@kweaver-ai/kweaver-sdk 0.7.2 → 0.7.4

package/dist/commands/model.js

@@ -0,0 +1,1315 @@
+import { readFile } from "node:fs/promises";
+import { createInterface } from "node:readline";
+import { readBundledModelTemplateFile } from "../bundled-model-templates.js";
+import { ensureValidToken, formatHttpError, with401RefreshRetry } from "../auth/oauth.js";
+import { resolveBusinessDomain } from "../config/store.js";
+import { addLlmModel, addSmallModel, deleteLlmModels, deleteSmallModels, editLlmModel, editSmallModel, getLlmModel, getSmallModel, listLlmModels, listSmallModels, testLlmModel, testSmallModel, } from "../api/models.js";
+import { modelChatCompletions, modelEmbeddings, modelRerank } from "../api/model-invocation.js";
+import { formatCallOutput } from "./call.js";
+const DEFAULT_LIST_LIMIT = 30;
+/** Strip global flags; fill default business domain. */
+export function parseModelGlobalFlags(args) {
+    let businessDomain = "";
+    let mfManagerBaseUrl;
+    let mfApiBaseUrl;
+    let pretty = true;
+    const rest = [];
+    for (let i = 0; i < args.length; i += 1) {
+        const a = args[i];
+        if ((a === "-bd" || a === "--biz-domain") && args[i + 1]) {
+            businessDomain = args[i + 1];
+            i += 1;
+            continue;
+        }
+        if (a === "--mf-base-url" && args[i + 1]) {
+            mfManagerBaseUrl = args[i + 1];
+            i += 1;
+            continue;
+        }
+        if (a === "--mf-api-base-url" && args[i + 1]) {
+            mfApiBaseUrl = args[i + 1];
+            i += 1;
+            continue;
+        }
+        if (a === "--json") {
+            pretty = false;
+            continue;
+        }
+        if (a === "--compact") {
+            pretty = false;
+            continue;
+        }
+        if (a === "--pretty") {
+            pretty = true;
+            continue;
+        }
+        rest.push(a);
+    }
+    if (!businessDomain) {
+        businessDomain = resolveBusinessDomain();
+    }
+    return { rest, businessDomain, mfManagerBaseUrl, mfApiBaseUrl, pretty };
+}
+function printModelUsage() {
+    console.log(`kweaver model
+
+Usage:
+  kweaver model llm   list [--keyword X] [--type llm|rlm|vu] [--series S] [--api-model M] [--page N] [--limit N] [--json] [-bd value]
+  kweaver model llm   get <model_id> [--json] [-bd value]
+  kweaver model llm   add --body-file <path.json> [--upstream-url <url>] [--api-model <id>] [--api-key <secret>|--api-key-file <path>] [--json] [-bd value]
+  kweaver model llm   edit [<model_id>] --body-file <path.json> [--upstream-url <url>] [--api-model <id>] [--api-key <secret>|--api-key-file <path>] [--json] [-bd value]
+                        (optional leading model_id overrides body.model_id after merge)
+                        OR: kweaver model llm edit <model_id> [sparse flags] (GET /llm/get then merge only flags you pass)
+                        Sparse flags: --name, --series, --type|-t, --max-model-len, --quota, --model-config-file, upstream flags (same as --body-file)
+  kweaver model llm   delete <model_id> [<model_id> ...] [-y] [-bd value]
+  kweaver model llm   test --body-file <path.json> [--upstream-url <url>] [--api-model <id>] [--api-key <secret>|--api-key-file <path>] [--json] [-bd value]
+  kweaver model llm   chat <model_id> (-m|--message) "text" [--model-name <registry_model_name>] [--skip-model-name-resolve] [--stream] [--no-stream] [--verbose] [--temperature N] [--max-tokens N] [--mf-api-base-url url] [-bd value]
+  kweaver model llm   --template [--json]                     (offline: bundled llm registration JSON stub)
+
+  kweaver model small list [--name X] [--type embedding|reranker] [--series S] [--page N] [--limit N] [--json] [-bd value]
+  kweaver model small get <model_id> [--json] [-bd value]
+  kweaver model small add --name N --type embedding|reranker --batch-size N
+                        (--model-config-file <path.json> | --adapter --adapter-code-file <path.py>)
+                        [--upstream-url <url>] [--api-model <id>] [--api-key <secret>|--api-key-file <path>]
+                        [--max-tokens N] [--embedding-dim N] [--json] [-bd value]
+                        (embedding requires --max-tokens and --embedding-dim; upstream flags merge into model_config.api_* — not valid with --adapter)
+  kweaver model small edit <model_id> [--body-file <path.json> | partial flags]
+                        (without --body-file: loads current model via GET, then applies only the flags you pass)
+  kweaver model small delete <model_id> [<model_id> ...] [-y] [-bd value]
+  kweaver model small test [--body-file <path.json>|<model_id>] [--json] [-bd value]
+  kweaver model small embeddings <model_id> (-i|--input <text>) ... [--model-name <registry_model_name>] [--skip-model-name-resolve] [--mf-api-base-url url] [-bd value]
+                        (runtime: POST mf-model-api /small-model/embeddings — like llm chat for vectors)
+  kweaver model small rerank <model_id> (-q|--query) <text> (-d|--document <text>) ... [--model-name <registry_model_name>] [--skip-model-name-resolve] [--mf-api-base-url url] [-bd value]
+                        (runtime: POST mf-model-api /small-model/reranker)
+  kweaver model small --template [--json]                   (offline: bundled small model_config JSON stub)
+
+Global:
+  --mf-base-url <url>       Override origin for mf-model-manager (default: KWEAVER_BASE_URL or KWEAVER_MF_MODEL_MANAGER_URL)
+  --mf-api-base-url <url>   Override origin for mf-model-api / chat (default: KWEAVER_BASE_URL or KWEAVER_MF_MODEL_API_URL)
+  -bd, --biz-domain        Business domain (default from config)
+
+Upstream secrets:
+  Prefer --api-key-file over --api-key (shell history). For LLM add/edit/test, flags merge into body.model_config as api_url, api_model, api_key (creating model_config if missing). Small-model upstream flags merge into model_config the same way.
