configure llama.cpp with /login ${LLAMA_PROVIDER_ID}`, \"warning\");\n\t\treturn undefined;\n\t}\n\tconst configuredUrl = result.env?.LLAMA_BASE_URL;\n\tconst serverUrl = normalizeLlamaServerUrl(\n\t\ttypeof configuredUrl === \"string\" && configuredUrl ? configuredUrl : (result.auth.baseUrl ?? \"\"),\n\t);\n\treturn new LlamaClient(serverUrl, result.auth.apiKey);\n}\n\nexport default function llamaExtension(pi: ExtensionAPI): void {\n\tconst provider = createLlamaProvider();\n\tpi.registerProvider(provider.provider);\n\n\tconst syncCatalog = async (\n\t\tctx: ExtensionCommandContext,\n\t\tclient: LlamaClient,\n\t\tcatalog?: LlamaModelInfo[],\n\t): Promise<LlamaModelInfo[]> => {\n\t\tconst signal = AbortSignal.timeout(15_000);\n\t\tconst current = catalog ?? (await client.list({ signal }));\n\t\tprovider.setCatalog(current, client.serverUrl);\n\t\tconst result = await ctx.modelRegistry.refresh({\n\t\t\tproviders: [LLAMA_PROVIDER_ID],\n\t\t\t// /llama already contacted the configured llama.cpp server, so keep this refresh live even in PI_OFFLINE.\n\t\t\tallowNetwork: true,\n\t\t\tsignal,\n\t\t});\n\t\tif (result.aborted) throw new Error(\"Model catalog refresh timed out.\");\n\t\tconst refreshError = result.errors.get(LLAMA_PROVIDER_ID);\n\t\tif (refreshError) throw refreshError;\n\t\treturn current;\n\t};\n\n\tconst loadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tcatalog: LlamaModelInfo[],\n\t\ttarget: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tconst loaded = catalog.filter((model) => model.id !== target.id && modelIsLoaded(model));\n\t\tlet replace = false;\n\t\tif (loaded.length > 0) {\n\t\t\tconst choice = await ui.select(`${loaded.length} model${loaded.length === 1 ? \" is\" : \"s are\"} loaded`, [\n\t\t\t\t\"Unload all and load\",\n\t\t\t\t\"Keep loaded and load\",\n\t\t\t\t\"Cancel\",\n\t\t\t]);\n\t\t\tif (!choice || choice === \"Cancel\") return;\n\t\t\treplace = choice === \"Unload all and load\";\n\t\t}\n\n\t\tconst restoreLoaded = async (): Promise<void> => {\n\t\t\tctx.ui.notify(\"Restoring previously loaded models\");\n\t\t\tfor (const model of loaded) await client.loadAndWait(model.id, () => {});\n\t\t\tawait syncCatalog(ctx, client);\n\t\t};\n\t\tif (replace) {\n\t\t\tfor (const model of loaded) await client.unloadAndWait(model.id);\n\t\t}\n\n\t\ttry {\n\t\t\tconst result = await runWithProgress(ui, {\n\t\t\t\ttitle: \"Loading model\",\n\t\t\t\tmodel: target.id,\n\t\t\t\tinitialMessage: \"Starting…\",\n\t\t\t\tcancelTitle: \"Stop loading?\",\n\t\t\t\tcancelMessage: target.id,\n\t\t\t\trun: (signal, update) => client.loadAndWait(target.id, update, signal),\n\t\t\t\tcancel: () => client.unload(target.id),\n\t\t\t});\n\t\t\tif (result.cancelled) {\n\t\t\t\tif (replace) await restoreLoaded();\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst refreshed = await syncCatalog(ctx, client);\n\t\t\tconst loadedModel = refreshed.find((model) => model.id === target.id);\n\t\t\tctx.ui.notify(\n\t\t\t\tloadedModel?.status.value === \"loaded\" ? `Loaded ${target.id}` : `Load started for ${target.id}`,\n\t\t\t);\n\t\t} catch (error) {\n\t\t\tif (replace) {\n\t\t\t\ttry {\n\t\t\t\t\tawait restoreLoaded();\n\t\t\t\t} catch {\n\t\t\t\t\t// Preserve the original load error.\n\t\t\t\t}\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t};\n\n\tconst unloadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tmodel: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tif (!(await ui.confirm(\"Unload model?\", model.id))) return;\n\t\tawait client.unloadAndWait(model.id);\n\t\tawait syncCatalog(ctx, client);\n\t\tctx.ui.notify(`Unloaded ${model.id}`);\n\t};\n\n\tconst downloadModel = async (ctx: ExtensionCommandContext, ui: LlamaUi, client: LlamaClient): Promise<void> => {\n\t\tconst huggingFace = new HuggingFaceClient(await findHuggingFaceToken());\n\t\tconst selected = await ui.searchModels((query, signal) => huggingFace.search(query, signal));\n\t\tif (!selected) return;\n\t\tconst parsed = parseHuggingFaceModel(selected);\n\t\tui.showStatus(\"Loading model details\", parsed.repository);\n\t\tconst details = await huggingFace.details(parsed.repository);\n\t\tif (details.gated) {\n\t\t\tconst approval = details.gated === \"manual\" ? \"Manual approval is required\" : \"Accept the access terms\";\n\t\t\tconst choice = await ui.select(\n\t\t\t\t`Hugging Face access required\\n${details.id}\\n\\n${approval} at:\\nhttps://huggingface.co/${details.id}\\n\\nThe llama.cpp server needs HF_TOKEN with access.`,\n\t\t\t\t[\"Continue\", \"Back\"],\n\t\t\t);\n\t\t\tif (choice !== \"Continue\") return;\n\t\t}\n\t\tlet quantization = parsed.quantization;\n\t\tif (!quantization && details.quantizations.length > 0) {\n\t\t\tconst options = details.quantizations.map((entry) => {\n\t\t\t\tconst detail = [\n\t\t\t\t\tentry.size === undefined ? undefined : formatBytes(entry.size),\n\t\t\t\t\tentry.name === \"Q4_K_M\" ? \"recommended\" : undefined,\n\t\t\t\t]\n\t\t\t\t\t.filter((value): value is string => Boolean(value))\n\t\t\t\t\t.join(\" · \");\n\t\t\t\treturn detail ? `${entry.name} · ${detail}` : entry.name;\n\t\t\t});\n\t\t\tconst choice = await ui.select(`Select quantization\\n${details.id}`, options);\n\t\t\tif (!choice) return;\n\t\t\tquantization = details.quantizations[options.indexOf(choice)]?.name;\n\t\t\tif (!quantization) return;\n\t\t}\n\t\tconst model = quantization ? `${details.id}:${quantization}` : details.id;\n\t\tconst result = await runWithProgress(ui, {\n\t\t\ttitle: \"Downloading model\",\n\t\t\tmodel,\n\t\t\tinitialMessage: \"Starting…\",\n\t\t\tcancelTitle: \"Stop download?\",\n\t\t\tcancelMessage: model,\n\t\t\trun: (signal, update) => client.downloadAndWait(model, update, signal),\n\t\t\tcancel: () => client.unload(model),\n\t\t});\n\t\tif (result.cancelled) return;\n\t\tawait syncCatalog(ctx, client, result.value);\n\t\tctx.ui.notify(`Downloaded ${model}`);\n\t};\n\n\tpi.registerCommand(\"llama\", {\n\t\tdescription: \"Manage llama.cpp router models\",\n\t\thandler: async (_args, ctx) => {\n\t\t\tif (ctx.mode !