iobroker.assistant 0.1.2

package/build/lib/llm.js

@@ -0,0 +1,303 @@
+"use strict";
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.LlmAgent = exports.PROVIDER_PRESETS = void 0;
+exports.resolveProvider = resolveProvider;
+const openai_1 = __importDefault(require("openai"));
+const sdk_1 = __importDefault(require("@anthropic-ai/sdk"));
+/** Per-provider SDK, endpoint and default model. Gemini/DeepSeek expose OpenAI-compatible endpoints. */
+exports.PROVIDER_PRESETS = {
+    openai: { sdk: 'openai', baseUrl: '', defaultModel: 'gpt-4o-mini' },
+    anthropic: { sdk: 'anthropic', baseUrl: '', defaultModel: 'claude-sonnet-4-6' },
+    gemini: {
+        sdk: 'openai',
+        baseUrl: 'https://generativelanguage.googleapis.com/v1beta/openai/',
+        defaultModel: 'gemini-2.0-flash',
+    },
+    deepseek: { sdk: 'openai', baseUrl: 'https://api.deepseek.com', defaultModel: 'deepseek-chat' },
+    custom: { sdk: 'openai', baseUrl: '', defaultModel: '' },
+};
+/** Resolve the effective SDK, base URL and default model for a provider (custom uses `configBaseUrl`). */
+function resolveProvider(provider, configBaseUrl) {
+    const preset = exports.PROVIDER_PRESETS[provider || 'openai'] || exports.PROVIDER_PRESETS.openai;
+    return {
+        sdk: preset.sdk,
+        baseUrl: provider === 'custom' ? configBaseUrl || '' : preset.baseUrl,
+        defaultModel: preset.defaultModel,
+    };
+}
+/**
+ * Provider-agnostic LLM agent with tool-calling.
+ *
+ * `ask(question)` runs the model, executes any tool calls it requests, feeds the
+ * results back, and repeats until the model returns a final text answer.
+ *
+ * The request/response plumbing of the vendor SDKs is intentionally loosely typed
+ * (`any`) so the code stays robust across minor SDK type changes; the public API
+ * (options, `ask`, tools) is fully typed.
+ */
+class LlmAgent {
+    provider;
+    model;
+    systemPrompt;
+    maxTokens;
+    tools;
+    toolMap;
+    log;
+    maxRounds = 6; // safety cap against tool-call loops
+    openai;
+    anthropic;
+    constructor(opts) {
+        this.provider = opts.provider === 'anthropic' ? 'anthropic' : 'openai';
+        this.model = opts.model;
+        this.systemPrompt = opts.systemPrompt || '';
+        this.maxTokens = opts.maxTokens || 1024;
+        this.tools = opts.tools || [];
+        this.log = opts.log;
+        this.toolMap = Object.fromEntries(this.tools.map(t => [t.name, t]));
+        if (this.provider === 'anthropic') {
+            this.anthropic = new sdk_1.default({ apiKey: opts.apiKey, baseURL: opts.baseUrl || undefined });
+        }
+        else {
+            this.openai = new openai_1.default({ apiKey: opts.apiKey, baseURL: opts.baseUrl || undefined });
+        }
+    }
+    /**
+     * `systemPrompt` overrides the configured one for this call (e.g. to inject a device list).
+     * `history` are prior turns (user/assistant) prepended so the model can resolve follow-ups.
+     */
+    async ask(question, systemPrompt, history = []) {
+        this.log.debug(`llm ask [${this.provider}/${this.model}] (${history.length} history): ${question}`);
+        const started = Date.now();
+        const sys = systemPrompt ?? this.systemPrompt;
+        const answer = this.provider === 'anthropic'
+            ? await this.askAnthropic(question, sys, history)
+            : await this.askOpenAI(question, sys, history);
+        this.log.debug(`llm answer (${Date.now() - started} ms): ${answer}`);
+        return answer;
+    }
+    /** Compact, truncated JSON for debug logging of LLM requests/responses. */
+    dump(label, payload) {
+        if (this.log.level !== 'silly' && this.log.level !== 'debug') {
+            return; // avoid stringifying large payloads when debug logging is off
+        }
+        const max = this.log.level === 'silly' ? 8000 : 1500;
+        let text;
+        try {
+            text = JSON.stringify(payload);
+        }
+        catch {
+            text = String(payload);
+        }
+        this.log.debug(`${label}: ${text.length > max ? `${text.slice(0, max)}…[${text.length}]` : text}`);
+    }
+    /** Translate a short device/room name into `targetLanguage` (single completion, no tools). */
+    async translate(text, targetLanguage) {
+        const trimmed = (text || '').trim();
+        if (!trimmed) {
+            return '';
+        }
+        const prompt = `Translate the following smart-home device or room name into ${targetLanguage}. ` +
+            `Reply with ONLY the translation — no quotes, no punctuation, no explanation.\n\nName: ${trimmed}`;
+        if (this.provider === 'anthropic') {
+            const client = this.anthropic;
+            const resp = await client.messages.create({
+                model: this.model,
+                max_tokens: 64,
+                messages: [{ role: 'user', content: prompt }],
+            });
+            return resp.content
+                .filter((b) => b.type === 'text')
+                .map((b) => b.text)
+                .join('')
+                .trim();
+        }
+        const client = this.openai;
+        const resp = await client.chat.completions.create({
+            model: this.model,
+            max_tokens: 64,
+            messages: [{ role: 'user', content: prompt }],
+        });
+        return String(resp.choices[0].message.content || '').trim();
+    }
+    /** Lightweight validation of the provider credentials (used by the settings Test button). */
+    async testConnection() {
+        try {
+            if (this.provider === 'anthropic') {
+                const client = this.anthropic;
+                await client.messages.create({
+                    model: this.model,
+                    max_tokens: 1,
+                    messages: [{ role: 'user', content: 'ping' }],
+                });
+            }
+            else {
+                const client = this.openai;
+                await client.models.list();
+            }
+            return { ok: true };
+        }
+        catch (e) {
+            return { ok: false, error: e.message };
+        }
+    }
+    /** List the provider's available model ids (for the settings autocomplete). */
+    async listModels() {
+        if (this.provider === 'anthropic') {
+            const client = this.anthropic;
+            const res = (await client.models.list());
+            return (res.data || []).map(m => m.id);
+        }
+        const client = this.openai;
+        const res = (await client.models.list());
+        // OpenAI lists many non-chat models (embeddings, tts, whisper, …) — filter to chat-capable-ish.
+        const EXCLUDE = /embedding|whisper|tts|audio|dall-e|moderation|image|realtime|transcribe|search|davinci|babbage|codex/i;
+        return (res.data || [])
+            .map(m => String(m.id))
+            .filter(id => !EXCLUDE.test(id))
+            .sort();
+    }
+    async runTool(name, args) {
+        const tool = this.toolMap[name];
+        if (!tool) {
+            return { error: `unknown tool: ${name}` };
+        }
+        try {
+            const result = await tool.run(args || {});
+            this.log.info(`tool ${name}(${JSON.stringify(args)}) → ${JSON.stringify(result).slice(0, 200)}`);
+            return result;
+        }
+        catch (e) {
+            this.log.warn(`tool ${name} failed: ${e.message}`);
+            return { error: e.message };
+        }
+    }
+    // ── OpenAI (Chat Completions + function tools) ──────────────────────────
+    async askOpenAI(question, systemPrompt, history = []) {
+        const client = this.openai;
+        const tools = this.tools.map(t => ({
+            type: 'function',
+            function: { name: t.name, description: t.description, parameters: t.parameters },
+        }));
+        const messages = [];
+        if (systemPrompt) {
+            messages.push({ role: 'system', content: systemPrompt });
+        }
+        for (const t of history) {
+            messages.push({ role: t.role, content: t.content });
+        }
+        messages.push({ role: 'user', content: question });
+        for (let round = 0; round < this.maxRounds; round++) {
+            const request = {
+                model: this.model,
+                messages,
+                tools: tools.length ? tools : undefined,
+                max_tokens: this.maxTokens,
+            };
+            this.dump(`llm → openai round ${round}`, request);
+            const resp = await client.chat.completions.create(request);
+            const msg = resp.choices[0].message;
+            this.dump(`llm ← openai round ${round}`, { message: msg, usage: resp.usage });
+            messages.push(msg);
+            if (!msg.tool_calls || !msg.tool_calls.length) {
+                return String(msg.content || '').trim();
+            }
+            for (const tc of msg.tool_calls) {
+                let args = {};
+                try {
+                    args = JSON.parse(tc.function.arguments || '{}');
+                }
+                catch {
+                    /* leave empty */
+                }
+                const result = await this.runTool(tc.function.name, args);
+                messages.push({ role: 'tool', tool_call_id: tc.id, content: JSON.stringify(result) });
+            }
+        }
+        return 'Ich konnte die Anfrage nicht abschließen (zu viele Tool-Schritte).';
+    }
+    /**
+     * Keep a single cache breakpoint on the last content block of the last message (Anthropic allows
+     * ≤4 total; we also mark system + last tool). Strips prior message breakpoints so we never exceed it.