+
+Bundled templates:
+  model llm --template | model small --template — print offline JSON stub (no auth).`);
+}
+async function printBundledModelBranchTemplate(branch, g) {
+    try {
+        const text = await readBundledModelTemplateFile(branch, "basic");
+        console.log(formatCallOutput(text, g.pretty));
+        return 0;
+    }
+    catch (e) {
+        console.error(e instanceof Error ? e.message : String(e));
+        return 1;
+    }
+}
+function confirmYes(prompt) {
+    return new Promise((resolve) => {
+        const rl = createInterface({ input: process.stdin, output: process.stdout });
+        rl.question(`${prompt} [y/N] `, (answer) => {
+            rl.close();
+            const t = answer.trim().toLowerCase();
+            resolve(t === "y" || t === "yes");
+        });
+    });
+}
+async function readJsonBodyFile(path) {
+    const raw = await readFile(path, "utf-8");
+    return JSON.parse(raw);
+}
+async function readApiKeyCredential(flags) {
+    const inline = flags.apiKey ?? "";
+    const path = flags.apiKeyFile ?? "";
+    if (inline && path) {
+        throw new Error("Use only one of --api-key or --api-key-file.");
+    }
+    if (path) {
+        return (await readFile(path, "utf-8")).trim();
+    }
+    return inline;
+}
+/** Merge cloud upstream settings into small-model ``model_config`` (api_url / api_model / api_key). */
+function mergeUpstreamIntoSmallModelConfig(mc, upstreamUrl, apiModel, apiKeyPlain) {
+    if (upstreamUrl)
+        mc.api_url = upstreamUrl;
+    if (apiModel)
+        mc.api_model = apiModel;
+    if (apiKeyPlain)
+        mc.api_key = apiKeyPlain;
+}
+async function mergeUpstreamIntoLlmModelConfig(body, opts) {
+    let mc = body.model_config;
+    if (!mc || typeof mc !== "object" || Array.isArray(mc)) {
+        mc = {};
+        body.model_config = mc;
+    }
+    const key = await readApiKeyCredential(opts);
+    mergeUpstreamIntoSmallModelConfig(mc, opts.upstreamUrl, opts.apiModel, key);
+}
+async function readLlmJsonBodyWithUpstreamMerge(tail, verb) {
+    let bodyFile = "";
+    const uf = {};
+    for (let i = 0; i < tail.length; i += 1) {
+        const a = tail[i];
+        if ((a === "--body-file" || a === "-f") && tail[i + 1]) {
+            bodyFile = tail[++i];
+            continue;
+        }
+        if ((a === "--upstream-url" || a === "--api-url") && tail[i + 1]) {
+            uf.upstreamUrl = tail[++i];
+            continue;
+        }
+        if (a === "--api-model" && tail[i + 1]) {
+            uf.apiModel = tail[++i];
+            continue;
+        }
+        if (a === "--api-key" && tail[i + 1]) {
+            uf.apiKey = tail[++i];
+            continue;
+        }
+        if ((a === "--api-key-file" || a === "--secret-file") && tail[i + 1]) {
+            uf.apiKeyFile = tail[++i];
+            continue;
+        }
+        throw new Error(`Unknown flag for model llm ${verb}: ${a}`);
+    }
+    if (!bodyFile) {
+        throw new Error(`kweaver model llm ${verb} requires --body-file <path.json>`);
+    }
+    const body = await readJsonBodyFile(bodyFile);
+    await mergeUpstreamIntoLlmModelConfig(body, uf);
+    return body;
+}
+function tailHasLlmBodyFileFlag(tail) {
+    for (let i = 0; i < tail.length; i += 1) {
+        const a = tail[i];
+        if ((a === "--body-file" || a === "-f") && tail[i + 1])
+            return true;
+    }
+    return false;
+}
+/** Exported for unit tests. */
+export function parsedLlmSparseEditHasUpdates(p) {
+    return (p.name !== undefined ||
+        p.modelSeries !== undefined ||
+        p.modelType !== undefined ||
+        p.maxModelLen !== undefined ||
+        p.quota !== undefined ||
+        p.modelConfigFile !== undefined ||
+        p.upstreamUrl !== undefined ||
+        p.apiModel !== undefined ||
+        p.apiKey !== undefined ||
+        p.apiKeyFile !== undefined);
+}
+/** Exported for unit tests. */
+export function parseLlmSparseEditFlags(tail) {
+    const o = {};
+    for (let i = 0; i < tail.length; i += 1) {
+        const a = tail[i];
+        if (a === "--name" && tail[i + 1]) {
+            o.name = tail[++i];
+            continue;
+        }
+        if (a === "--series" && tail[i + 1]) {
+            o.modelSeries = tail[++i];
+            continue;
+        }
+        if ((a === "--type" || a === "-t") && tail[i + 1]) {
+            o.modelType = tail[++i];
+            continue;
+        }
+        if (a === "--max-model-len" && tail[i + 1]) {
+            const n = Number(tail[++i]);
+            if (!Number.isFinite(n)) {
+                throw new Error("--max-model-len must be a finite number.");
+            }
+            o.maxModelLen = n;
+            continue;
+        }
+        if (a === "--quota") {
+            const v = tail[i + 1];
+            if (v === "true" || v === "1") {
+                o.quota = true;
+                i += 1;
+            }
+            else if (v === "false" || v === "0") {
+                o.quota = false;
+                i += 1;
+            }
+            else {
+                o.quota = true;
+            }
+            continue;
+        }
+        if ((a === "--model-config-file" || a === "--config-file") && tail[i + 1]) {
+            o.modelConfigFile = tail[++i];
+            continue;
+        }
+        if ((a === "--upstream-url" || a === "--api-url") && tail[i + 1]) {
+            o.upstreamUrl = tail[++i];
+            continue;
+        }
+        if (a === "--api-model" && tail[i + 1]) {
+            o.apiModel = tail[++i];
+            continue;
+        }
+        if (a === "--api-key" && tail[i + 1]) {
+            o.apiKey = tail[++i];
+            continue;
+        }
+        if ((a === "--api-key-file" || a === "--secret-file") && tail[i + 1]) {
+            o.apiKeyFile = tail[++i];
+            continue;
+        }
+        throw new Error(`Unknown flag for model llm edit: ${a}`);
+    }
+    return o;
+}
+/**
+ * Normalize GET /llm/get JSON into a body suitable for POST /llm/edit.