== \"tui\") {\n\t\t\t\tctx.ui.notify(\"/llama is available in interactive mode\", \"warning\");\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst client = await configuredClient(ctx);\n\t\t\tif (!client) return;\n\t\t\tawait showLlamaUi(ctx, async (ui) => {\n\t\t\t\tconst readCatalog = async (): Promise<LlamaModelInfo[] | undefined> => {\n\t\t\t\t\twhile (true) {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\treturn await syncCatalog(ctx, client);\n\t\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\t\tif ((await ui.connectionError(client.serverUrl, connectionErrorMessage(error))) === \"close\") {\n\t\t\t\t\t\t\t\treturn undefined;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t};\n\n\t\t\t\tlet catalog = await readCatalog();\n\t\t\t\tif (!catalog) return;\n\t\t\t\twhile (true) {\n\t\t\t\t\tconst action = await ui.showModels(client.serverUrl, catalog);\n\t\t\t\t\tif (action.type === \"close\") return;\n\t\t\t\t\tlet actionError: unknown;\n\t\t\t\t\ttry {\n\t\t\t\t\t\tif (action.type === \"download\") await downloadModel(ctx, ui, client);\n\t\t\t\t\t\telse if (modelIsLoaded(action.model)) await unloadModel(ctx, ui, client, action.model);\n\t\t\t\t\t\telse if (action.model.status.value === \"unloaded\")\n\t\t\t\t\t\t\tawait loadModel(ctx, ui, client, catalog, action.model);\n\t\t\t\t\t\telse ctx.ui.notify(`${action.model.id} is ${action.model.status.value}`, \"warning\");\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tactionError = error;\n\t\t\t\t\t}\n\t\t\t\t\tconst refreshed = await readCatalog();\n\t\t\t\t\tif (!refreshed) return;\n\t\t\t\t\tcatalog = refreshed;\n\t\t\t\t\tif (actionError && !isConnectionError(actionError)) {\n\t\t\t\t\t\tctx.ui.notify(actionError instanceof Error ? actionError.message : String(actionError), \"error\");\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t});\n\t\t},\n\t});\n}\n"]}

package/dist/runtime/extensions/llama/index.js

@@ -0,0 +1,219 @@
+import { formatBytes, LlamaClient, normalizeLlamaServerUrl } from "./client.js";
+import { findHuggingFaceToken, HuggingFaceClient } from "./huggingface.js";
+import { createLlamaProvider, LLAMA_PROVIDER_ID } from "./provider.js";
+import { runWithProgress, showLlamaUi } from "./ui.js";
+function modelIsLoaded(model) {
+    return model.status.value === "loaded" || model.status.value === "sleeping";
+}
+function isConnectionError(error) {
+    if (!(error instanceof Error))
+        return false;
+    const message = `${error.name} ${error.message}`.toLowerCase();
+    return message.includes("fetch failed") || message.includes("timeout") || message.includes("network");
+}
+function connectionErrorMessage(error) {
+    if (isConnectionError(error))
+        return "Could not connect to the server.";
+    return error instanceof Error ? error.message : String(error);
+}
+function parseHuggingFaceModel(value) {
+    const colon = value.indexOf(":", value.indexOf("/") + 1);
+    return colon < 0
+        ? { repository: value }
+        : { repository: value.slice(0, colon), quantization: value.slice(colon + 1) };
+}
+async function configuredClient(ctx) {
+    const result = await ctx.modelRegistry.getProviderAuth(LLAMA_PROVIDER_ID);
+    if (!result) {
+        ctx.ui.notify(`Configure llama.cpp with /login ${LLAMA_PROVIDER_ID}`, "warning");
+        return undefined;
+    }
+    const configuredUrl = result.env?.LLAMA_BASE_URL;
+    const serverUrl = normalizeLlamaServerUrl(typeof configuredUrl === "string" && configuredUrl ? configuredUrl : (result.auth.baseUrl ?? ""));
+    return new LlamaClient(serverUrl, result.auth.apiKey);
+}
+export default function llamaExtension(pi) {
+    const provider = createLlamaProvider();
+    pi.registerProvider(provider.provider);
+    const syncCatalog = async (ctx, client, catalog) => {
+        const signal = AbortSignal.timeout(15_000);
+        const current = catalog ?? (await client.list({ signal }));
+        provider.setCatalog(current, client.serverUrl);
+        const result = await ctx.modelRegistry.refresh({
+            providers: [LLAMA_PROVIDER_ID],
+            // /llama already contacted the configured llama.cpp server, so keep this refresh live even in PI_OFFLINE.
+            allowNetwork: true,
+            signal,
+        });
+        if (result.aborted)
+            throw new Error("Model catalog refresh timed out.");
+        const refreshError = result.errors.get(LLAMA_PROVIDER_ID);
+        if (refreshError)
+            throw refreshError;
+        return current;
+    };
+    const loadModel = async (ctx, ui, client, catalog, target) => {
+        const loaded = catalog.filter((model) => model.id !== target.id && modelIsLoaded(model));
+        let replace = false;
+        if (loaded.length > 0) {
+            const choice = await ui.select(`${loaded.length} model${loaded.length === 1 ? " is" : "s are"} loaded`, [
+                "Unload all and load",
+                "Keep loaded and load",
+                "Cancel",
+            ]);
+            if (!choice || choice === "Cancel")
+                return;
+            replace = choice === "Unload all and load";
+        }
+        const restoreLoaded = async () => {
+            ctx.ui.notify("Restoring previously loaded models");
+            for (const model of loaded)
+                await client.loadAndWait(model.id, () => { });
+            await syncCatalog(ctx, client);
+        };
+        if (replace) {
+            for (const model of loaded)
+                await client.unloadAndWait(model.id);
+        }
+        try {
+            const result = await runWithProgress(ui, {
+                title: "Loading model",
+                model: target.id,
+                initialMessage: "Starting…",
+                cancelTitle: "Stop loading?",
+                cancelMessage: target.id,
+                run: (signal, update) => client.loadAndWait(target.id, update, signal),
+                cancel: () => client.unload(target.id),
+            });
+            if (result.cancelled) {
+                if (replace)
+                    await restoreLoaded();
+                return;
+            }
+            const refreshed = await syncCatalog(ctx, client);
+            const loadedModel = refreshed.find((model) => model.id === target.id);
+            ctx.ui.notify(loadedModel?.status.value === "loaded" ? `Loaded ${target.id}` : `Load started for ${target.id}`);
+        }
+        catch (error) {
+            if (replace) {
+                try {
+                    await restoreLoaded();
+                }
+                catch {
+                    // Preserve the original load error.