+     */
+    markLastMessageForCache(messages) {
+        for (const m of messages) {
+            if (Array.isArray(m.content)) {
+                for (const b of m.content) {
+                    if (b && typeof b === 'object') {
+                        delete b.cache_control;
+                    }
+                }
+            }
+        }
+        const last = messages[messages.length - 1];
+        if (!last) {
+            return;
+        }
+        if (typeof last.content === 'string') {
+            last.content = [{ type: 'text', text: last.content, cache_control: { type: 'ephemeral' } }];
+        }
+        else if (Array.isArray(last.content) && last.content.length) {
+            const b = last.content[last.content.length - 1];
+            if (b && typeof b === 'object') {
+                b.cache_control = { type: 'ephemeral' };
+            }
+        }
+    }
+    // ── Anthropic (Messages API + tool_use) ─────────────────────────────────
+    async askAnthropic(question, systemPrompt, history = []) {
+        const client = this.anthropic;
+        const tools = this.tools.map(t => ({
+            name: t.name,
+            description: t.description,
+            input_schema: t.parameters,
+        }));
+        // Prompt caching: mark the last tool → the static prefix (system + all tools) is cached and
+        // read back on every later round/request (5-min TTL), cutting latency & cost on the big context.
+        if (tools.length) {
+            tools[tools.length - 1] = { ...tools[tools.length - 1], cache_control: { type: 'ephemeral' } };
+        }
+        const system = systemPrompt
+            ? [{ type: 'text', text: systemPrompt, cache_control: { type: 'ephemeral' } }]
+            : undefined;
+        const messages = [
+            ...history.map(t => ({ role: t.role, content: t.content })),
+            { role: 'user', content: question },
+        ];
+        for (let round = 0; round < this.maxRounds; round++) {
+            // Cache the growing conversation prefix too (e.g. the large list_devices result): move a
+            // single cache breakpoint to the last content block of the last message each round.
+            this.markLastMessageForCache(messages);
+            const request = {
+                model: this.model,
+                max_tokens: this.maxTokens,
+                system,
+                tools: tools.length ? tools : undefined,
+                messages,
+            };
+            this.dump(`llm → anthropic round ${round}`, request);
+            const resp = await client.messages.create(request);
+            this.dump(`llm ← anthropic round ${round}`, { content: resp.content, usage: resp.usage });
+            const toolUses = resp.content.filter((b) => b.type === 'tool_use');
+            if (!toolUses.length) {
+                return resp.content
+                    .filter((b) => b.type === 'text')
+                    .map((b) => b.text)
+                    .join('')
+                    .trim();
+            }
+            messages.push({ role: 'assistant', content: resp.content });
+            const toolResults = [];
+            for (const tu of toolUses) {
+                const result = await this.runTool(tu.name, tu.input);
+                toolResults.push({ type: 'tool_result', tool_use_id: tu.id, content: JSON.stringify(result) });
+            }
+            messages.push({ role: 'user', content: toolResults });
+        }
+        return 'Ich konnte die Anfrage nicht abschließen (zu viele Tool-Schritte).';
+    }
+}
+exports.LlmAgent = LlmAgent;
+//# sourceMappingURL=llm.js.map

package/build/lib/llm.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"llm.js","sourceRoot":"","sources":["../../src/lib/llm.ts"],"names":[],"mappings":";;;;;;AA8BA,0CAUC;AAxCD,oDAA4B;AAC5B,4DAA0C;AAe1C,wGAAwG;AAC3F,QAAA,gBAAgB,GAAqC;IAC9D,MAAM,EAAE,EAAE,GAAG,EAAE,QAAQ,EAAE,OAAO,EAAE,EAAE,EAAE,YAAY,EAAE,aAAa,EAAE;IACnE,SAAS,EAAE,EAAE,GAAG,EAAE,WAAW,EAAE,OAAO,EAAE,EAAE,EAAE,YAAY,EAAE,mBAAmB,EAAE;IAC/E,MAAM,EAAE;QACJ,GAAG,EAAE,QAAQ;QACb,OAAO,EAAE,0DAA0D;QACnE,YAAY,EAAE,kBAAkB;KACnC;IACD,QAAQ,EAAE,EAAE,GAAG,EAAE,QAAQ,EAAE,OAAO,EAAE,0BAA0B,EAAE,YAAY,EAAE,eAAe,EAAE;IAC/F,MAAM,EAAE,EAAE,GAAG,EAAE,QAAQ,EAAE,OAAO,EAAE,EAAE,EAAE,YAAY,EAAE,EAAE,EAAE;CAC3D,CAAC;AAEF,0GAA0G;AAC1G,SAAgB,eAAe,CAC3B,QAA4B,EAC5B,aAAsB;IAEtB,MAAM,MAAM,GAAG,wBAAgB,CAAE,QAAqB,IAAI,QAAQ,CAAC,IAAI,wBAAgB,CAAC,MAAM,CAAC;IAC/F,OAAO;QACH,GAAG,EAAE,MAAM,CAAC,GAAG;QACf,OAAO,EAAE,QAAQ,KAAK,QAAQ,CAAC,CAAC,CAAC,aAAa,IAAI,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO;QACrE,YAAY,EAAE,MAAM,CAAC,YAAY;KACpC,CAAC;AACN,CAAC;AAaD;;;;;;;;;GASG;AACH,MAAa,QAAQ;IACD,QAAQ,CAAyB;IACjC,KAAK,CAAS;IAEb,YAAY,CAAS;IACrB,SAAS,CAAS;IAClB,KAAK,CAAS;IACd,OAAO,CAAuB;IAC9B,GAAG,CAAkB;IACrB,SAAS,GAAG,CAAC,CAAC,CAAC,qCAAqC;IAEpD,MAAM,CAAU;IAChB,SAAS,CAAa;IAEvC,YAAY,IAAqB;QAC7B,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,KAAK,WAAW,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,QAAQ,CAAC;QACvE,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;QACxB,IAAI,CAAC,YAAY,GAAG,IAAI,CAAC,YAAY,IAAI,EAAE,CAAC;QAC5C,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,IAAI,CAAC;QACxC,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,EAAE,CAAC;QAC9B,IAAI,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC;QACpB,IAAI,CAAC,OAAO,GAAG,MAAM,CAAC,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;QAEpE,IAAI,IAAI,CAAC,QAAQ,KAAK,WAAW,EAAE,CAAC;YAChC,IAAI,CAAC,SAAS,GAAG,IAAI,aAAS,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,IAAI,SAAS,EAAE,CAAC,CAAC;QAChG,CAAC;aAAM,CAAC;YACJ,IAAI,CAAC,MAAM,GAAG,IAAI,gBAAM,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,IAAI,SAAS,EAAE,CAAC,CAAC;QAC1F,CAAC;IACL,CAAC;IAED;;;OAGG;IACI,KAAK,CAAC,GAAG,CACZ,QAAgB,EAChB,YAAqB,EACrB,UAA6D,EAAE;QAE/D,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,YAAY,IAAI,CAAC,QAAQ,IAAI,IAAI,CAAC,KAAK,MAAM,OAAO,CAAC,MAAM,cAAc,QAAQ,EAAE,CAAC,CAAC;QACpG,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC3B,MAAM,GAAG,GAAG,YAAY,IAAI,IAAI,CAAC,YAAY,CAAC;QAC9C,MAAM,MAAM,GACR,IAAI,CAAC,QAAQ,KAAK,WAAW;YACzB,CAAC,CAAC,MAAM,IAAI,CAAC,YAAY,CAAC,QAAQ,EAAE,GAAG,EAAE,OAAO,CAAC;YACjD,CAAC,CAAC,MAAM,IAAI,CAAC,SAAS,CAAC,QAAQ,EAAE,GAAG,EAAE,OAAO,CAAC,CAAC;QACvD,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,eAAe,IAAI,CAAC,GAAG,EAAE,GAAG,OAAO,SAAS,MAAM,EAAE,CAAC,CAAC;QACrE,OAAO,MAAM,CAAC;IAClB,CAAC;IAED,2EAA2E;IACnE,IAAI,CAAC,KAAa,EAAE,OAAgB;QACxC,IAAI,IAAI,CAAC,GAAG,CAAC,KAAK,KAAK,OAAO,IAAI,IAAI,CAAC,GAAG,CAAC,KAAK,KAAK,OAAO,EAAE,CAAC;YAC3D,OAAO,CAAC,8DAA8D;QAC1E,CAAC;QACD,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,KAAK,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC;QACrD,IAAI,IAAY,CAAC;QACjB,IAAI,CAAC;YACD,IAAI,GAAG,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC;QACnC,CAAC;QAAC,MAAM,CAAC;YACL,IAAI,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC;QAC3B,CAAC;QACD,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,KAAK,KAAK,IAAI,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,KAAK,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;IACvG,CAAC;IAED,8FAA8F;IACvF,KAAK,CAAC,SAAS,CAAC,IAAY,EAAE,cAAsB;QACvD,MAAM,OAAO,GAAG,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACpC,IAAI,CAAC,OAAO,EAAE,CAAC;YACX,OAAO,EAAE,CAAC;QACd,CAAC;QACD,MAAM,MAAM,GACR,+DAA+D,cAAc,IAAI;YACjF,yFAAyF,OAAO,EAAE,CAAC;QACvG,IAAI,IAAI,CAAC,QAAQ,KAAK,WAAW,EAAE,CAAC;YAChC,MAAM,MAAM,GAAG,IAAI,CAAC,SAAsB,CAAC;YAC3C,MAAM,IAAI,GAAQ,MAAM,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC;gBAC3C,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,UAAU,EAAE,EAAE;gBACd,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC;aAChD,CAAC,CAAC;YACH,OAAO,IAAI,CAAC,OAAO;iBACd,MAAM,CAAC,CAAC,CAAM,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC;iBACrC,GAAG,CAAC,CAAC,CAAM,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;iBACvB,IAAI,CAAC,EAAE,CAAC;iBACR,IAAI,EAAE,CAAC;QAChB,CAAC;QACD,MAAM,MAAM,GAAG,IAAI,CAAC,MAAgB,CAAC;QACrC,MAAM,IAAI,GAAQ,MAAM,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