+ * Exported for unit tests.
+ */
+export function llmModelGetToEditBase(raw) {
+    if (!raw || typeof raw !== "object") {
+        throw new Error("llm/get returned invalid response.");
+    }
+    const obj = raw;
+    const mid = obj.model_id;
+    if (mid == null || String(mid).trim() === "") {
+        throw new Error("llm/get response missing model_id.");
+    }
+    const mcRaw = obj.model_config;
+    const modelConfig = mcRaw != null && typeof mcRaw === "object" && !Array.isArray(mcRaw)
+        ? { ...mcRaw }
+        : {};
+    const out = {
+        ...obj,
+        model_id: String(mid),
+        model_config: modelConfig,
+    };
+    // GET /llm/get sometimes omits `quota`; mf-model-manager edit handler expects it (KeyError otherwise).
+    const q = out.quota;
+    if (q === undefined || q === null) {
+        out.quota = false;
+    }
+    else if (typeof q === "number") {
+        out.quota = q !== 0;
+    }
+    else if (typeof q === "string") {
+        const s = q.trim().toLowerCase();
+        out.quota = s === "true" || s === "1";
+    }
+    return out;
+}
+/** Apply sparse llm edit flags onto a normalized record from GET. Exported for unit tests. */
+export async function mergeLlmEditOntoExistingBase(base, p) {
+    const out = {
+        ...base,
+        model_config: base.model_config != null &&
+            typeof base.model_config === "object" &&
+            !Array.isArray(base.model_config)
+            ? { ...base.model_config }
+            : {},
+    };
+    if (p.name !== undefined)
+        out.model_name = p.name;
+    if (p.modelSeries !== undefined)
+        out.model_series = p.modelSeries;
+    if (p.modelType !== undefined)
+        out.model_type = p.modelType;
+    if (p.maxModelLen !== undefined)
+        out.max_model_len = p.maxModelLen;
+    if (p.quota !== undefined)
+        out.quota = p.quota;
+    const wantsMc = p.modelConfigFile !== undefined ||
+        p.upstreamUrl !== undefined ||
+        p.apiModel !== undefined ||
+        p.apiKey !== undefined ||
+        p.apiKeyFile !== undefined;
+    if (wantsMc) {
+        const mc = out.model_config;
+        if (p.modelConfigFile) {
+            const rawJson = await readFile(p.modelConfigFile, "utf-8");
+            Object.assign(mc, JSON.parse(rawJson));
+        }
+        const apiKeyPlain = await readApiKeyCredential(p);
+        mergeUpstreamIntoSmallModelConfig(mc, p.upstreamUrl, p.apiModel, apiKeyPlain);
+        out.model_config = mc;
+    }
+    return out;
+}
+async function mergeLlmEditFromPartialFlags(modelId, p, mfOpts) {
+    const raw = await getLlmModel({ ...mfOpts, modelId });
+    const base = llmModelGetToEditBase(raw);
+    if (String(base.model_id) !== String(modelId)) {
+        base.model_id = modelId;
+    }
+    return mergeLlmEditOntoExistingBase(base, p);
+}
+async function runLlmList(g, tail) {
+    let keyword = "";
+    let type = "";
+    let series = "all";
+    let apiModel = "";
+    let page = 1;
+    let limit = DEFAULT_LIST_LIMIT;
+    let quota;
+    for (let i = 0; i < tail.length; i += 1) {
+        const a = tail[i];
+        if ((a === "--keyword" || a === "-k") && tail[i + 1]) {
+            keyword = tail[++i];
+            continue;
+        }
+        if ((a === "--type" || a === "-t") && tail[i + 1]) {
+            type = tail[++i];
+            continue;
+        }
+        if (a === "--series" && tail[i + 1]) {
+            series = tail[++i];
+            continue;
+        }
+        if (a === "--api-model" && tail[i + 1]) {
+            apiModel = tail[++i];
+            continue;
+        }
+        if (a === "--page" && tail[i + 1]) {
+            page = Math.max(1, Number(tail[++i]) || 1);
+            continue;
+        }
+        if (a === "--limit" && tail[i + 1]) {
+            limit = Math.max(1, Number(tail[++i]) || DEFAULT_LIST_LIMIT);
+            continue;
+        }
+        if (a === "--quota") {
+            const v = tail[i + 1];
+            if (v === "true" || v === "1") {
+                quota = true;
+                i += 1;
+            }
+            else if (v === "false" || v === "0") {
+                quota = false;
+                i += 1;
+            }
+            else {
+                quota = true;
+            }
+            continue;
+        }
+        console.error(`Unknown flag for model llm list: ${a}`);
+        return 1;
+    }
+    const token = await ensureValidToken();
+    const data = await listLlmModels({
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+        page,
+        size: limit,
+        name: keyword,
+        modelType: type,
+        series,
+        apiModel,
+        quota,
+    });
+    console.log(formatCallOutput(JSON.stringify(data), g.pretty));
+    return 0;
+}
+async function runSmallList(g, tail) {
+    let modelName = "";
+    let modelType = "";
+    let modelSeries = "";
+    let page = 1;
+    let limit = DEFAULT_LIST_LIMIT;
+    for (let i = 0; i < tail.length; i += 1) {
+        const a = tail[i];
+        if ((a === "--name" || a === "--keyword" || a === "-k") && tail[i + 1]) {
+            modelName = tail[++i];
+            continue;
+        }
+        if ((a === "--type" || a === "-t") && tail[i + 1]) {
+            modelType = tail[++i];
+            continue;
+        }
+        if (a === "--series" && tail[i + 1]) {