+                }
+            }
+            throw error;
+        }
+    };
+    const unloadModel = async (ctx, ui, client, model) => {
+        if (!(await ui.confirm("Unload model?", model.id)))
+            return;
+        await client.unloadAndWait(model.id);
+        await syncCatalog(ctx, client);
+        ctx.ui.notify(`Unloaded ${model.id}`);
+    };
+    const downloadModel = async (ctx, ui, client) => {
+        const huggingFace = new HuggingFaceClient(await findHuggingFaceToken());
+        const selected = await ui.searchModels((query, signal) => huggingFace.search(query, signal));
+        if (!selected)
+            return;
+        const parsed = parseHuggingFaceModel(selected);
+        ui.showStatus("Loading model details", parsed.repository);
+        const details = await huggingFace.details(parsed.repository);
+        if (details.gated) {
+            const approval = details.gated === "manual" ? "Manual approval is required" : "Accept the access terms";
+            const choice = await ui.select(`Hugging Face access required\n${details.id}\n\n${approval} at:\nhttps://huggingface.co/${details.id}\n\nThe llama.cpp server needs HF_TOKEN with access.`, ["Continue", "Back"]);
+            if (choice !== "Continue")
+                return;
+        }
+        let quantization = parsed.quantization;
+        if (!quantization && details.quantizations.length > 0) {
+            const options = details.quantizations.map((entry) => {
+                const detail = [
+                    entry.size === undefined ? undefined : formatBytes(entry.size),
+                    entry.name === "Q4_K_M" ? "recommended" : undefined,
+                ]
+                    .filter((value) => Boolean(value))
+                    .join(" · ");
+                return detail ? `${entry.name} · ${detail}` : entry.name;
+            });
+            const choice = await ui.select(`Select quantization\n${details.id}`, options);
+            if (!choice)
+                return;
+            quantization = details.quantizations[options.indexOf(choice)]?.name;
+            if (!quantization)
+                return;
+        }
+        const model = quantization ? `${details.id}:${quantization}` : details.id;
+        const result = await runWithProgress(ui, {
+            title: "Downloading model",
+            model,
+            initialMessage: "Starting…",
+            cancelTitle: "Stop download?",
+            cancelMessage: model,
+            run: (signal, update) => client.downloadAndWait(model, update, signal),
+            cancel: () => client.unload(model),
+        });
+        if (result.cancelled)
+            return;
+        await syncCatalog(ctx, client, result.value);
+        ctx.ui.notify(`Downloaded ${model}`);
+    };
+    pi.registerCommand("llama", {
+        description: "Manage llama.cpp router models",
+        handler: async (_args, ctx) => {
+            if (ctx.mode !== "tui") {
+                ctx.ui.notify("/llama is available in interactive mode", "warning");
+                return;
+            }
+            const client = await configuredClient(ctx);
+            if (!client)
+                return;
+            await showLlamaUi(ctx, async (ui) => {
+                const readCatalog = async () => {
+                    while (true) {
+                        try {
+                            return await syncCatalog(ctx, client);
+                        }
+                        catch (error) {
+                            if ((await ui.connectionError(client.serverUrl, connectionErrorMessage(error))) === "close") {
+                                return undefined;
+                            }
+                        }
+                    }
+                };
+                let catalog = await readCatalog();
+                if (!catalog)
+                    return;
+                while (true) {
+                    const action = await ui.showModels(client.serverUrl, catalog);
+                    if (action.type === "close")
+                        return;
+                    let actionError;
+                    try {
+                        if (action.type === "download")
+                            await downloadModel(ctx, ui, client);
+                        else if (modelIsLoaded(action.model))
+                            await unloadModel(ctx, ui, client, action.model);
+                        else if (action.model.status.value === "unloaded")
+                            await loadModel(ctx, ui, client, catalog, action.model);
+                        else
+                            ctx.ui.notify(`${action.model.id} is ${action.model.status.value}`, "warning");
+                    }
+                    catch (error) {
+                        actionError = error;
+                    }
+                    const refreshed = await readCatalog();
+                    if (!refreshed)
+                        return;
+                    catalog = refreshed;
+                    if (actionError && !isConnectionError(actionError)) {
+                        ctx.ui.notify(actionError instanceof Error ? actionError.message : String(actionError), "error");
+                    }
+                }
+            });
+        },
+    });
+}
+//# sourceMappingURL=index.js.map