,MAAM,CAAC;YACnD,KAAK,EAAE,IAAI,CAAC,KAAK;YACjB,UAAU,EAAE,EAAE;YACd,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC;SAChD,CAAC,CAAC;QACH,OAAO,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IAChE,CAAC;IAED,6FAA6F;IACtF,KAAK,CAAC,cAAc;QACvB,IAAI,CAAC;YACD,IAAI,IAAI,CAAC,QAAQ,KAAK,WAAW,EAAE,CAAC;gBAChC,MAAM,MAAM,GAAG,IAAI,CAAC,SAAsB,CAAC;gBAC3C,MAAM,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC;oBACzB,KAAK,EAAE,IAAI,CAAC,KAAK;oBACjB,UAAU,EAAE,CAAC;oBACb,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC;iBAChD,CAAC,CAAC;YACP,CAAC;iBAAM,CAAC;gBACJ,MAAM,MAAM,GAAG,IAAI,CAAC,MAAgB,CAAC;gBACrC,MAAM,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;YAC/B,CAAC;YACD,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC;QACxB,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACT,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAG,CAAW,CAAC,OAAO,EAAE,CAAC;QACtD,CAAC;IACL,CAAC;IAED,+EAA+E;IACxE,KAAK,CAAC,UAAU;QACnB,IAAI,IAAI,CAAC,QAAQ,KAAK,WAAW,EAAE,CAAC;YAChC,MAAM,MAAM,GAAG,IAAI,CAAC,SAAsB,CAAC;YAC3C,MAAM,GAAG,GAAG,CAAC,MAAM,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAA2C,CAAC;YACnF,OAAO,CAAC,GAAG,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QAC3C,CAAC;QACD,MAAM,MAAM,GAAG,IAAI,CAAC,MAAgB,CAAC;QACrC,MAAM,GAAG,GAAG,CAAC,MAAM,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAA2C,CAAC;QACnF,gGAAgG;QAChG,MAAM,OAAO,GACT,uGAAuG,CAAC;QAC5G,OAAO,CAAC,GAAG,CAAC,IAAI,IAAI,EAAE,CAAC;aAClB,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;aACtB,MAAM,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;aAC/B,IAAI,EAAE,CAAC;IAChB,CAAC;IAEO,KAAK,CAAC,OAAO,CAAC,IAAY,EAAE,IAA6B;QAC7D,MAAM,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QAChC,IAAI,CAAC,IAAI,EAAE,CAAC;YACR,OAAO,EAAE,KAAK,EAAE,iBAAiB,IAAI,EAAE,EAAE,CAAC;QAC9C,CAAC;QACD,IAAI,CAAC;YACD,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,GAAG,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC;YAC1C,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,IAAI,IAAI,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,OAAO,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;YACjG,OAAO,MAAM,CAAC;QAClB,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACT,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,IAAI,YAAa,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;YAC9D,OAAO,EAAE,KAAK,EAAG,CAAW,CAAC,OAAO,EAAE,CAAC;QAC3C,CAAC;IACL,CAAC;IAED,2EAA2E;IACnE,KAAK,CAAC,SAAS,CACnB,QAAgB,EAChB,YAAoB,EACpB,UAA6D,EAAE;QAE/D,MAAM,MAAM,GAAG,IAAI,CAAC,MAAgB,CAAC;QACrC,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;YAC/B,IAAI,EAAE,UAAmB;YACzB,QAAQ,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,WAAW,EAAE,CAAC,CAAC,WAAW,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE;SACnF,CAAC,CAAC,CAAC;QAEJ,MAAM,QAAQ,GAAU,EAAE,CAAC;QAC3B,IAAI,YAAY,EAAE,CAAC;YACf,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,YAAY,EAAE,CAAC,CAAC;QAC7D,CAAC;QACD,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;YACtB,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;QACxD,CAAC;QACD,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC,CAAC;QAEnD,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,SAAS,EAAE,KAAK,EAAE,EAAE,CAAC;YAClD,MAAM,OAAO,GAAG;gBACZ,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,QAAQ;gBACR,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS;gBACvC,UAAU,EAAE,IAAI,CAAC,SAAS;aAC7B,CAAC;YACF,IAAI,CAAC,IAAI,CAAC,sBAAsB,KAAK,EAAE,EAAE,OAAO,CAAC,CAAC;YAClD,MAAM,IAAI,GAAQ,MAAM,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YAChE,MAAM,GAAG,GAAG,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;YACpC,IAAI,CAAC,IAAI,CAAC,sBAAsB,KAAK,EAAE,EAAE,EAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;YAC9E,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YAEnB,IAAI,CAAC,GAAG,CAAC,UAAU,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,MAAM,EAAE,CAAC;gBAC5C,OAAO,MAAM,CAAC,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;YAC5C,CAAC;YAED,KAAK,MAAM,EAAE,IAAI,GAAG,CAAC,UAAU,EAAE,CAAC;gBAC9B,IAAI,IAAI,GAA4B,EAAE,CAAC;gBACvC,IAAI,CAAC;oBACD,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,IAAI,IAAI,CAAC,CAAC;gBACrD,CAAC;gBAAC,MAAM,CAAC;oBACL,iBAAiB;gBACrB,CAAC;gBACD,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,QAAQ,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;gBAC1D,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,YAAY,EAAE,EAAE,CAAC,EAAE,EAAE,OAAO,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;YAC1F,CAAC;QACL,CAAC;QACD,OAAO,oEAAoE,CAAC;IAChF,CAAC;IAED;;;OAGG;IACK,uBAAuB,CAAC,QAAe;QAC3C,KAAK,MAAM,CAAC,IAAI,QAAQ,EAAE,CAAC;YACvB,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;gBAC3B,KAAK,MAAM,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,CAAC;oBACxB,IAAI,CAAC,IAAI,OAAO,CAAC,KAAK,QAAQ,EAAE,CAAC;wBAC7B,OAAO,CAAC,CAAC,aAAa,CAAC;oBAC3B,CAAC;gBACL,CAAC;YACL,CAAC;QACL,CAAC;QACD,MAAM,IAAI,GAAG,QAAQ,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QAC3C,IAAI,CAAC,IAAI,EAAE,CAAC;YACR,OAAO;QACX,CAAC;QACD,IAAI,OAAO,IAAI,CAAC,OAAO,KAAK,QAAQ,EAAE,CAAC;YACnC,IAAI,CAAC,OAAO,GAAG,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,CAAC,OAAO,EAAE,aAAa,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,EAAE,CAAC,CAAC;QAChG,CAAC;aAAM,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC;YAC5D,MAAM,CAAC,GAAG,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;YAChD,IAAI,CAAC,IAAI,OAAO,CAAC,KAAK,QAAQ,EAAE,CAAC;gBAC7B,CAAC,CAAC,aAAa,GAAG,EAAE,IAAI,EAAE,WAAW,EAAE,CAAC;YAC5C,CAAC;QACL,CAAC;IACL,CAAC;IAED,2EAA2E;IACnE,KAAK,CAAC,YAAY,CACtB,QAAgB,EAChB,YAAoB,EACpB,UAA6D,EAAE;QAE/D,MAAM,MAAM,GAAG,IAAI,CAAC,SAAsB,CAAC;QAC3C,MAAM,KAAK,GAAU,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;YACtC,IAAI,EAAE,CAAC,CAAC,IAAI;YACZ,WAAW,EAAE,CAAC,CAAC,WAAW;YAC1B,YAAY,EAAE,CAAC,CAAC,UAAwC;SAC3D,CAAC,CAAC,CAAC;QACJ,4FAA4F;QAC5F,iGAAiG;QACjG,IAAI,KAAK,CAAC,MAAM,EAAE,CAAC;YACf,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,EAAE,aAAa,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,EAAE,CAAC;QACnG,CAAC;QACD,MAAM,MAAM,GAAQ,YAAY;YAC5B,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,YAAY,EAAE,aAAa,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,EAAE,CAAC;YAC9E,CAAC,CAAC,SAAS,CAAC;QAEhB,MAAM,QAAQ,GAAU;YACpB,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;YAC3D,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE;SACtC,CAAC;QAEF,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,SAAS,EAAE,KAAK,EAAE,EAAE,CAAC;YAClD,yFAAyF;YACzF,oFAAoF;YACpF,IAAI,CAAC,uBAAuB,CAAC,QAAQ,CAAC,CAAC;YACvC,MAAM,OAAO,GAAG;gBACZ,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,UAAU,EAAE,IAAI,CAAC,SAAS;gBAC1B,MAAM;gBACN,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS;gBACvC,QAAQ;aACX,CAAC;YACF,IAAI,CAAC,IAAI,CAAC,yBAAyB,KAAK,EAAE,EAAE,OAAO,CAAC,CAAC;YACrD,MAAM,IAAI,GAAQ,MAAM,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YACxD,IAAI,CAAC,IAAI,CAAC,yBAAyB,KAAK,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;YAE1F,MAAM,QAAQ,GAAG,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAM,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,UAAU,CAAC,CAAC;YACxE,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC;gBACnB,OAAO,IAAI,CAAC,OAAO;qBACd,MAAM,CAAC,CAAC,CAAM,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC;qBACrC,GAAG,CAAC,CAAC,CAAM,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;qBACvB,IAAI,CAAC,EAAE,CAAC;qBACR,IAAI,EAAE,CAAC;YAChB,CAAC;YAED,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,WAAW,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;YAC5D,MAAM,WAAW,GAAG,EAAE,CAAC;YACvB,KAAK,MAAM,EAAE,IAAI,QAAQ,EAAE,CAAC;gBACxB,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,KAAK,CAAC,CAAC;gBACrD,WAAW,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,aAAa,EAAE,WAAW,EAAE,EAAE,CAAC,EAAE,EAAE,OAAO,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;YACnG,CAAC;YACD,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,EAAE,CAAC,CAAC;QAC1D,CAAC;QACD,OAAO,oEAAoE,CAAC;IAChF,CAAC;CACJ;AAjSD,4BAiSC","sourcesContent":["import OpenAI from 'openai';\nimport Anthropic from '@anthropic-ai/sdk';\nimport type { Tool } from './tools';\n\n/** All selectable LLM providers. gemini/deepseek/custom all use the OpenAI-compatible SDK path. */\nexport type Provider = 'openai' | 'anthropic' | 'gemini' | 'deepseek' | 'custom';\n\ninterface ProviderPreset {\n    /** Which vendor SDK to use. */\n    sdk: 'openai' | 'anthropic';\n    /** Fixed API base URL ('' = SDK default; for 'custom' the user's baseUrl is used instead). */\n    baseUrl: string;\n    /** Default model when none is configured. */\n    defaultModel: string;\n}\n\n/** Per-provider SDK, endpoint and default model. Gemini/DeepSeek expose OpenAI-compatible endpoints. */\nexport const PROVIDER_PRESETS: Record<Provider, ProviderPreset> = {\n    openai: { sdk: 'openai', baseUrl: '', defaultModel: 'gpt-4o-mini' },\n    anthropic: { sdk: 'anthropic', baseUrl: '', defaultModel: 'claude-sonnet-4-6' },\n    gemini: {\n        sdk: 'openai',\n        baseUrl: 'https://generativelanguage.googleapis.com/v1beta/openai/',\n        defaultModel: 'gemini-2.0-flash',\n    },\n    deepseek: { sdk: 'openai', baseUrl: 'https://api.deepseek.com', defaultModel: 'deepseek-chat' },\n    custom: { sdk: 'openai', baseUrl: '', defaultModel: '' },\n};\n\n/** Resolve the effective SDK, base URL and default model for a provider (custom uses `configBaseUrl`). */\nexport function resolveProvider(\n    provider: string | undefined,\n    configBaseUrl?: string,\n): { sdk: 'openai' | 'anthropic'; baseUrl: string; defaultModel: string } {\n    const preset = PROVIDER_PRESETS[(provider as Provider) || 'openai'] || PROVIDER_PRESETS.openai;\n    return {\n        sdk: preset.sdk,\n        baseUrl: provider === 'custom' ? configBaseUrl || '' : preset.baseUrl,\n        defaultModel: preset.defaultModel,\n    };\n}\n\nexport interface LlmAgentOptions {\n    provider: Provider;\n    apiKey: string;\n    model: string;\n    baseUrl?: string;\n    systemPrompt?: string;\n    maxTokens?: number;\n    tools: Tool[];\n    log: ioBroker.Logger;\n}\n\n/**\n * Provider-agnostic LLM agent with tool-calling.\n *\n * `ask(question)` runs the model, executes any tool calls it requests, feeds the\n * results back, and repeats until the model returns a final text answer.\n *\n * The request/response plumbing of the vendor SDKs is intentionally loosely typed\n * (`any`) so the code stays robust across minor SDK type changes; the public API\n * (options, `ask`, tools) is fully typed.\n */\nexport class LlmAgent {\n    public readonly provider: 'openai' | 'anthropic';\n    public readonly model: string;\n\n    private readonly systemPrompt: string;\n    private readonly maxTokens: number;\n    private readonly tools: Tool[];\n    private readonly toolMap: Record<string, Tool>;\n    private readonly log: ioBroker.Logger;\n    private readonly maxRounds = 6; // safety cap against tool-call loops\n\n    private readonly openai?: OpenAI;\n    private readonly anthropic?: Anthropic;\n\n    constructor(opts: LlmAgentOptions) {\n        this.provider = opts.provider === 'anthropic' ? 'anthropic' : 'openai';\n        this.model = opts.model;\n        this.systemPrompt = opts.systemPrompt || '';\n        this.maxTokens = opts.maxTokens || 1024;\n        this.tools = opts.tools || [];\n        this.log = opts.log;\n        this.toolMap = Object.fromEntries(this.tools.map(t => [t.name, t]));\n\n        if (this.provider === 'anthropic') {\n            this.anthropic = new Anthropic({ apiKey: opts.apiKey, baseURL: opts.baseUrl || undefined });\n        } else {\n            this.openai = new OpenAI({ apiKey: opts.apiKey, baseURL: opts.baseUrl || undefined });\n        }\n    }\n\n    /**\n     * `systemPrompt` overrides the configured one for this call (e.g. to inject a device list).\n     * `history` are prior turns (user/assistant) prepended so the model can resolve follow-ups.\n     */\n    public async ask(\n        question: string,\n        systemPrompt?: string,\n        history: { role: 'user' | 'assistant'; content: string }[] = [],\n    ): Promise<string> {\n        this.log.debug(`llm ask [${this.provider}/${this.model}] (${history.length} history): ${question}`);\n        const started = Date.now();\n        const sys = systemPrompt ?? this.systemPrompt;\n        const answer =\n            this.provider === 'anthropic'\n                ? await this.askAnthropic(question, sys, history)\n                : await this.askOpenAI(question, sys, history);\n        this.log.debug(`llm answer (${Date.now() - started} ms): ${answer}`);\n        return answer;\n    }\n\n    /** Compact, truncated JSON for debug logging of LLM requests/responses. */\n    private dump(label: string, payload: unknown): void {\n        if (this.log.level !== 'silly' && this.log.level !== 'debug') {\n            return; // avoid stringifying large payloads when debug logging is off\n        }\n        const max = this.log.level === 'silly' ? 8000 : 1500;\n        let text: string;\n        try {\n            text = JSON.stringify(payload);\n        } catch {\n            text = String(payload);\n        }\n        this.log.debug(`${label}: ${text.length > max ? `${text.slice(0, max)}…[${text.length}]` : text}`);\n    }\n\n    /** Translate a short device/room name into `targetLanguage` (single completion, no tools). */\n    public async translate(text: string, targetLanguage: string): Promise<string> {\n        const trimmed = (text || '').trim();\n        if (!trimmed) {\n            return '';\n        }\n        const prompt =\n            `Translate the following smart-home device or room name into ${targetLanguage}. ` +\n            `Reply with ONLY the translation — no quotes, no punctuation, no explanation.\\n\\nName: ${trimmed}`;\n        if (this.provider === 'anthropic') {\n            const client = this.anthropic as Anthropic;\n            const resp: any = await client.messages.create({\n                model: this.model,\n                max_tokens: 64,\n                messages: [{ role: 'user', content: prompt }],\n            });\n            return resp.content\n                .filter((b: any) => b.type === 'text')\n                .map((b: any) => b.text)\n                .join('')\n                .trim();\n        }\n        const client = this.openai as OpenAI;\n        const resp: any = await client.chat.completions.create({\n            model: this.model,\n            max_tokens: 64,\n            messages: [{ role: 'user', content: prompt }],\n        });\n        return String(resp.choices[0].message.content || '').trim();\n    }\n\n    /** Lightweight validation of the provider credentials (used by the settings Test button). */\n    public async testConnection(): Promise<{ ok: boolean; error?: string }> {\n        try {\n            if (this.provider === 'anthropic') {\n                const client = this.anthropic as Anthropic;\n                await client.messages.create({\n                    model: this.model,\n                    max_tokens: 1,\n                    messages: [{ role: 'user', content: 'ping' }],\n                });\n            } else {\n                const client = this.openai as OpenAI;\n                await client.models.list();\n            }\n            return { ok: true };\n        } catch (e) {\n            return { ok: false, error: (e as Error).message };\n        }\n    }\n\n    /** List the provider's available model ids (for the settings autocomplete). */\n    public async listModels(): Promise<string[]> {\n        if (this.provider === 'anthropic') {\n            const client = this.anthropic as Anthropic;\n            const res = (await client.models.list()) as unknown as { data?: { id: string }[] };\n            return (res.data || []).map(m => m.id);\n        }\n        const client = this.openai as OpenAI;\n        const res = (await client.models.list()) as unknown as { data?: { id: string }[] };\n        // OpenAI lists many non-chat models (embeddings, tts, whisper, …) — filter to chat-capable-ish.