+            modelSeries = tail[++i];
+            continue;
+        }
+        if (a === "--page" && tail[i + 1]) {
+            page = Math.max(1, Number(tail[++i]) || 1);
+            continue;
+        }
+        if (a === "--limit" && tail[i + 1]) {
+            limit = Math.max(1, Number(tail[++i]) || DEFAULT_LIST_LIMIT);
+            continue;
+        }
+        console.error(`Unknown flag for model small list: ${a}`);
+        return 1;
+    }
+    const token = await ensureValidToken();
+    const data = await listSmallModels({
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+        page,
+        size: limit,
+        modelName,
+        modelType,
+        modelSeries,
+    });
+    console.log(formatCallOutput(JSON.stringify(data), g.pretty));
+    return 0;
+}
+async function runLlmGet(g, tail) {
+    const id = tail[0];
+    if (!id || id.startsWith("-")) {
+        console.error("Usage: kweaver model llm get <model_id>");
+        return 1;
+    }
+    const token = await ensureValidToken();
+    const data = await getLlmModel({
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+        modelId: id,
+    });
+    console.log(formatCallOutput(JSON.stringify(data), g.pretty));
+    return 0;
+}
+async function runSmallGet(g, tail) {
+    const id = tail[0];
+    if (!id || id.startsWith("-")) {
+        console.error("Usage: kweaver model small get <model_id>");
+        return 1;
+    }
+    const token = await ensureValidToken();
+    const data = await getSmallModel({
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+        modelId: id,
+    });
+    console.log(formatCallOutput(JSON.stringify(data), g.pretty));
+    return 0;
+}
+async function runLlmAdd(g, tail) {
+    let body;
+    try {
+        body = await readLlmJsonBodyWithUpstreamMerge(tail, "add");
+    }
+    catch (e) {
+        console.error(e instanceof Error ? e.message : String(e));
+        return 1;
+    }
+    const token = await ensureValidToken();
+    const data = await addLlmModel({
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+        body,
+    });
+    console.log(formatCallOutput(JSON.stringify(data), g.pretty));
+    return 0;
+}
+async function runLlmEdit(g, tail) {
+    const token = await ensureValidToken();
+    const mfOpts = {
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+    };
+    let body;
+    try {
+        if (tailHasLlmBodyFileFlag(tail)) {
+            let slice = tail;
+            let forcedModelId;
+            if (tail[0] && !tail[0].startsWith("-")) {
+                forcedModelId = tail[0];
+                slice = tail.slice(1);
+            }
+            body = await readLlmJsonBodyWithUpstreamMerge(slice, "edit");
+            if (forcedModelId)
+                body.model_id = forcedModelId;
+        }
+        else {
+            const modelId = tail[0];
+            if (!modelId || modelId.startsWith("-")) {
+                console.error("Usage: kweaver model llm edit <model_id> [sparse flags] | [--body-file <path.json> ...] (see kweaver model --help)");
+                return 1;
+            }
+            const parsed = parseLlmSparseEditFlags(tail.slice(1));
+            if (!parsedLlmSparseEditHasUpdates(parsed)) {
+                throw new Error("llm edit: pass --body-file or at least one of --name, --series, --type, --max-model-len, --quota, --model-config-file, --upstream-url, --api-model, --api-key/--api-key-file");
+            }
+            body = await mergeLlmEditFromPartialFlags(modelId, parsed, mfOpts);
+        }
+    }
+    catch (e) {
+        console.error(e instanceof Error ? e.message : String(e));
+        return 1;
+    }
+    const data = await editLlmModel({
+        ...mfOpts,
+        body,
+    });
+    console.log(formatCallOutput(JSON.stringify(data), g.pretty));
+    return 0;
+}
+async function runLlmTest(g, tail) {
+    let body;
+    try {
+        body = await readLlmJsonBodyWithUpstreamMerge(tail, "test");
+    }
+    catch (e) {
+        console.error(e instanceof Error ? e.message : String(e));
+        return 1;
+    }
+    const token = await ensureValidToken();
+    const data = await testLlmModel({
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+        body,
+    });
+    console.log(formatCallOutput(JSON.stringify(data), g.pretty));
+    return 0;
+}
+async function runLlmDelete(g, tail) {
+    let skipConfirm = false;
+    const ids = [];
+    for (const a of tail) {
+        if (a === "-y" || a === "--yes") {
+            skipConfirm = true;
+            continue;
+        }
+        if (!a.startsWith("-"))
+            ids.push(a);
+    }
+    if (ids.length === 0) {
+        console.error("Usage: kweaver model llm delete <model_id> [...] [-y]");
+        return 1;
+    }
+    if (!skipConfirm) {
+        const ok = await confirmYes(`Delete LLM model(s) ${ids.join(", ")}?`);
+        if (!ok) {
+            console.error("Cancelled.");
+            return 1;
+        }
+    }
+    const token = await ensureValidToken();
+    const data = await deleteLlmModels({
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+        modelIds: ids,
+    });
+    console.log(formatCallOutput(JSON.stringify(data), g.pretty));
+    return 0;
+}
+/**
+ * Returns registry **model_name** from ``GET /llm/get`` JSON when present. Exported for unit tests.