package/dist/runtime/extensions/llama/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/extensions/llama/index.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,WAAW,EAAuB,uBAAuB,EAAE,MAAM,aAAa,CAAC;AACrG,OAAO,EAAE,oBAAoB,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAC3E,OAAO,EAAE,mBAAmB,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AACvE,OAAO,EAAgB,eAAe,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAErE,SAAS,aAAa,CAAC,KAAqB,EAAW;IACtD,OAAO,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,CAAC;AAAA,CAC5E;AAED,SAAS,iBAAiB,CAAC,KAAc,EAAW;IACnD,IAAI,CAAC,CAAC,KAAK,YAAY,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IAC5C,MAAM,OAAO,GAAG,GAAG,KAAK,CAAC,IAAI,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC,WAAW,EAAE,CAAC;IAC/D,OAAO,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;AAAA,CACtG;AAED,SAAS,sBAAsB,CAAC,KAAc,EAAU;IACvD,IAAI,iBAAiB,CAAC,KAAK,CAAC;QAAE,OAAO,kCAAkC,CAAC;IACxE,OAAO,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;AAAA,CAC9D;AAED,SAAS,qBAAqB,CAAC,KAAa,EAAiD;IAC5F,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,EAAE,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC;IACzD,OAAO,KAAK,GAAG,CAAC;QACf,CAAC,CAAC,EAAE,UAAU,EAAE,KAAK,EAAE;QACvB,CAAC,CAAC,EAAE,UAAU,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE,YAAY,EAAE,KAAK,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC,EAAE,CAAC;AAAA,CAC/E;AAED,KAAK,UAAU,gBAAgB,CAAC,GAA4B,EAAoC;IAC/F,MAAM,MAAM,GAAG,MAAM,GAAG,CAAC,aAAa,CAAC,eAAe,CAAC,iBAAiB,CAAC,CAAC;IAC1E,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,mCAAmC,iBAAiB,EAAE,EAAE,SAAS,CAAC,CAAC;QACjF,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,MAAM,aAAa,GAAG,MAAM,CAAC,GAAG,EAAE,cAAc,CAAC;IACjD,MAAM,SAAS,GAAG,uBAAuB,CACxC,OAAO,aAAa,KAAK,QAAQ,IAAI,aAAa,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,IAAI,EAAE,CAAC,CAChG,CAAC;IACF,OAAO,IAAI,WAAW,CAAC,SAAS,EAAE,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AAAA,CACtD;AAED,MAAM,CAAC,OAAO,UAAU,cAAc,CAAC,EAAgB,EAAQ;IAC9D,MAAM,QAAQ,GAAG,mBAAmB,EAAE,CAAC;IACvC,EAAE,CAAC,gBAAgB,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IAEvC,MAAM,WAAW,GAAG,KAAK,EACxB,GAA4B,EAC5B,MAAmB,EACnB,OAA0B,EACE,EAAE,CAAC;QAC/B,MAAM,MAAM,GAAG,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC3C,MAAM,OAAO,GAAG,OAAO,IAAI,CAAC,MAAM,MAAM,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC;QAC3D,QAAQ,CAAC,UAAU,CAAC,OAAO,EAAE,MAAM,CAAC,SAAS,CAAC,CAAC;QAC/C,MAAM,MAAM,GAAG,MAAM,GAAG,CAAC,aAAa,CAAC,OAAO,CAAC;YAC9C,SAAS,EAAE,CAAC,iBAAiB,CAAC;YAC9B,0GAA0G;YAC1G,YAAY,EAAE,IAAI;YAClB,MAAM;SACN,CAAC,CAAC;QACH,IAAI,MAAM,CAAC,OAAO;YAAE,MAAM,IAAI,KAAK,CAAC,kCAAkC,CAAC,CAAC;QACxE,MAAM,YAAY,GAAG,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,iBAAiB,CAAC,CAAC;QAC1D,IAAI,YAAY;YAAE,MAAM,YAAY,CAAC;QACrC,OAAO,OAAO,CAAC;IAAA,CACf,CAAC;IAEF,MAAM,SAAS,GAAG,KAAK,EACtB,GAA4B,EAC5B,EAAW,EACX,MAAmB,EACnB,OAAyB,EACzB,MAAsB,EACN,EAAE,CAAC;QACnB,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE,IAAI,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC;QACzF,IAAI,OAAO,GAAG,KAAK,CAAC;QACpB,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACvB,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,SAAS,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,OAAO,SAAS,EAAE;gBACvG,qBAAqB;gBACrB,sBAAsB;gBACtB,QAAQ;aACR,CAAC,CAAC;YACH,IAAI,CAAC,MAAM,IAAI,MAAM,KAAK,QAAQ;gBAAE,OAAO;YAC3C,OAAO,GAAG,MAAM,KAAK,qBAAqB,CAAC;QAC5C,CAAC;QAED,MAAM,aAAa,GAAG,KAAK,IAAmB,EAAE,CAAC;YAChD,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,oCAAoC,CAAC,CAAC;YACpD,KAAK,MAAM,KAAK,IAAI,MAAM;gBAAE,MAAM,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,EAAE,EAAE,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC,CAAC;YACzE,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAAA,CAC/B,CAAC;QACF,IAAI,OAAO,EAAE,CAAC;YACb,KAAK,MAAM,KAAK,IAAI,MAAM;gBAAE,MAAM,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAClE,CAAC;QAED,IAAI,CAAC;YACJ,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,EAAE,EAAE;gBACxC,KAAK,EAAE,eAAe;gBACtB,KAAK,EAAE,MAAM,CAAC,EAAE;gBAChB,cAAc,EAAE,aAAW;gBAC3B,WAAW,EAAE,eAAe;gBAC5B,aAAa,EAAE,MAAM,CAAC,EAAE;gBACxB,GAAG,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC;gBACtE,MAAM,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC;aACtC,CAAC,CAAC;YACH,IAAI,MAAM,CAAC,SAAS,EAAE,CAAC;gBACtB,IAAI,OAAO;oBAAE,MAAM,aAAa,EAAE,CAAC;gBACnC,OAAO;YACR,CAAC;YACD,MAAM,SAAS,GAAG,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;YACjD,MAAM,WAAW,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE,CAAC,CAAC;YACtE,GAAG,CAAC,EAAE,CAAC,MAAM,CACZ,WAAW,EAAE,MAAM,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,UAAU,MAAM,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,oBAAoB,MAAM,CAAC,EAAE,EAAE,CAChG,CAAC;QACH,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,IAAI,OAAO,EAAE,CAAC;gBACb,IAAI,CAAC;oBACJ,MAAM,aAAa,EAAE,CAAC;gBACvB,CAAC;gBAAC,MAAM,CAAC;oBACR,oCAAoC;gBACrC,CAAC;YACF,CAAC;YACD,MAAM,KAAK,CAAC;QACb,CAAC;IAAA,CACD,CAAC;IAEF,MAAM,WAAW,GAAG,KAAK,EACxB,GAA4B,EAC5B,EAAW,EACX,MAAmB,EACnB,KAAqB,EACL,EAAE,CAAC;QACnB,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,eAAe,EAAE,KAAK,CAAC,EAAE,CAAC,CAAC;YAAE,OAAO;QAC3D,MAAM,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACrC,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAC/B,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,YAAY,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC;IAAA,CACtC,CAAC;IAEF,MAAM,aAAa,GAAG,KAAK,EAAE,GAA4B,EAAE,EAAW,EAAE,MAAmB,EAAiB,EAAE,CAAC;QAC9G,MAAM,WAAW,GAAG,IAAI,iBAAiB,CAAC,MAAM,oBAAoB,EAAE,CAAC,CAAC;QACxE,MAAM,QAAQ,GAAG,MAAM,EAAE,CAAC,YAAY,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,WAAW,CAAC,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC;QAC7F,IAAI,CAAC,QAAQ;YAAE,OAAO;QACtB,MAAM,MAAM,GAAG,qBAAqB,CAAC,QAAQ,CAAC,CAAC;QAC/C,EAAE,CAAC,UAAU,CAAC,uBAAuB,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC;QAC1D,MAAM,OAAO,GAAG,MAAM,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;QAC7D,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;YACnB,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,6BAA6B,CAAC,CAAC,CAAC,yBAAyB,CAAC;YACxG,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAC7B,iCAAiC,OAAO,CAAC,EAAE,OAAO,QAAQ,gCAAgC,OAAO,CAAC,EAAE,sDAAsD,EAC1J,CAAC,UAAU,EAAE,MAAM,CAAC,CACpB,CAAC;YACF,IAAI,MAAM,KAAK,UAAU;gBAAE,OAAO;QACnC,CAAC;QACD,IAAI,YAAY,GAAG,MAAM,CAAC,YAAY,CAAC;QACvC,IAAI,CAAC,YAAY,IAAI,OAAO,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACvD,MAAM,OAAO,GAAG,OAAO,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;gBACpD,MAAM,MAAM,GAAG;oBACd,KAAK,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC,KAAK,CAAC,IAAI,CAAC;oBAC9D,KAAK,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS;iBACnD;qBACC,MAAM,CAAC,CAAC,KAAK,EAAmB,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;qBAClD,IAAI,CAAC,MAAK,CAAC,CAAC;gBACd,OAAO,MAAM,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,IAAI,OAAM,MAAM,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC;YAAA,CACzD,CAAC,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAAC,wBAAwB,OAAO,CAAC,EAAE,EAAE,EAAE,OAAO,CAAC,CAAC;YAC9E,IAAI,CAAC,MAAM;gBAAE,OAAO;YACpB,YAAY,GAAG,OAAO,CAAC,aAAa,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,EAAE,IAAI,CAAC;YACpE,IAAI,CAAC,YAAY;gBAAE,OAAO;QAC3B,CAAC;QACD,MAAM,KAAK,GAAG,YAAY,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,EAAE,IAAI,YAAY,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;QAC1E,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,EAAE,EAAE;YACxC,KAAK,EAAE,mBAAmB;YAC1B,KAAK;YACL,cAAc,EAAE,aAAW;YAC3B,WAAW,EAAE,gBAAgB;YAC7B,aAAa,EAAE,KAAK;YACpB,GAAG,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC;YACtE,MAAM,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC;SAClC,CAAC,CAAC;QACH,IAAI,MAAM,CAAC,SAAS;YAAE,OAAO;QAC7B,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;QAC7C,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,cAAc,KAAK,EAAE,CAAC,CAAC;IAAA,CACrC,CAAC;IAEF,EAAE,CAAC,eAAe,CAAC,OAAO,EAAE;QAC3B,WAAW,EAAE,gCAAgC;QAC7C,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,GAAG,EAAE,EAAE,CAAC;YAC9B,IAAI,GAAG,CAAC,IAAI,KAAK,KAAK,EAAE,CAAC;gBACxB,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,yCAAyC,EAAE,SAAS,CAAC,CAAC;gBACpE,OAAO;YACR,CAAC;YACD,MAAM,MAAM,GAAG,MAAM,gBAAgB,CAAC,GAAG,CAAC,CAAC;YAC3C,IAAI,CAAC,MAAM;gBAAE,OAAO;YACpB,MAAM,WAAW,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,EAAE,EAAE,CAAC;gBACpC,MAAM,WAAW,GAAG,KAAK,IAA2C,EAAE,CAAC;oBACtE,OAAO,IAAI,EAAE,CAAC;wBACb,IAAI,CAAC;4BACJ,OAAO,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;wBACvC,CAAC;wBAAC,OAAO,KAAK,EAAE,CAAC;4BAChB,IAAI,CAAC,MAAM,EAAE,CAAC,eAAe,CAAC,MAAM,CAAC,SAAS,EAAE,sBAAsB,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,OAAO,EAAE,CAAC;gCAC7F,OAAO,SAAS,CAAC;4BAClB,CAAC;wBACF,CAAC;oBACF,CAAC;gBAAA,CACD,CAAC;gBAEF,IAAI,OAAO,GAAG,MAAM,WAAW,EAAE,CAAC;gBAClC,IAAI,CAAC,OAAO;oBAAE,OAAO;gBACrB,OAAO,IAAI,EAAE,CAAC;oBACb,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,UAAU,CAAC,MAAM,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;oBAC9D,IAAI,MAAM,CAAC,IAAI,KAAK,OAAO;wBAAE,OAAO;oBACpC,IAAI,WAAoB,CAAC;oBACzB,IAAI,CAAC;wBACJ,IAAI,MAAM,CAAC,IAAI,KAAK,UAAU;4BAAE,MAAM,aAAa,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC;6BAChE,IAAI,aAAa,CAAC,MAAM,CAAC,KAAK,CAAC;4BAAE,MAAM,WAAW,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;6BAClF,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU;4BAChD,MAAM,SAAS,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;;4BACpD,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,EAAE,OAAO,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAC;oBACrF,CAAC;oBAAC,OAAO,KAAK,EAAE,CAAC;wBAChB,WAAW,GAAG,KAAK,CAAC;oBACrB,CAAC;oBACD,MAAM,SAAS,GAAG,MAAM,WAAW,EAAE,CAAC;oBACtC,IAAI,CAAC,SAAS;wBAAE,OAAO;oBACvB,OAAO,GAAG,SAAS,CAAC;oBACpB,IAAI,WAAW,IAAI,CAAC,iBAAiB,CAAC,WAAW,CAAC,EAAE,CAAC;wBACpD,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,WAAW,YAAY,KAAK,CAAC,CAAC,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,WAAW,CAAC,EAAE,OAAO,CAAC,CAAC;oBAClG,CAAC;gBACF,CAAC;YAAA,CACD,CAAC,CAAC;QAAA,CACH;KACD,CAAC,CAAC;AAAA,CACH","sourcesContent":["import type { ExtensionAPI, ExtensionCommandContext } from \"../../core/extensions/types.ts\";\nimport { formatBytes, LlamaClient, type LlamaModelInfo, normalizeLlamaServerUrl } from \"./client.ts\";\nimport { findHuggingFaceToken, HuggingFaceClient } from \"./huggingface.ts\";\nimport { createLlamaProvider, LLAMA_PROVIDER_ID } from \"./provider.ts\";\nimport { type LlamaUi, runWithProgress, showLlamaUi } from \"./ui.ts\";\n\nfunction modelIsLoaded(model: LlamaModelInfo): boolean {\n\treturn model.status.value === \"loaded\" || model.status.value === \"sleeping\";\n}\n\nfunction isConnectionError(error: unknown): boolean {\n\tif (!(error instanceof Error)) return false;\n\tconst message = `${error.name} ${error.message}`.toLowerCase();\n\treturn message.includes(\"fetch failed\") || message.includes(\"timeout\") || message.includes(\"network\");\n}\n\nfunction connectionErrorMessage(error: unknown): string {\n\tif (isConnectionError(error)) return \"Could not connect to the server.