\n        const EXCLUDE =\n            /embedding|whisper|tts|audio|dall-e|moderation|image|realtime|transcribe|search|davinci|babbage|codex/i;\n        return (res.data || [])\n            .map(m => String(m.id))\n            .filter(id => !EXCLUDE.test(id))\n            .sort();\n    }\n\n    private async runTool(name: string, args: Record<string, unknown>): Promise<unknown> {\n        const tool = this.toolMap[name];\n        if (!tool) {\n            return { error: `unknown tool: ${name}` };\n        }\n        try {\n            const result = await tool.run(args || {});\n            this.log.info(`tool ${name}(${JSON.stringify(args)}) → ${JSON.stringify(result).slice(0, 200)}`);\n            return result;\n        } catch (e) {\n            this.log.warn(`tool ${name} failed: ${(e as Error).message}`);\n            return { error: (e as Error).message };\n        }\n    }\n\n    // ── OpenAI (Chat Completions + function tools) ──────────────────────────\n    private async askOpenAI(\n        question: string,\n        systemPrompt: string,\n        history: { role: 'user' | 'assistant'; content: string }[] = [],\n    ): Promise<string> {\n        const client = this.openai as OpenAI;\n        const tools = this.tools.map(t => ({\n            type: 'function' as const,\n            function: { name: t.name, description: t.description, parameters: t.parameters },\n        }));\n\n        const messages: any[] = [];\n        if (systemPrompt) {\n            messages.push({ role: 'system', content: systemPrompt });\n        }\n        for (const t of history) {\n            messages.push({ role: t.role, content: t.content });\n        }\n        messages.push({ role: 'user', content: question });\n\n        for (let round = 0; round < this.maxRounds; round++) {\n            const request = {\n                model: this.model,\n                messages,\n                tools: tools.length ? tools : undefined,\n                max_tokens: this.maxTokens,\n            };\n            this.dump(`llm → openai round ${round}`, request);\n            const resp: any = await client.chat.completions.create(request);\n            const msg = resp.choices[0].message;\n            this.dump(`llm ← openai round ${round}`, { message: msg, usage: resp.usage });\n            messages.push(msg);\n\n            if (!msg.tool_calls || !msg.tool_calls.length) {\n                return String(msg.content || '').trim();\n            }\n\n            for (const tc of msg.tool_calls) {\n                let args: Record<string, unknown> = {};\n                try {\n                    args = JSON.parse(tc.function.arguments || '{}');\n                } catch {\n                    /* leave empty */\n                }\n                const result = await this.runTool(tc.function.name, args);\n                messages.push({ role: 'tool', tool_call_id: tc.id, content: JSON.stringify(result) });\n            }\n        }\n        return 'Ich konnte die Anfrage nicht abschließen (zu viele Tool-Schritte).';\n    }\n\n    /**\n     * Keep a single cache breakpoint on the last content block of the last message (Anthropic allows\n     * ≤4 total; we also mark system + last tool). Strips prior message breakpoints so we never exceed it.\n     */\n    private markLastMessageForCache(messages: any[]): void {\n        for (const m of messages) {\n            if (Array.isArray(m.content)) {\n                for (const b of m.content) {\n                    if (b && typeof b === 'object') {\n                        delete b.cache_control;\n                    }\n                }\n            }\n        }\n        const last = messages[messages.length - 1];\n        if (!last) {\n            return;\n        }\n        if (typeof last.content === 'string') {\n            last.content = [{ type: 'text', text: last.content, cache_control: { type: 'ephemeral' } }];\n        } else if (Array.isArray(last.content) && last.content.length) {\n            const b = last.content[last.content.length - 1];\n            if (b && typeof b === 'object') {\n                b.cache_control = { type: 'ephemeral' };\n            }\n        }\n    }\n\n    // ── Anthropic (Messages API + tool_use) ─────────────────────────────────\n    private async askAnthropic(\n        question: string,\n        systemPrompt: string,\n        history: { role: 'user' | 'assistant'; content: string }[] = [],\n    ): Promise<string> {\n        const client = this.anthropic as Anthropic;\n        const tools: any[] = this.tools.map(t => ({\n            name: t.name,\n            description: t.description,\n            input_schema: t.parameters as Anthropic.Tool.InputSchema,\n        }));\n        // Prompt caching: mark the last tool → the static prefix (system + all tools) is cached and\n        // read back on every later round/request (5-min TTL), cutting latency & cost on the big context.\n        if (tools.length) {\n            tools[tools.length - 1] = { ...tools[tools.length - 1], cache_control: { type: 'ephemeral' } };\n        }\n        const system: any = systemPrompt\n            ? [{ type: 'text', text: systemPrompt, cache_control: { type: 'ephemeral' } }]\n            : undefined;\n\n        const messages: any[] = [\n            ...history.map(t => ({ role: t.role, content: t.content })),\n            { role: 'user', content: question },\n        ];\n\n        for (let round = 0; round < this.maxRounds; round++) {\n            // Cache the growing conversation prefix too (e.g. the large list_devices result): move a\n            // single cache breakpoint to the last content block of the last message each round.\n            this.markLastMessageForCache(messages);\n            const request = {\n                model: this.model,\n                max_tokens: this.maxTokens,\n                system,\n                tools: tools.length ? tools : undefined,\n                messages,\n            };\n            this.dump(`llm → anthropic round ${round}`, request);\n            const resp: any = await client.messages.create(request);\n            this.dump(`llm ← anthropic round ${round}`, { content: resp.content, usage: resp.usage });\n\n            const toolUses = resp.content.filter((b: any) => b.type === 'tool_use');\n            if (!toolUses.length) {\n                return resp.content\n                    .filter((b: any) => b.type === 'text')\n                    .map((b: any) => b.text)\n                    .join('')\n                    .trim();\n            }\n\n            messages.push({ role: 'assistant', content: resp.content });\n            const toolResults = [];\n            for (const tu of toolUses) {\n                const result = await this.runTool(tu.name, tu.input);\n                toolResults.push({ type: 'tool_result', tool_use_id: tu.id, content: JSON.stringify(result) });\n            }\n            messages.push({ role: 'user', content: toolResults });\n        }\n        return 'Ich konnte die Anfrage nicht abschließen (zu viele Tool-Schritte).';\n    }\n}\n"]}

package/build/lib/localLlm.js

@@ -0,0 +1,221 @@
+"use strict";
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.LocalLlm = exports.DEFAULT_LOCAL_MODEL_URL = exports.HANDOFF = void 0;
+exports.isLocalLlmInstalled = isLocalLlmInstalled;
+exports.isModelDownloaded = isModelDownloaded;
+exports.installLocalLlm = installLocalLlm;
+exports.isHandoff = isHandoff;
+const node_child_process_1 = require("node:child_process");
+const node_module_1 = require("node:module");
+const node_url_1 = require("node:url");
+const path = __importStar(require("node:path"));
+const fs = __importStar(require("node:fs"));
+/**
+ * Tier-1a local LLM via `node-llama-cpp` — installed **on demand** (not a package.json dependency, so
+ * users who don't enable it never pay the large native download). Everything lives in the writable
+ * instance data dir (`<iobroker-data>/assistant.<n>/`), which survives adapter upgrades; on a fresh
+ * upgrade the module is simply re-installed on request.