+ */
+export function llmGetRecordModelName(raw) {
+    if (!raw || typeof raw !== "object")
+        return "";
+    const mn = raw.model_name;
+    return typeof mn === "string" && mn.trim().length > 0 ? mn.trim() : "";
+}
+/** Registry **model_name** from ``GET /small-model/get``. Exported for unit tests. */
+export function smallGetRecordModelName(raw) {
+    if (!raw || typeof raw !== "object")
+        return "";
+    const mn = raw.model_name;
+    return typeof mn === "string" && mn.trim().length > 0 ? mn.trim() : "";
+}
+async function runLlmChat(g, tail) {
+    const modelId = tail[0];
+    if (!modelId || modelId.startsWith("-")) {
+        console.error('Usage: kweaver model llm chat <model_id> (-m|--message) "text" [--model-name <registry_model_name>] [--skip-model-name-resolve] ...');
+        return 1;
+    }
+    let message = "";
+    let registryModelName = "";
+    let skipModelNameResolve = false;
+    let stream = true;
+    let verbose = false;
+    let temperature;
+    let maxTokens;
+    for (let i = 1; i < tail.length; i += 1) {
+        const a = tail[i];
+        if ((a === "--model-name" || a === "--register-name") && tail[i + 1]) {
+            registryModelName = tail[++i];
+            continue;
+        }
+        if (a === "--skip-model-name-resolve") {
+            skipModelNameResolve = true;
+            continue;
+        }
+        if ((a === "-m" || a === "--message") && tail[i + 1]) {
+            message = tail[++i];
+            continue;
+        }
+        if (a === "--stream") {
+            stream = true;
+            continue;
+        }
+        if (a === "--no-stream") {
+            stream = false;
+            continue;
+        }
+        if (a === "--verbose" || a === "-v") {
+            verbose = true;
+            continue;
+        }
+        if (a === "--temperature" && tail[i + 1]) {
+            temperature = Number(tail[++i]);
+            continue;
+        }
+        if (a === "--max-tokens" && tail[i + 1]) {
+            maxTokens = Number(tail[++i]);
+            continue;
+        }
+        console.error(`Unknown flag for model llm chat: ${a}`);
+        return 1;
+    }
+    if (!message) {
+        console.error("kweaver model llm chat requires -m / --message");
+        return 1;
+    }
+    const token = await ensureValidToken();
+    const mfManagerOpts = {
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+    };
+    let resolvedModelName = registryModelName.trim();
+    if (resolvedModelName.length === 0 && !skipModelNameResolve) {
+        try {
+            const raw = await getLlmModel({ ...mfManagerOpts, modelId });
+            resolvedModelName = llmGetRecordModelName(raw);
+        }
+        catch (e) {
+            console.error(e instanceof Error ? e.message : String(e));
+            return 1;
+        }
+    }
+    const result = await modelChatCompletions({
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfApiBaseUrl: g.mfApiBaseUrl,
+        modelId,
+        ...(resolvedModelName.length > 0 ? { modelName: resolvedModelName } : {}),
+        messages: [{ role: "user", content: message }],
+        stream,
+        temperature,
+        maxTokens,
+        verbose,
+    });
+    if (stream) {
+        process.stdout.write(result.text);
+        if (!result.text.endsWith("\n"))
+            process.stdout.write("\n");
+    }
+    else {
+        console.log(result.text);
+    }
+    return 0;
+}
+/** True when sparse CLI flags should perform an edit (excluding --body-file). Exported for unit tests. */
+export function parsedSmallFlagsHasEditUpdates(p) {
+    return (p.name !== undefined ||
+        p.modelType !== undefined ||
+        p.batchSize !== undefined ||
+        p.maxTokens !== undefined ||
+        p.embeddingDim !== undefined ||
+        p.modelConfigFile !== undefined ||
+        p.adapter === true ||
+        p.adapterCodeFile !== undefined ||
+        p.upstreamUrl !== undefined ||
+        p.apiModel !== undefined ||
+        p.apiKey !== undefined ||
+        p.apiKeyFile !== undefined);
+}
+function coerceFiniteNumber(value, label) {
+    const n = typeof value === "number" ? value : Number(value);
+    if (!Number.isFinite(n)) {
+        throw new Error(`${label} must be a finite number, got: ${String(value)}`);
+    }
+    return n;
+}
+/**
+ * Normalize GET /small-model/get JSON into a body suitable for POST /small-model/edit.
+ * Exported for unit tests.
+ */
+export function smallModelGetToEditBase(raw) {
+    if (!raw || typeof raw !== "object") {
+        throw new Error("small-model/get returned invalid response.");
+    }
+    const o = raw;
+    const mid = o.model_id;
+    if (mid == null || String(mid).trim() === "") {
+        throw new Error("small-model/get response missing model_id.");
+    }
+    const mcRaw = o.model_config;
+    const mc = mcRaw != null && typeof mcRaw === "object" && !Array.isArray(mcRaw)
+        ? { ...mcRaw }
+        : {};
+    const body = {
+        model_id: String(mid),
+        model_name: typeof o.model_name === "string" ? o.model_name : String(o.model_name ?? ""),
+        model_type: typeof o.model_type === "string" ? o.model_type : String(o.model_type ?? ""),
+        batch_size: coerceFiniteNumber(o.batch_size, "batch_size"),
+        model_config: mc,
+        adapter: o.adapter === true,
+        adapter_code: typeof o.adapter_code === "string" ? o.adapter_code : "",
+    };
+    if (o.max_tokens != null && String(o.max_tokens) !== "") {
+        body.max_tokens = coerceFiniteNumber(o.max_tokens, "max_tokens");
+    }
+    if (o.embedding_dim != null && String(o.embedding_dim) !== "") {
+        body.embedding_dim = coerceFiniteNumber(o.embedding_dim, "embedding_dim");
+    }
+    return body;
+}
+/**
+ * Apply sparse edit flags onto a normalized small-model record (from GET). Exported for unit tests.