\";\n\treturn error instanceof Error ? error.message : String(error);\n}\n\nfunction parseHuggingFaceModel(value: string): { repository: string; quantization?: string } {\n\tconst colon = value.indexOf(\":\", value.indexOf(\"/\") + 1);\n\treturn colon < 0\n\t\t? { repository: value }\n\t\t: { repository: value.slice(0, colon), quantization: value.slice(colon + 1) };\n}\n\nasync function configuredClient(ctx: ExtensionCommandContext): Promise<LlamaClient | undefined> {\n\tconst result = await ctx.modelRegistry.getProviderAuth(LLAMA_PROVIDER_ID);\n\tif (!result) {\n\t\tctx.ui.notify(`Configure llama.cpp with /login ${LLAMA_PROVIDER_ID}`, \"warning\");\n\t\treturn undefined;\n\t}\n\tconst configuredUrl = result.env?.LLAMA_BASE_URL;\n\tconst serverUrl = normalizeLlamaServerUrl(\n\t\ttypeof configuredUrl === \"string\" && configuredUrl ? configuredUrl : (result.auth.baseUrl ?? \"\"),\n\t);\n\treturn new LlamaClient(serverUrl, result.auth.apiKey);\n}\n\nexport default function llamaExtension(pi: ExtensionAPI): void {\n\tconst provider = createLlamaProvider();\n\tpi.registerProvider(provider.provider);\n\n\tconst syncCatalog = async (\n\t\tctx: ExtensionCommandContext,\n\t\tclient: LlamaClient,\n\t\tcatalog?: LlamaModelInfo[],\n\t): Promise<LlamaModelInfo[]> => {\n\t\tconst signal = AbortSignal.timeout(15_000);\n\t\tconst current = catalog ?? (await client.list({ signal }));\n\t\tprovider.setCatalog(current, client.serverUrl);\n\t\tconst result = await ctx.modelRegistry.refresh({\n\t\t\tproviders: [LLAMA_PROVIDER_ID],\n\t\t\t// /llama already contacted the configured llama.cpp server, so keep this refresh live even in PI_OFFLINE.\n\t\t\tallowNetwork: true,\n\t\t\tsignal,\n\t\t});\n\t\tif (result.aborted) throw new Error(\"Model catalog refresh timed out.\");\n\t\tconst refreshError = result.errors.get(LLAMA_PROVIDER_ID);\n\t\tif (refreshError) throw refreshError;\n\t\treturn current;\n\t};\n\n\tconst loadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tcatalog: LlamaModelInfo[],\n\t\ttarget: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tconst loaded = catalog.filter((model) => model.id !== target.id && modelIsLoaded(model));\n\t\tlet replace = false;\n\t\tif (loaded.length > 0) {\n\t\t\tconst choice = await ui.select(`${loaded.length} model${loaded.length === 1 ? \" is\" : \"s are\"} loaded`, [\n\t\t\t\t\"Unload all and load\",\n\t\t\t\t\"Keep loaded and load\",\n\t\t\t\t\"Cancel\",\n\t\t\t]);\n\t\t\tif (!choice || choice === \"Cancel\") return;\n\t\t\treplace = choice === \"Unload all and load\";\n\t\t}\n\n\t\tconst restoreLoaded = async (): Promise<void> => {\n\t\t\tctx.ui.notify(\"Restoring previously loaded models\");\n\t\t\tfor (const model of loaded) await client.loadAndWait(model.id, () => {});\n\t\t\tawait syncCatalog(ctx, client);\n\t\t};\n\t\tif (replace) {\n\t\t\tfor (const model of loaded) await client.unloadAndWait(model.id);\n\t\t}\n\n\t\ttry {\n\t\t\tconst result = await runWithProgress(ui, {\n\t\t\t\ttitle: \"Loading model\",\n\t\t\t\tmodel: target.id,\n\t\t\t\tinitialMessage: \"Starting…\",\n\t\t\t\tcancelTitle: \"Stop loading?\",\n\t\t\t\tcancelMessage: target.id,\n\t\t\t\trun: (signal, update) => client.loadAndWait(target.id, update, signal),\n\t\t\t\tcancel: () => client.unload(target.id),\n\t\t\t});\n\t\t\tif (result.cancelled) {\n\t\t\t\tif (replace) await restoreLoaded();\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst refreshed = await syncCatalog(ctx, client);\n\t\t\tconst loadedModel = refreshed.find((model) => model.id === target.id);\n\t\t\tctx.ui.notify(\n\t\t\t\tloadedModel?.status.value === \"loaded\" ? `Loaded ${target.id}` : `Load started for ${target.id}`,\n\t\t\t);\n\t\t} catch (error) {\n\t\t\tif (replace) {\n\t\t\t\ttry {\n\t\t\t\t\tawait restoreLoaded();\n\t\t\t\t} catch {\n\t\t\t\t\t// Preserve the original load error.\n\t\t\t\t}\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t};\n\n\tconst unloadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tmodel: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tif (!(await ui.confirm(\"Unload model?\", model.id))) return;\n\t\tawait client.unloadAndWait(model.id);\n\t\tawait syncCatalog(ctx, client);\n\t\tctx.ui.notify(`Unloaded ${model.id}`);\n\t};\n\n\tconst downloadModel = async (ctx: ExtensionCommandContext, ui: LlamaUi, client: LlamaClient): Promise<void> => {\n\t\tconst huggingFace = new HuggingFaceClient(await findHuggingFaceToken());\n\t\tconst selected = await ui.searchModels((query, signal) => huggingFace.search(query, signal));\n\t\tif (!selected) return;\n\t\tconst parsed = parseHuggingFaceModel(selected);\n\t\tui.showStatus(\"Loading model details\", parsed.repository);\n\t\tconst details = await huggingFace.details(parsed.repository);\n\t\tif (details.gated) {\n\t\t\tconst approval = details.gated === \"manual\" ? \"Manual approval is required\" : \"Accept the access terms\";\n\t\t\tconst choice = await ui.select(\n\t\t\t\t`Hugging Face access required\\n${details.id}\\n\\n${approval} at:\\nhttps://huggingface.co/${details.id}\\n\\nThe llama.cpp server needs HF_TOKEN with access.`,\n\t\t\t\t[\"Continue\", \"Back\"],\n\t\t\t);\n\t\t\tif (choice !== \"Continue\") return;\n\t\t}\n\t\tlet quantization = parsed.quantization;\n\t\tif (!quantization && details.quantizations.length > 0) {\n\t\t\tconst options = details.quantizations.map((entry) => {\n\t\t\t\tconst detail = [\n\t\t\t\t\tentry.size === undefined ? undefined : formatBytes(entry.size),\n\t\t\t\t\tentry.name === \"Q4_K_M\" ? \"recommended\" : undefined,\n\t\t\t\t]\n\t\t\t\t\t.filter((value): value is string => Boolean(value))\n\t\t\t\t\t.join(\" · \");\n\t\t\t\treturn detail ? `${entry.name} · ${detail}` : entry.name;\n\t\t\t});\n\t\t\tconst choice = await ui.select(`Select quantization\\n${details.id}`, options);\n\t\t\tif (!choice) return;\n\t\t\tquantization = details.quantizations[options.indexOf(choice)]?.name;\n\t\t\tif (!quantization) return;\n\t\t}\n\t\tconst model = quantization ? `${details.id}:${quantization}` : details.id;\n\t\tconst result = await runWithProgress(ui, {\n\t\t\ttitle: \"Downloading model\",\n\t\t\tmodel,\n\t\t\tinitialMessage: \"Starting…\",\n\t\t\tcancelTitle: \"Stop download?