+ *
+ * The model runs **tool-free**: it answers general questions and, when a request needs live ioBroker
+ * state or it is unsure, it emits the {@link HANDOFF} sentinel so the orchestrator escalates to the
+ * cloud LLM (which has the tools). Small models handle a "say HANDOFF" rule far more reliably than
+ * real tool-calling.
+ */
+/** Sentinel the local model returns to escalate a request to the cloud LLM. */
+exports.HANDOFF = 'HANDOFF';
+/** Default local model: small enough for arm64 (Raspi/NanoPi), tool-free chat, ~1 GB. */
+exports.DEFAULT_LOCAL_MODEL_URL = 'https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct-GGUF/resolve/main/qwen2.5-1.5b-instruct-q4_k_m.gguf';
+const NPM = process.platform === 'win32' ? 'npm.cmd' : 'npm';
+/** Directory the on-demand npm install targets (`<dataDir>/node_modules/node-llama-cpp`). */
+function moduleDir(dataDir) {
+    return path.join(dataDir, 'node_modules', 'node-llama-cpp');
+}
+/** Local file path a model URL is cached to. */
+function modelPathFor(dataDir, url) {
+    const file = url.split('/').pop() || 'model.gguf';
+    return path.join(dataDir, 'models', file);
+}
+/** True once `node-llama-cpp` has been installed into the data dir. */
+function isLocalLlmInstalled(dataDir) {
+    try {
+        return fs.existsSync(path.join(moduleDir(dataDir), 'package.json'));
+    }
+    catch {
+        return false;
+    }
+}
+/** True once the model file has been downloaded. */
+function isModelDownloaded(dataDir, url) {
+    try {
+        return fs.existsSync(modelPathFor(dataDir, url));
+    }
+    catch {
+        return false;
+    }
+}
+/**
+ * Install `node-llama-cpp` (with its prebuilt native binary) into the data dir via a spawned `npm
+ * install`. Streams npm output to `onProgress`. Idempotent-ish: npm is a no-op if already present.
+ */
+function installLocalLlm(dataDir, onProgress) {
+    return new Promise((resolve, reject) => {
+        fs.mkdirSync(dataDir, { recursive: true });
+        // A local package.json makes npm install into <dataDir>/node_modules (and not walk up).
+        const pkg = path.join(dataDir, 'package.json');
+        if (!fs.existsSync(pkg)) {
+            fs.writeFileSync(pkg, JSON.stringify({ name: 'assistant-local-llm', private: true, version: '1.0.0' }));
+        }
+        const child = (0, node_child_process_1.spawn)(NPM, ['install', 'node-llama-cpp@3', '--omit=dev', '--no-audit', '--no-fund'], {
+            cwd: dataDir,
+            shell: process.platform === 'win32', // npm.cmd needs a shell on Windows
+        });
+        child.stdout.on('data', d => onProgress(String(d).trim()));
+        child.stderr.on('data', d => onProgress(String(d).trim()));
+        child.on('error', reject);
+        child.on('close', code => (code === 0 ? resolve() : reject(new Error(`npm install exited with code ${code}`))));
+    });
+}
+/**
+ * A loaded local model. `load()` imports the on-demand module, downloads the GGUF if missing, and
+ * spins up a llama context. `ask()` returns the model's answer (or {@link HANDOFF}).
+ */
+class LocalLlm {
+    opts;
+    session = null;
+    model = null;
+    context = null;
+    constructor(opts) {
+        this.opts = opts;
+    }
+    /** Full system prompt = the strict HANDOFF rule (device work must escalate to the tool-enabled cloud LLM). */
+    systemPrompt() {
+        // The local model has NO tools and cannot touch the smart home. Anything device-related MUST hand off,
+        // including control commands — otherwise a small model tends to "helpfully" refuse instead of escalating.
+        return (`You are a smart-home assistant, but you have NO ability to read or control anything in the home. ` +
+            `If the user wants to control or query ANYTHING in the home — e.g. turn on/off, switch, dim, ` +
+            `open/close, start/stop, set, activate, or ask for the value/state/status of a light, socket, ` +
+            `sensor, device, room, or the home — OR if you are unsure — you MUST reply with EXACTLY "${exports.HANDOFF}" ` +
+            `and nothing else. Do NOT apologise, do NOT explain, do NOT say you are an AI. ` +
+            `Reply directly ONLY for general knowledge, small talk, or questions unrelated to the home.`);
+    }
+    async load(onProgress) {
+        const dir = moduleDir(this.opts.dataDir);
+        if (!fs.existsSync(path.join(dir, 'package.json'))) {
+            throw new Error('node-llama-cpp is not installed — run the install first');
+        }
+        // Resolve the package entry from the data dir and import it (v3 is ESM-only → dynamic import).
+        const req = (0, node_module_1.createRequire)(path.join(this.opts.dataDir, 'package.json'));
+        const entry = req.resolve('node-llama-cpp');
+        const mod = await import((0, node_url_1.pathToFileURL)(entry).href);
+        const { getLlama, LlamaChatSession, createModelDownloader } = mod;
+        const modelPath = modelPathFor(this.opts.dataDir, this.opts.modelUrl);
+        if (!fs.existsSync(modelPath)) {
+            this.opts.log.info(`Downloading local model → ${modelPath}`);
+            fs.mkdirSync(path.dirname(modelPath), { recursive: true });
+            const downloader = await createModelDownloader({
+                modelUri: this.opts.modelUrl,
+                dirPath: path.dirname(modelPath),
+                fileName: path.basename(modelPath),
+                onProgress: (p) => {
+                    if (p.totalSize) {
+                        onProgress?.(`download ${Math.round((100 * p.downloadedSize) / p.totalSize)}%`);
+                    }
+                },
+            });
+            await downloader.download();
+        }
+        const llama = await getLlama();
+        this.model = await llama.loadModel({ modelPath });
+        this.context = await this.model.createContext();
+        this.session = new LlamaChatSession({
+            contextSequence: this.context.getSequence(),
+            systemPrompt: this.systemPrompt(),
+        });
+        this.opts.log.info('Local model loaded.');
+    }
+    /** Ask the local model. Returns its answer, or {@link HANDOFF} to escalate to the cloud LLM. */
+    async ask(question) {
+        if (!this.session) {
+            return exports.HANDOFF;
+        }
+        // Reset per-question so prompts stay independent (best-effort across node-llama-cpp versions).
+        try {
+            this.session.setChatHistory?.([]);
+        }
+        catch {
+            /* ignore */
+        }
+        const answer = await this.session.prompt(question, { maxTokens: this.opts.maxTokens || 512 });
+        return (answer || '').trim();
+    }
+    async dispose() {
+        try {
+            await this.context?.dispose?.();
+            await this.model?.dispose?.();
+        }
+        catch {
+            /* ignore */
+        }
+        this.session = this.context = this.model = null;
+    }
+}
+exports.LocalLlm = LocalLlm;
+/**
+ * Refusal patterns (de/en/ru) a small tool-free model emits when it *should* have handed off — e.g.
+ * "I'm an AI and can't control real devices". Treated as a hand-off so the tool-enabled cloud LLM runs.