+ */
+export async function mergeSmallEditOntoExistingBase(base, p) {
+    const out = {
+        ...base,
+        model_config: base.model_config != null &&
+            typeof base.model_config === "object" &&
+            !Array.isArray(base.model_config)
+            ? { ...base.model_config }
+            : {},
+    };
+    delete out.change;
+    if (p.name !== undefined)
+        out.model_name = p.name;
+    if (p.modelType !== undefined)
+        out.model_type = p.modelType;
+    if (p.batchSize !== undefined)
+        out.batch_size = p.batchSize;
+    if (p.maxTokens !== undefined)
+        out.max_tokens = p.maxTokens;
+    if (p.embeddingDim !== undefined)
+        out.embedding_dim = p.embeddingDim;
+    const existingAdapter = base.adapter === true;
+    const wantsConfigInputs = p.modelConfigFile !== undefined ||
+        p.upstreamUrl !== undefined ||
+        p.apiModel !== undefined ||
+        p.apiKey !== undefined ||
+        p.apiKeyFile !== undefined;
+    const wantsAdapterInputs = p.adapter === true || p.adapterCodeFile !== undefined;
+    if (existingAdapter && wantsConfigInputs) {
+        throw new Error("This small model uses adapter mode; use --adapter-code-file to replace code, or --body-file for a full JSON body. Upstream flags and --model-config-file are not merged for adapter models.");
+    }
+    if (wantsAdapterInputs) {
+        if (existingAdapter) {
+            if (p.adapterCodeFile) {
+                out.adapter = true;
+                out.adapter_code = await readFile(p.adapterCodeFile, "utf-8");
+            }
+            return out;
+        }
+        if (!p.adapterCodeFile) {
+            throw new Error("small edit: switching to adapter mode requires --adapter-code-file (Python source).");
+        }
+        out.adapter = true;
+        out.adapter_code = await readFile(p.adapterCodeFile, "utf-8");
+        out.model_config = {};
+        return out;
+    }
+    if (wantsConfigInputs) {
+        const mc = out.model_config;
+        if (p.modelConfigFile) {
+            const rawJson = await readFile(p.modelConfigFile, "utf-8");
+            Object.assign(mc, JSON.parse(rawJson));
+        }
+        const apiKeyPlain = await readApiKeyCredential(p);
+        mergeUpstreamIntoSmallModelConfig(mc, p.upstreamUrl, p.apiModel, apiKeyPlain);
+        out.model_config = mc;
+        out.adapter = false;
+        out.adapter_code = "";
+        if (apiKeyPlain.length > 0)
+            out.change = true;
+        return out;
+    }
+    return out;
+}
+async function mergeSmallEditFromPartialFlags(modelId, p, mfOpts) {
+    const raw = await getSmallModel({ ...mfOpts, modelId });
+    const base = smallModelGetToEditBase(raw);
+    if (String(base.model_id) !== String(modelId)) {
+        base.model_id = modelId;
+    }
+    return mergeSmallEditOntoExistingBase(base, p);
+}
+/** Parse small-model add/edit flags from argv tail (after action). Exported for unit tests. */
+export function parseSmallAddFlags(tail) {
+    const o = {};
+    for (let i = 0; i < tail.length; i += 1) {
+        const a = tail[i];
+        if (a === "--name" && tail[i + 1]) {
+            o.name = tail[++i];
+            continue;
+        }
+        if (a === "--type" && tail[i + 1]) {
+            const t = tail[++i];
+            if (t !== "embedding" && t !== "reranker") {
+                throw new Error(`--type must be embedding or reranker, got: ${t}`);
+            }
+            o.modelType = t;
+            continue;
+        }
+        if (a === "--batch-size" && tail[i + 1]) {
+            o.batchSize = Number(tail[++i]);
+            continue;
+        }
+        if (a === "--max-tokens" && tail[i + 1]) {
+            o.maxTokens = Number(tail[++i]);
+            continue;
+        }
+        if (a === "--embedding-dim" && tail[i + 1]) {
+            o.embeddingDim = Number(tail[++i]);
+            continue;
+        }
+        if ((a === "--model-config-file" || a === "--config-file") && tail[i + 1]) {
+            o.modelConfigFile = tail[++i];
+            continue;
+        }
+        if (a === "--adapter") {
+            o.adapter = true;
+            continue;
+        }
+        if ((a === "--adapter-code-file" || a === "--code-file") && tail[i + 1]) {
+            o.adapterCodeFile = tail[++i];
+            continue;
+        }
+        if ((a === "--upstream-url" || a === "--api-url") && tail[i + 1]) {
+            o.upstreamUrl = tail[++i];
+            continue;
+        }
+        if (a === "--api-model" && tail[i + 1]) {
+            o.apiModel = tail[++i];
+            continue;
+        }
+        if (a === "--api-key" && tail[i + 1]) {
+            o.apiKey = tail[++i];
+            continue;
+        }
+        if ((a === "--api-key-file" || a === "--secret-file") && tail[i + 1]) {
+            o.apiKeyFile = tail[++i];
+            continue;
+        }
+        if ((a === "--body-file" || a === "-f") && tail[i + 1]) {
+            o.bodyFile = tail[++i];
+            continue;
+        }
+        throw new Error(`Unknown flag for small model: ${a}`);
+    }
+    return o;
+}
+export async function buildSmallBodyFromFlags(p, modelId) {
+    if (p.bodyFile) {
+        return readJsonBodyFile(p.bodyFile);
+    }
+    if (!p.name || !p.modelType || p.batchSize == null || Number.isNaN(p.batchSize)) {
+        throw new Error("small add/edit requires --name, --type, --batch-size or --body-file");
+    }
+    const body = {
+        model_name: p.name,
+        model_type: p.modelType,
+        batch_size: p.batchSize,
+    };
+    if (modelId)
+        body.model_id = modelId;
+    if (p.modelType === "embedding") {
+        if (p.maxTokens == null || p.embeddingDim == null) {
+            throw new Error("embedding type requires --max-tokens and --embedding-dim");
+        }
+        body.max_tokens = p.maxTokens;
+        body.embedding_dim = p.embeddingDim;
+    }
+    if (p.adapter) {
+        if (p.upstreamUrl || p.apiModel || p.apiKey || p.apiKeyFile) {