\",\n\t\t\tcancelMessage: model,\n\t\t\trun: (signal, update) => client.downloadAndWait(model, update, signal),\n\t\t\tcancel: () => client.unload(model),\n\t\t});\n\t\tif (result.cancelled) return;\n\t\tawait syncCatalog(ctx, client, result.value);\n\t\tctx.ui.notify(`Downloaded ${model}`);\n\t};\n\n\tpi.registerCommand(\"llama\", {\n\t\tdescription: \"Manage llama.cpp router models\",\n\t\thandler: async (_args, ctx) => {\n\t\t\tif (ctx.mode !== \"tui\") {\n\t\t\t\tctx.ui.notify(\"/llama is available in interactive mode\", \"warning\");\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst client = await configuredClient(ctx);\n\t\t\tif (!client) return;\n\t\t\tawait showLlamaUi(ctx, async (ui) => {\n\t\t\t\tconst readCatalog = async (): Promise<LlamaModelInfo[] | undefined> => {\n\t\t\t\t\twhile (true) {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\treturn await syncCatalog(ctx, client);\n\t\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\t\tif ((await ui.connectionError(client.serverUrl, connectionErrorMessage(error))) === \"close\") {\n\t\t\t\t\t\t\t\treturn undefined;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t};\n\n\t\t\t\tlet catalog = await readCatalog();\n\t\t\t\tif (!catalog) return;\n\t\t\t\twhile (true) {\n\t\t\t\t\tconst action = await ui.showModels(client.serverUrl, catalog);\n\t\t\t\t\tif (action.type === \"close\") return;\n\t\t\t\t\tlet actionError: unknown;\n\t\t\t\t\ttry {\n\t\t\t\t\t\tif (action.type === \"download\") await downloadModel(ctx, ui, client);\n\t\t\t\t\t\telse if (modelIsLoaded(action.model)) await unloadModel(ctx, ui, client, action.model);\n\t\t\t\t\t\telse if (action.model.status.value === \"unloaded\")\n\t\t\t\t\t\t\tawait loadModel(ctx, ui, client, catalog, action.model);\n\t\t\t\t\t\telse ctx.ui.notify(`${action.model.id} is ${action.model.status.value}`, \"warning\");\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tactionError = error;\n\t\t\t\t\t}\n\t\t\t\t\tconst refreshed = await readCatalog();\n\t\t\t\t\tif (!refreshed) return;\n\t\t\t\t\tcatalog = refreshed;\n\t\t\t\t\tif (actionError && !isConnectionError(actionError)) {\n\t\t\t\t\t\tctx.ui.notify(actionError instanceof Error ? actionError.message : String(actionError), \"error\");\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t});\n\t\t},\n\t});\n}\n"]}

package/dist/runtime/extensions/llama/provider.d.ts

@@ -0,0 +1,12 @@
+import type { Provider } from "@earendil-works/pi-ai";
+import { type LlamaModelInfo } from "./client.ts";
+export declare const LLAMA_PROVIDER_ID = "llama.cpp";
+export declare const DEFAULT_LLAMA_SERVER_URL = "http://127.0.0.1:8080";
+export interface LlamaProviderController {
+    provider: Provider<"openai-completions">;
+    setCatalog(models: readonly LlamaModelInfo[], serverUrl: string, options?: {
+        routerAutoload?: boolean;
+    }): void;
+}
+export declare function createLlamaProvider(): LlamaProviderController;
+//# sourceMappingURL=provider.d.ts.map

package/dist/runtime/extensions/llama/provider.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"provider.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/provider.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAKX,QAAQ,EAGR,MAAM,uBAAuB,CAAC;AAE/B,OAAO,EAAe,KAAK,cAAc,EAA8C,MAAM,aAAa,CAAC;AAE3G,eAAO,MAAM,iBAAiB,cAAc,CAAC;AAC7C,eAAO,MAAM,wBAAwB,0BAA0B,CAAC;AA4DhE,MAAM,WAAW,uBAAuB;IACvC,QAAQ,EAAE,QAAQ,CAAC,oBAAoB,CAAC,CAAC;IACzC,UAAU,CAAC,MAAM,EAAE,SAAS,cAAc,EAAE,EAAE,SAAS,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,cAAc,CAAC,EAAE,OAAO,CAAA;KAAE,GAAG,IAAI,CAAC;CAC/G;AAED,wBAAgB,mBAAmB,IAAI,uBAAuB,CAqG7D","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 { LlamaClient, type LlamaModelInfo, llamaInferenceUrl, normalizeLlamaServerUrl } 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): Model<\"openai-completions\"> {\n\tconst reportedContextWindow = model.meta?.n_ctx ?? model.meta?.n_ctx_train;\n\tconst contextWindow = reportedContextWindow && reportedContextWindow > 0 ? reportedContextWindow : 128000;\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: false,\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},\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 = catalog\n\t\t\t\t.filter((model) => modelIsSelectable(model, routerAutoload))\n\t\t\t\t.map((model) => toPiModel(model, serverUrl));\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"]}

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

@@ -0,0 +1,146 @@
+import { stream, streamSimple } from "@earendil-works/pi-ai/compat";
+import { LlamaClient, llamaInferenceUrl, normalizeLlamaServerUrl } from "./client.js";
+export const LLAMA_PROVIDER_ID = "llama.cpp";
+export const DEFAULT_LLAMA_SERVER_URL = "http://127.0.0.1:8080";
+function credentialServerUrl(credential) {
+    const value = credential?.env?.LLAMA_BASE_URL;
+    return typeof value === "string" && value.trim() ? normalizeLlamaServerUrl(value) : undefined;
+}
+async function resolveServerUrl(ctx, credential) {
+    const configured = credentialServerUrl(credential) ?? (await ctx.env("LLAMA_BASE_URL"))?.trim();
+    return configured ? normalizeLlamaServerUrl(configured) : undefined;
+}
+function modelIsSelectable(model, routerAutoload) {
+    if (model.status.value === "loaded")
+        return true;
+    // llama.cpp reports idle-slept models as "sleeping"; requests wake them automatically.