+ */
+const REFUSAL = [
+    /\bhandoff\b/i,
+    // en
+    /\bi('?m| am)\s+an?\s+ai\b/i,
+    /\bi\s+(can('?t|not)|am\s+unable\s+to)\b.*\b(turn|switch|control|access|interact|device|light)\b/i,
+    // de
+    /\bich\s+bin\s+eine?\s+(ki|künstliche)\b/i,
+    /\bkann\s+.*\bnicht\b.*\b(schalten|steuern|einschalten|zugreifen|geräte?|licht)\b/i,
+    // ru (\w/\b are ASCII-only → use Cyrillic classes / plain substrings)
+    /искусственн[а-яё]*\s+интеллект/i,
+    /не\s+(могу|имею)/i,
+];
+/** True if the model's answer signals (or implies) a hand-off to the cloud LLM. */
+function isHandoff(answer) {
+    const trimmed = answer.trim();
+    const up = trimmed.toUpperCase();
+    if (up === exports.HANDOFF || up.startsWith(`${exports.HANDOFF} `) || up === `"${exports.HANDOFF}"`) {
+        return true;
+    }
+    return REFUSAL.some(re => re.test(trimmed));
+}
+//# sourceMappingURL=localLlm.js.map

package/build/lib/localLlm.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"localLlm.js","sourceRoot":"","sources":["../../src/lib/localLlm.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAmDA,kDAMC;AAGD,8CAMC;AAMD,0CAiBC;AA6GD,8BAOC;AA7MD,2DAA2C;AAC3C,6CAA4C;AAC5C,uCAAyC;AACzC,gDAAkC;AAClC,4CAA8B;AAE9B;;;;;;;;;;GAUG;AAEH,+EAA+E;AAClE,QAAA,OAAO,GAAG,SAAS,CAAC;AAEjC,yFAAyF;AAC5E,QAAA,uBAAuB,GAChC,uGAAuG,CAAC;AAE5G,MAAM,GAAG,GAAG,OAAO,CAAC,QAAQ,KAAK,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC;AAc7D,6FAA6F;AAC7F,SAAS,SAAS,CAAC,OAAe;IAC9B,OAAO,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,cAAc,EAAE,gBAAgB,CAAC,CAAC;AAChE,CAAC;AAED,gDAAgD;AAChD,SAAS,YAAY,CAAC,OAAe,EAAE,GAAW;IAC9C,MAAM,IAAI,GAAG,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,IAAI,YAAY,CAAC;IAClD,OAAO,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,IAAI,CAAC,CAAC;AAC9C,CAAC;AAED,uEAAuE;AACvE,SAAgB,mBAAmB,CAAC,OAAe;IAC/C,IAAI,CAAC;QACD,OAAO,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,EAAE,cAAc,CAAC,CAAC,CAAC;IACxE,CAAC;IAAC,MAAM,CAAC;QACL,OAAO,KAAK,CAAC;IACjB,CAAC;AACL,CAAC;AAED,oDAAoD;AACpD,SAAgB,iBAAiB,CAAC,OAAe,EAAE,GAAW;IAC1D,IAAI,CAAC;QACD,OAAO,EAAE,CAAC,UAAU,CAAC,YAAY,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC,CAAC;IACrD,CAAC;IAAC,MAAM,CAAC;QACL,OAAO,KAAK,CAAC;IACjB,CAAC;AACL,CAAC;AAED;;;GAGG;AACH,SAAgB,eAAe,CAAC,OAAe,EAAE,UAAkC;IAC/E,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACnC,EAAE,CAAC,SAAS,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAC3C,wFAAwF;QACxF,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;QAC/C,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;YACtB,EAAE,CAAC,aAAa,CAAC,GAAG,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,IAAI,EAAE,qBAAqB,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC;QAC5G,CAAC;QACD,MAAM,KAAK,GAAG,IAAA,0BAAK,EAAC,GAAG,EAAE,CAAC,SAAS,EAAE,kBAAkB,EAAE,YAAY,EAAE,YAAY,EAAE,WAAW,CAAC,EAAE;YAC/F,GAAG,EAAE,OAAO;YACZ,KAAK,EAAE,OAAO,CAAC,QAAQ,KAAK,OAAO,EAAE,mCAAmC;SAC3E,CAAC,CAAC;QACH,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC,EAAE,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;QAC3D,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC,EAAE,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;QAC3D,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;QAC1B,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,gCAAgC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IACpH,CAAC,CAAC,CAAC;AACP,CAAC;AAED;;;GAGG;AACH,MAAa,QAAQ;IAKY;IAJrB,OAAO,GAAQ,IAAI,CAAC;IACpB,KAAK,GAAQ,IAAI,CAAC;IAClB,OAAO,GAAQ,IAAI,CAAC;IAE5B,YAA6B,IAAqB;QAArB,SAAI,GAAJ,IAAI,CAAiB;IAAG,CAAC;IAEtD,8GAA8G;IACtG,YAAY;QAChB,uGAAuG;QACvG,0GAA0G;QAC1G,OAAO,CACH,mGAAmG;YACnG,8FAA8F;YAC9F,+FAA+F;YAC/F,2FAA2F,eAAO,IAAI;YACtG,gFAAgF;YAChF,4FAA4F,CAC/F,CAAC;IACN,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,UAAmC;QAC1C,MAAM,GAAG,GAAG,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QACzC,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,cAAc,CAAC,CAAC,EAAE,CAAC;YACjD,MAAM,IAAI,KAAK,CAAC,yDAAyD,CAAC,CAAC;QAC/E,CAAC;QACD,+FAA+F;QAC/F,MAAM,GAAG,GAAG,IAAA,2BAAa,EAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC,CAAC;QACxE,MAAM,KAAK,GAAG,GAAG,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC;QAC5C,MAAM,GAAG,GAAQ,MAAM,MAAM,CAAC,IAAA,wBAAa,EAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC;QACzD,MAAM,EAAE,QAAQ,EAAE,gBAAgB,EAAE,qBAAqB,EAAE,GAAG,GAAG,CAAC;QAElE,MAAM,SAAS,GAAG,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QACtE,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,SAAS,CAAC,EAAE,CAAC;YAC5B,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,6BAA6B,SAAS,EAAE,CAAC,CAAC;YAC7D,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YAC3D,MAAM,UAAU,GAAG,MAAM,qBAAqB,CAAC;gBAC3C,QAAQ,EAAE,IAAI,CAAC,IAAI,CAAC,QAAQ;gBAC5B,OAAO,EAAE,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC;gBAChC,QAAQ,EAAE,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC;gBAClC,UAAU,EAAE,CAAC,CAAgD,EAAE,EAAE;oBAC7D,IAAI,CAAC,CAAC,SAAS,EAAE,CAAC;wBACd,UAAU,EAAE,CAAC,YAAY,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,cAAc,CAAC,GAAG,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;oBACpF,CAAC;gBACL,CAAC;aACJ,CAAC,CAAC;YACH,MAAM,UAAU,CAAC,QAAQ,EAAE,CAAC;QAChC,CAAC;QAED,MAAM,KAAK,GAAG,MAAM,QAAQ,EAAE,CAAC;QAC/B,IAAI,CAAC,KAAK,GAAG,MAAM,KAAK,CAAC,SAAS,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC;QAClD,IAAI,CAAC,OAAO,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,CAAC;QAChD,IAAI,CAAC,OAAO,GAAG,IAAI,gBAAgB,CAAC;YAChC,eAAe,EAAE,IAAI,CAAC,OAAO,CAAC,WAAW,EAAE;YAC3C,YAAY,EAAE,IAAI,CAAC,YAAY,EAAE;SACpC,CAAC,CAAC;QACH,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAC;IAC9C,CAAC;IAED,gGAAgG;IAChG,KAAK,CAAC,GAAG,CAAC,QAAgB;QACtB,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC;YAChB,OAAO,eAAO,CAAC;QACnB,CAAC;QACD,+FAA+F;QAC/F,IAAI,CAAC;YACD,IAAI,CAAC,OAAO,CAAC,cAAc,EAAE,CAAC,EAAE,CAAC,CAAC;QACtC,CAAC;QAAC,MAAM,CAAC;YACL,YAAY;QAChB,CAAC;QACD,MAAM,MAAM,GAAW,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,CAAC,IAAI,CAAC,SAAS,IAAI,GAAG,EAAE,CAAC,CAAC;QACtG,OAAO,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IACjC,CAAC;IAED,KAAK,CAAC,OAAO;QACT,IAAI,CAAC;YACD,MAAM,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC;YAChC,MAAM,IAAI,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE,CAAC;QAClC,CAAC;QAAC,MAAM,CAAC;YACL,YAAY;QAChB,CAAC;QACD,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC;IACpD,CAAC;CACJ;AAnFD,4BAmFC;AAED;;;GAGG;AACH,MAAM,OAAO,GAAG;IACZ,cAAc;IACd,KAAK;IACL,4BAA4B;IAC5B,kGAAkG;IAClG,KAAK;IACL,0CAA0C;IAC1C,mFAAmF;IACnF,sEAAsE;IACtE,iCAAiC;IACjC,mBAAmB;CACtB,CAAC;AAEF,mFAAmF;AACnF,SAAgB,SAAS,CAAC,MAAc;IACpC,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC;IAC9B,MAAM,EAAE,GAAG,OAAO,CAAC,WAAW,EAAE,CAAC;IACjC,IAAI,EAAE,KAAK,eAAO,IAAI,EAAE,CAAC,UAAU,CAAC,GAAG,eAAO,GAAG,CAAC,IAAI,EAAE,KAAK,IAAI,eAAO,GAAG,EAAE,CAAC;QAC1E,OAAO,IAAI,CAAC;IAChB,CAAC;IACD,OAAO,OAAO,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC;AAChD,CAAC","sourcesContent":["import { spawn } from 'node:child_process';\nimport { createRequire } from 'node:module';\nimport { pathToFileURL } from 'node:url';\nimport * as path from 'node:path';\nimport * as fs from 'node:fs';\n\n/**\n * Tier-1a local LLM via `node-llama-cpp` — installed **on demand** (not a package.json dependency, so\n * users who don't enable it never pay the large native download). Everything lives in the writable\n * instance data dir (`<iobroker-data>/assistant.<n>/`), which survives adapter upgrades; on a fresh\n * upgrade the module is simply re-installed on request.\n *\n * The model runs **tool-free**: it answers general questions and, when a request needs live ioBroker\n * state or it is unsure, it emits the {@link HANDOFF} sentinel so the orchestrator escalates to the\n * cloud LLM (which has the tools). Small models handle a \"say HANDOFF\" rule far more reliably than\n * real tool-calling.\n */\n\n/** Sentinel the local model returns to escalate a request to the cloud LLM. */\nexport const HANDOFF = 'HANDOFF';\n\n/** Default local model: small enough for arm64 (Raspi/NanoPi), tool-free chat, ~1 GB. */\nexport const DEFAULT_LOCAL_MODEL_URL =\n    'https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct-GGUF/resolve/main/qwen2.5-1.5b-instruct-q4_k_m.gguf';\n\nconst NPM = process.platform === 'win32' ? 