+            throw new Error("Upstream flags (--upstream-url, --api-model, --api-key/--api-key-file) cannot be used with --adapter. Use direct config (--model-config-file with optional upstream flags, or only upstream flags), or embed URL/key inside adapter_code.");
+        }
+        if (!p.adapterCodeFile)
+            throw new Error("--adapter requires --adapter-code-file");
+        const code = await readFile(p.adapterCodeFile, "utf-8");
+        body.adapter = true;
+        body.adapter_code = code;
+        return body;
+    }
+    const apiKeyPlain = await readApiKeyCredential(p);
+    const mc = {};
+    if (p.modelConfigFile) {
+        const raw = await readFile(p.modelConfigFile, "utf-8");
+        Object.assign(mc, JSON.parse(raw));
+    }
+    mergeUpstreamIntoSmallModelConfig(mc, p.upstreamUrl, p.apiModel, apiKeyPlain);
+    if (Object.keys(mc).length === 0) {
+        throw new Error("Provide --model-config-file and/or --upstream-url with --api-key or --api-key-file (and usually --api-model) for cloud-hosted small models.");
+    }
+    body.model_config = mc;
+    body.adapter = false;
+    return body;
+}
+async function runSmallAdd(g, tail) {
+    let parsed;
+    try {
+        parsed = parseSmallAddFlags(tail);
+    }
+    catch (e) {
+        console.error(e instanceof Error ? e.message : String(e));
+        return 1;
+    }
+    let body;
+    try {
+        body = await buildSmallBodyFromFlags(parsed);
+    }
+    catch (e) {
+        console.error(e instanceof Error ? e.message : String(e));
+        return 1;
+    }
+    const token = await ensureValidToken();
+    const data = await addSmallModel({
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+        body,
+    });
+    console.log(formatCallOutput(JSON.stringify(data), g.pretty));
+    return 0;
+}
+async function runSmallEdit(g, tail) {
+    const modelId = tail[0];
+    if (!modelId || modelId.startsWith("-")) {
+        console.error("Usage: kweaver model small edit <model_id> [--body-file ... | flags...]");
+        return 1;
+    }
+    let parsed;
+    try {
+        parsed = parseSmallAddFlags(tail.slice(1));
+    }
+    catch (e) {
+        console.error(e instanceof Error ? e.message : String(e));
+        return 1;
+    }
+    const token = await ensureValidToken();
+    const mfOpts = {
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+    };
+    let body;
+    try {
+        if (parsed.bodyFile) {
+            body = await readJsonBodyFile(parsed.bodyFile);
+            body.model_id = modelId;
+        }
+        else if (!parsedSmallFlagsHasEditUpdates(parsed)) {
+            throw new Error("small edit: pass --body-file or at least one of --name, --type, --batch-size, --max-tokens, --embedding-dim, --model-config-file, --adapter/--adapter-code-file, --upstream-url, --api-model, --api-key/--api-key-file");
+        }
+        else {
+            body = await mergeSmallEditFromPartialFlags(modelId, parsed, mfOpts);
+        }
+    }
+    catch (e) {
+        console.error(e instanceof Error ? e.message : String(e));
+        return 1;
+    }
+    const data = await editSmallModel({
+        ...mfOpts,
+        body,
+    });
+    console.log(formatCallOutput(JSON.stringify(data), g.pretty));
+    return 0;
+}
+async function runSmallDelete(g, tail) {
+    let skipConfirm = false;
+    const ids = [];
+    for (const a of tail) {
+        if (a === "-y" || a === "--yes") {
+            skipConfirm = true;
+            continue;
+        }
+        if (!a.startsWith("-"))
+            ids.push(a);
+    }
+    if (ids.length === 0) {
+        console.error("Usage: kweaver model small delete <model_id> [...] [-y]");
+        return 1;
+    }
+    if (!skipConfirm) {
+        const ok = await confirmYes(`Delete small model(s) ${ids.join(", ")}?`);
+        if (!ok) {
+            console.error("Cancelled.");
+            return 1;
+        }
+    }
+    const token = await ensureValidToken();
+    const data = await deleteSmallModels({
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+        modelIds: ids,
+    });
+    console.log(formatCallOutput(JSON.stringify(data), g.pretty));
+    return 0;
+}
+async function runSmallTest(g, tail) {
+    let bodyFile = "";
+    const positional = [];
+    for (let i = 0; i < tail.length; i += 1) {
+        const a = tail[i];
+        if ((a === "--body-file" || a === "-f") && tail[i + 1]) {
+            bodyFile = tail[++i];
+            continue;
+        }
+        if (a.startsWith("-")) {
+            console.error(`Unknown flag for model small test: ${a}`);
+            return 1;
+        }
+        positional.push(a);
+    }
+    let body;
+    if (bodyFile) {
+        body = await readJsonBodyFile(bodyFile);
+    }
+    else if (positional.length === 1) {
+        body = { model_id: positional[0] };
+    }
+    else {
+        console.error("Usage: kweaver model small test <model_id> | --body-file <path.json>");
+        return 1;
+    }
+    const token = await ensureValidToken();
+    const data = await testSmallModel({
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+        body,
+    });
+    console.log(formatCallOutput(JSON.stringify(data), g.pretty));
+    return 0;
+}
+async function runSmallEmbeddings(g, tail) {
+    const modelId = tail[0];
+    if (!modelId || modelId.startsWith("-")) {
+        console.error('Usage: kweaver model small embeddings <model_id> (-i|--input <text>) ... [--model-name <registry_model_name>] [--skip-model-name-resolve]');
+        return 1;
+    }
+    const inputs = [];
+    let registryModelName = "";
+    let skipModelNameResolve = false;
+    for (let i = 1; i < tail.length; i += 1) {
+        const a = tail[i];
+        if ((a === "--input" || a === "-i") && tail[i + 1]) {
+            inputs.push(tail[++i]);
+            continue;
+        }
+        if ((a === "--model-name" || a === "--register-name") && tail[i + 1]) {
+            registryModelName = tail[++i];
+            continue;
+        }
+        if (a === "--skip-model-name-resolve") {
+            skipModelNameResolve = true;
+            continue;
+        }
+        console.error(`Unknown flag for model small embeddings: ${a}`);