+    if (model.status.value === "sleeping")
+        return true;
+    // Unloaded presets are routable only when llama.cpp router autoload can load them on first use.
+    return routerAutoload && model.status.value === "unloaded" && !model.status.failed && model.source === "preset";
+}
+async function routerAutoloadEnabled(client, catalog, signal) {
+    if (!catalog.some((model) => model.status.value === "unloaded" && model.source === "preset"))
+        return false;
+    try {
+        return (await client.props({ signal })).models_autoload === true;
+    }
+    catch {
+        return false;
+    }
+}
+function toPiModel(model, serverUrl) {
+    const reportedContextWindow = model.meta?.n_ctx ?? model.meta?.n_ctx_train;
+    const contextWindow = reportedContextWindow && reportedContextWindow > 0 ? reportedContextWindow : 128000;
+    return {
+        id: model.id,
+        name: model.id,
+        api: "openai-completions",
+        provider: LLAMA_PROVIDER_ID,
+        baseUrl: llamaInferenceUrl(serverUrl),
+        reasoning: false,
+        input: model.architecture?.input_modalities?.includes("image") ? ["text", "image"] : ["text"],
+        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+        contextWindow,
+        maxTokens: contextWindow,
+        compat: {
+            supportsStore: false,
+            supportsDeveloperRole: false,
+            supportsReasoningEffort: false,
+            supportsUsageInStreaming: true,
+            supportsStrictMode: false,
+            maxTokensField: "max_tokens",
+        },
+    };
+}
+export function createLlamaProvider() {
+    let models = [];
+    const setCatalog = (catalog, serverUrl, options = {}) => {
+        models = catalog
+            .filter((model) => modelIsSelectable(model, options.routerAutoload === true))
+            .map((model) => toPiModel(model, serverUrl));
+    };
+    const provider = {
+        id: LLAMA_PROVIDER_ID,
+        name: "llama.cpp",
+        baseUrl: llamaInferenceUrl(DEFAULT_LLAMA_SERVER_URL),
+        auth: {
+            apiKey: {
+                name: "llama.cpp server",
+                login: async (interaction) => {
+                    const enteredUrl = await interaction.prompt({
+                        type: "text",
+                        message: "llama.cpp server URL",
+                        placeholder: process.env.LLAMA_BASE_URL ?? DEFAULT_LLAMA_SERVER_URL,
+                    });
+                    const serverUrl = normalizeLlamaServerUrl(enteredUrl.trim() || process.env.LLAMA_BASE_URL || DEFAULT_LLAMA_SERVER_URL);
+                    const apiKey = (await interaction.prompt({
+                        type: "secret",
+                        message: "API key (optional)",
+                    })).trim();
+                    await new LlamaClient(serverUrl, apiKey || undefined).list({ signal: interaction.signal });
+                    return {
+                        type: "api_key",
+                        key: apiKey || undefined,
+                        env: { LLAMA_BASE_URL: serverUrl },
+                    };
+                },
+                check: async ({ ctx, credential }) => {
+                    const serverUrl = await resolveServerUrl(ctx, credential);
+                    return serverUrl
+                        ? { type: "api_key", source: credential ? "stored credential" : "LLAMA_BASE_URL" }
+                        : undefined;
+                },
+                resolve: async ({ ctx, credential }) => {
+                    const serverUrl = await resolveServerUrl(ctx, credential);
+                    if (!serverUrl)
+                        return undefined;
+                    const apiKey = credential?.key ?? (await ctx.env("LLAMA_API_KEY")) ?? "local";
+                    return {
+                        auth: { apiKey, baseUrl: llamaInferenceUrl(serverUrl) },
+                        env: { ...credential?.env, LLAMA_BASE_URL: serverUrl },
+                        source: credential ? "stored credential" : "LLAMA_BASE_URL",
+                    };
+                },
+            },
+        },
+        getModels: () => models,
+        refreshModels: async (context) => {
+            if (context.stored) {
+                const restored = context.stored.models.filter((model) => model.provider === LLAMA_PROVIDER_ID && model.api === "openai-completions");
+                if (!(await context.publish({
+                    update: () => {
+                        models = restored;
+                    },
+                }))) {
+                    return;
+                }
+            }
+            if (!context.allowNetwork || context.signal.aborted || context.credential?.type !== "api_key")
+                return;
+            const serverUrl = credentialServerUrl(context.credential);
+            if (!serverUrl)
+                return;
+            const client = new LlamaClient(serverUrl, context.credential.key);
+            const catalog = await client.list({ signal: context.signal });
+            if (context.signal.aborted)
+                return;
+            const routerAutoload = await routerAutoloadEnabled(client, catalog, context.signal);
+            if (context.signal.aborted)
+                return;
+            const refreshed = catalog
+                .filter((model) => modelIsSelectable(model, routerAutoload))
+                .map((model) => toPiModel(model, serverUrl));
+            await context.publish({
+                persist: { models: refreshed, checkedAt: Date.now() },
+                update: () => {
+                    models = refreshed;
+                },
+            });
+        },
+        stream: (model, context, options) => stream(model, context, options),
+        streamSimple: (model, context, options) => streamSimple(model, context, options),
+    };
+    return { provider, setCatalog };
+}
+//# sourceMappingURL=provider.js.map