'npm.cmd' : 'npm';\n\ntype Logger = Pick<ioBroker.Logger, 'info' | 'warn' | 'debug' | 'error'>;\n\nexport interface LocalLlmOptions {\n    /** Writable instance data dir (from `adapter.getAbsoluteInstanceDataDir()`). */\n    dataDir: string;\n    /** Direct URL of a GGUF model file. */\n    modelUrl: string;\n    systemPrompt?: string;\n    maxTokens?: number;\n    log: Logger;\n}\n\n/** Directory the on-demand npm install targets (`<dataDir>/node_modules/node-llama-cpp`). */\nfunction moduleDir(dataDir: string): string {\n    return path.join(dataDir, 'node_modules', 'node-llama-cpp');\n}\n\n/** Local file path a model URL is cached to. */\nfunction modelPathFor(dataDir: string, url: string): string {\n    const file = url.split('/').pop() || 'model.gguf';\n    return path.join(dataDir, 'models', file);\n}\n\n/** True once `node-llama-cpp` has been installed into the data dir. */\nexport function isLocalLlmInstalled(dataDir: string): boolean {\n    try {\n        return fs.existsSync(path.join(moduleDir(dataDir), 'package.json'));\n    } catch {\n        return false;\n    }\n}\n\n/** True once the model file has been downloaded. */\nexport function isModelDownloaded(dataDir: string, url: string): boolean {\n    try {\n        return fs.existsSync(modelPathFor(dataDir, url));\n    } catch {\n        return false;\n    }\n}\n\n/**\n * Install `node-llama-cpp` (with its prebuilt native binary) into the data dir via a spawned `npm\n * install`. Streams npm output to `onProgress`. Idempotent-ish: npm is a no-op if already present.\n */\nexport function installLocalLlm(dataDir: string, onProgress: (line: string) => void): Promise<void> {\n    return new Promise((resolve, reject) => {\n        fs.mkdirSync(dataDir, { recursive: true });\n        // A local package.json makes npm install into <dataDir>/node_modules (and not walk up).\n        const pkg = path.join(dataDir, 'package.json');\n        if (!fs.existsSync(pkg)) {\n            fs.writeFileSync(pkg, JSON.stringify({ name: 'assistant-local-llm', private: true, version: '1.0.0' }));\n        }\n        const child = spawn(NPM, ['install', 'node-llama-cpp@3', '--omit=dev', '--no-audit', '--no-fund'], {\n            cwd: dataDir,\n            shell: process.platform === 'win32', // npm.cmd needs a shell on Windows\n        });\n        child.stdout.on('data', d => onProgress(String(d).trim()));\n        child.stderr.on('data', d => onProgress(String(d).trim()));\n        child.on('error', reject);\n        child.on('close', code => (code === 0 ? resolve() : reject(new Error(`npm install exited with code ${code}`))));\n    });\n}\n\n/**\n * A loaded local model. `load()` imports the on-demand module, downloads the GGUF if missing, and\n * spins up a llama context. `ask()` returns the model's answer (or {@link HANDOFF}).\n */\nexport class LocalLlm {\n    private session: any = null;\n    private model: any = null;\n    private context: any = null;\n\n    constructor(private readonly opts: LocalLlmOptions) {}\n\n    /** Full system prompt = the strict HANDOFF rule (device work must escalate to the tool-enabled cloud LLM). */\n    private systemPrompt(): string {\n        // The local model has NO tools and cannot touch the smart home. Anything device-related MUST hand off,\n        // including control commands — otherwise a small model tends to \"helpfully\" refuse instead of escalating.\n        return (\n            `You are a smart-home assistant, but you have NO ability to read or control anything in the home. ` +\n            `If the user wants to control or query ANYTHING in the home — e.g. turn on/off, switch, dim, ` +\n            `open/close, start/stop, set, activate, or ask for the value/state/status of a light, socket, ` +\n            `sensor, device, room, or the home — OR if you are unsure — you MUST reply with EXACTLY \"${HANDOFF}\" ` +\n            `and nothing else. Do NOT apologise, do NOT explain, do NOT say you are an AI. ` +\n            `Reply directly ONLY for general knowledge, small talk, or questions unrelated to the home.`\n        );\n    }\n\n    async load(onProgress?: (line: string) => void): Promise<void> {\n        const dir = moduleDir(this.opts.dataDir);\n        if (!fs.existsSync(path.join(dir, 'package.json'))) {\n            throw new Error('node-llama-cpp is not installed — run the install first');\n        }\n        // Resolve the package entry from the data dir and import it (v3 is ESM-only → dynamic import).\n        const req = createRequire(path.join(this.opts.dataDir, 'package.json'));\n        const entry = req.resolve('node-llama-cpp');\n        const mod: any = await import(pathToFileURL(entry).href);\n        const { getLlama, LlamaChatSession, createModelDownloader } = mod;\n\n        const modelPath = modelPathFor(this.opts.dataDir, this.opts.modelUrl);\n        if (!fs.existsSync(modelPath)) {\n            this.opts.log.info(`Downloading local model → ${modelPath}`);\n            fs.mkdirSync(path.dirname(modelPath), { recursive: true });\n            const downloader = await createModelDownloader({\n                modelUri: this.opts.modelUrl,\n                dirPath: path.dirname(modelPath),\n                fileName: path.basename(modelPath),\n                onProgress: (p: { downloadedSize: number; totalSize: number }) => {\n                    if (p.totalSize) {\n                        onProgress?.(`download ${Math.round((100 * p.downloadedSize) / p.totalSize)}%`);\n                    }\n                },\n            });\n            await downloader.download();\n        }\n\n        const llama = await getLlama();\n        this.model = await llama.loadModel({ modelPath });\n        this.context = await this.model.createContext();\n        this.session = new LlamaChatSession({\n            contextSequence: this.context.getSequence(),\n            systemPrompt: this.systemPrompt(),\n        });\n        this.opts.log.info('Local model loaded.');\n    }\n\n    /** Ask the local model. Returns its answer, or {@link HANDOFF} to escalate to the cloud LLM. */\n    async ask(question: string): Promise<string> {\n        if (!this.session) {\n            return HANDOFF;\n        }\n        // Reset per-question so prompts stay independent (best-effort across node-llama-cpp versions).\n        try {\n            this.session.setChatHistory?.([]);\n        } catch {\n            /* ignore */\n        }\n        const answer: string = await this.session.prompt(question, { maxTokens: this.opts.maxTokens || 512 });\n        return (answer || '').trim();\n    }\n\n    async dispose(): Promise<void> {\n        try {\n            await this.context?.dispose?.();\n            await this.model?.dispose?.();\n        } catch {\n            /* ignore */\n        }\n        this.session = this.context = this.model = null;\n    }\n}\n\n/**\n * Refusal patterns (de/en/ru) a small tool-free model emits when it *should* have handed off — e.g.\n * \"I'm an AI and can't control real devices\". Treated as a hand-off so the tool-enabled cloud LLM runs.\n */\nconst REFUSAL = [\n    /\\bhandoff\\b/i,\n    // en\n    /\\bi('?m| am)\\s+an?\\s+ai\\b/i,\n    /\\bi\\s+(can('?t|not)|am\\s+unable\\s+to)\\b.*\\b(turn|switch|control|access|interact|device|light)\\b/i,\n    // de\n    /\\bich\\s+bin\\s+eine?\\s+(ki|künstliche)\\b/i,\n    /\\bkann\\s+.*\\bnicht\\b.*\\b(schalten|steuern|einschalten|zugreifen|geräte?|licht)\\b/i,\n    // ru (\\w/\\b are ASCII-only → use Cyrillic classes / plain substrings)\n    /искусственн[а-яё]*\\s+интеллект/i,\n    /не\\s+(могу|имею)/i,\n];\n\n/** True if the model's answer signals (or implies) a hand-off to the cloud LLM. */\nexport function isHandoff(answer: string): boolean {\n    const trimmed = answer.trim();\n    const up = trimmed.toUpperCase();\n    if (up === HANDOFF || up.startsWith(`${HANDOFF} `) || up === `\"${HANDOFF}\"`) {\n        return true;\n    }\n    return REFUSAL.some(re => re.test(trimmed));\n}\n"]}