+        return 1;
+    }
+    if (inputs.length === 0) {
+        console.error("kweaver model small embeddings requires at least one -i / --input");
+        return 1;
+    }
+    const token = await ensureValidToken();
+    const mfManagerOpts = {
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+    };
+    let resolvedModelName = registryModelName.trim();
+    if (resolvedModelName.length === 0 && !skipModelNameResolve) {
+        try {
+            const raw = await getSmallModel({ ...mfManagerOpts, modelId });
+            resolvedModelName = smallGetRecordModelName(raw);
+        }
+        catch (e) {
+            console.error(e instanceof Error ? e.message : String(e));
+            return 1;
+        }
+    }
+    const data = await modelEmbeddings({
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfApiBaseUrl: g.mfApiBaseUrl,
+        modelId,
+        ...(resolvedModelName.length > 0 ? { modelName: resolvedModelName } : {}),
+        input: inputs,
+    });
+    console.log(formatCallOutput(JSON.stringify(data), g.pretty));
+    return 0;
+}
+async function runSmallRerank(g, tail) {
+    const modelId = tail[0];
+    if (!modelId || modelId.startsWith("-")) {
+        console.error('Usage: kweaver model small rerank <model_id> (-q|--query) <text> (-d|--document <text>) ... [--model-name <registry_model_name>] [--skip-model-name-resolve]');
+        return 1;
+    }
+    let query = "";
+    const documents = [];
+    let registryModelName = "";
+    let skipModelNameResolve = false;
+    for (let i = 1; i < tail.length; i += 1) {
+        const a = tail[i];
+        if ((a === "--query" || a === "-q") && tail[i + 1]) {
+            query = tail[++i];
+            continue;
+        }
+        if ((a === "--document" || a === "-d") && tail[i + 1]) {
+            documents.push(tail[++i]);
+            continue;
+        }
+        if ((a === "--model-name" || a === "--register-name") && tail[i + 1]) {
+            registryModelName = tail[++i];
+            continue;
+        }
+        if (a === "--skip-model-name-resolve") {
+            skipModelNameResolve = true;
+            continue;
+        }
+        console.error(`Unknown flag for model small rerank: ${a}`);
+        return 1;
+    }
+    if (!query) {
+        console.error("kweaver model small rerank requires -q / --query");
+        return 1;
+    }
+    if (documents.length === 0) {
+        console.error("kweaver model small rerank requires at least one -d / --document");
+        return 1;
+    }
+    const token = await ensureValidToken();
+    const mfManagerOpts = {
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfManagerBaseUrl: g.mfManagerBaseUrl,
+    };
+    let resolvedModelName = registryModelName.trim();
+    if (resolvedModelName.length === 0 && !skipModelNameResolve) {
+        try {
+            const raw = await getSmallModel({ ...mfManagerOpts, modelId });
+            resolvedModelName = smallGetRecordModelName(raw);
+        }
+        catch (e) {
+            console.error(e instanceof Error ? e.message : String(e));
+            return 1;
+        }
+    }
+    const data = await modelRerank({
+        baseUrl: token.baseUrl,
+        accessToken: token.accessToken,
+        businessDomain: g.businessDomain,
+        mfApiBaseUrl: g.mfApiBaseUrl,
+        modelId,
+        ...(resolvedModelName.length > 0 ? { modelName: resolvedModelName } : {}),
+        query,
+        documents,
+    });
+    console.log(formatCallOutput(JSON.stringify(data), g.pretty));
+    return 0;
+}
+export async function runModelCommand(args) {
+    const g = parseModelGlobalFlags(args);
+    const rest = g.rest;
+    if (rest.length === 0 || rest[0] === "--help" || rest[0] === "-h") {
+        printModelUsage();
+        return 0;
+    }
+    const branch = rest[0];
+    const action = rest[1];
+    const tail = rest.slice(2);
+    if (branch !== "llm" && branch !== "small") {
+        console.error(`Unknown model branch: ${branch} (expected llm or small)`);
+        printModelUsage();
+        return 1;
+    }
+    if (!action || action === "--help" || action === "-h") {
+        printModelUsage();
+        return 0;
+    }
+    if (action === "--template") {
+        if (tail.length > 0) {
+            console.error(`Usage: kweaver model ${branch} --template`);
+            return 1;
+        }
+        return printBundledModelBranchTemplate(branch, g);
+    }
+    const dispatch = async () => {
+        if (branch === "llm") {
+            if (action === "list")
+                return runLlmList(g, tail);
+            if (action === "get")
+                return runLlmGet(g, tail);
+            if (action === "add")
+                return runLlmAdd(g, tail);
+            if (action === "edit")
+                return runLlmEdit(g, tail);
+            if (action === "test")
+                return runLlmTest(g, tail);
+            if (action === "delete")
+                return runLlmDelete(g, tail);
+            if (action === "chat")
+                return runLlmChat(g, tail);
+        }
+        else {
+            if (action === "list")
+                return runSmallList(g, tail);
+            if (action === "get")
+                return runSmallGet(g, tail);
+            if (action === "add")
+                return runSmallAdd(g, tail);
+            if (action === "edit")
+                return runSmallEdit(g, tail);
+            if (action === "test")
+                return runSmallTest(g, tail);
+            if (action === "embeddings")
+                return runSmallEmbeddings(g, tail);
+            if (action === "rerank")
+                return runSmallRerank(g, tail);
+            if (action === "delete")
+                return runSmallDelete(g, tail);
+        }
+        console.error(`Unknown action: kweaver model ${branch} ${action}`);
+        printModelUsage();
+        return 1;
+    };
+    try {
+        return await with401RefreshRetry(async () => dispatch());
+    }
+    catch (e) {
+        console.error(formatHttpError(e));
+        return 1;
+    }
+}