@alexkroman1/aai 1.6.1 → 1.7.0

package/dist/host/runtime-barrel.js

@@ -1,16 +1,18 @@
 import { r as DEFAULT_SYSTEM_PROMPT } from "../types-KUgezM6u.js";
 import { _ as TOOL_EXECUTION_TIMEOUT_MS, a as DEFAULT_SHUTDOWN_TIMEOUT_MS, c as FETCH_TIMEOUT_MS, d as MAX_PAGE_CHARS, g as RUN_CODE_TIMEOUT_MS, h as PIPELINE_FLUSH_TIMEOUT_MS, l as MAX_HTML_BYTES, m as MAX_WS_PAYLOAD_BYTES, o as DEFAULT_STT_SAMPLE_RATE, p as MAX_VALUE_SIZE, s as DEFAULT_TTS_SAMPLE_RATE, t as AGENT_CSP } from "../constants-C2nirZUI.js";
 import { i as toolError, n as errorDetail, r as errorMessage, t as parseWsUpgradeParams } from "../ws-upgrade-BeOQ7fXL.js";
-import { ClientMessageSchema, buildReadyConfig, lenientParse } from "../sdk/protocol.js";
-import { a as toAgentConfig, c as makeSttError, i as agentToolsToSchemas, l as makeTtsError, n as EMPTY_PARAMS, s as assertProviderTriple } from "../_internal-types-DFL07G3f.js";
-import { a as ASSEMBLYAI_KIND, r as ELEVENLABS_KIND, s as DEEPGRAM_KIND, t as SONIOX_KIND } from "../soniox-DCQ3GqJq.js";
-import { a as RIME_KIND, n as CARTESIA_KIND } from "../cartesia-BfQPOQ7Y.js";
-import { a as MISTRAL_KIND, d as ANTHROPIC_KIND, l as GOOGLE_KIND, r as OPENAI_KIND, s as GROQ_KIND } from "../xai-jfQsxxPZ.js";
+import { ClientMessageSchema, VectorRequestSchema, buildReadyConfig, lenientParse } from "../sdk/protocol.js";
+import { a as toAgentConfig, c as makeSttError, i as agentToolsToSchemas, l as makeTtsError, n as EMPTY_PARAMS, s as assertProviderTriple } from "../_internal-types-CrnTi9Ew.js";
+import { a as MISTRAL_KIND, d as ANTHROPIC_KIND, l as GOOGLE_KIND, r as OPENAI_KIND, s as GROQ_KIND } from "../xai-BDI61Y2M.js";
+import { a as DEEPGRAM_KIND, r as ELEVENLABS_KIND, s as ASSEMBLYAI_KIND, t as SONIOX_KIND } from "../soniox-BQdL0mB5.js";
+import { a as CARTESIA_KIND, n as RIME_KIND } from "../rime-58p9mDR8.js";
+import { a as MEMORY_KV_KIND, r as REDIS_KV_KIND } from "../s3-BtCMvCod.js";
+import { r as IN_MEMORY_VECTOR_KIND, t as PINECONE_VECTOR_KIND } from "../pinecone-CeJ69aRs.js";
+import { createRequire } from "node:module";
 import { z } from "zod";
 import { convert } from "html-to-text";
 import vm from "node:vm";
-import pTimeout from "p-timeout";
-import { createStorage, prefixStorage } from "unstorage";
+import { createHash, randomUUID } from "node:crypto";
 import { createAnthropic } from "@ai-sdk/anthropic";
 import { createGoogleGenerativeAI } from "@ai-sdk/google";
 import { createGroq } from "@ai-sdk/groq";
@@ -22,9 +24,10 @@ import { createNanoEvents } from "nanoevents";
 import { DeepgramClient } from "@deepgram/sdk";
 import { ElevenLabsClient } from "@elevenlabs/elevenlabs-js";
 import { AudioFormat, CommitStrategy, RealtimeEvents } from "@elevenlabs/elevenlabs-js/wrapper/realtime/index.js";
-import WsWebSocket, { WebSocketServer } from "ws";
-import { randomUUID } from "node:crypto";
+import WebSocket, { WebSocketServer } from "ws";
 import { Cartesia } from "@cartesia/cartesia-js";
+import { createStorage, prefixStorage } from "unstorage";
+import pTimeout from "p-timeout";
 import { jsonSchema, stepCountIs, streamText, tool } from "ai";
 import fs from "node:fs";
 import http from "node:http";
@@ -356,35 +359,84 @@ function resolveAllBuiltins(names, opts) {
 	};
 }
 //#endregion
-//#region sdk/system-prompt.ts
-function getFormattedDate() {
-	return (/* @__PURE__ */ new Date()).toLocaleDateString("en-US", {
-		weekday: "long",
-		year: "numeric",
-		month: "long",
-		day: "numeric"
-	});
-}
-const VOICE_RULES = "\n\nCRITICAL OUTPUT RULES — you MUST follow these for EVERY response:\nYour response will be spoken aloud by a TTS system and displayed as plain text.\n- NEVER use markdown: no **, no *, no _, no #, no `, no [](), no ---\n- NEVER use bullet points (-, *, •) or numbered lists (1., 2.)\n- NEVER use code blocks or inline code\n- NEVER mention tools, search, APIs, or technical failures to the user. If a tool returns no results, just answer naturally without explaining why.\n- Write exactly as you would say it out loud to a friend\n- Use short conversational sentences. To list things, say \"First,\" \"Next,\" \"Finally,\"\n- Keep responses concise — 1 to 3 sentences max";
+//#region host/memory-vector.ts
 /**
-* Build the system prompt sent to the LLM from the agent configuration.
+* In-memory Vector implementation.
 *
-* Assembles the default system prompt, today's date, agent-specific instructions,
-* and optional sections for tool usage preamble and voice output rules.
-*
-* @param config - The serializable agent configuration (name, systemPrompt, etc.).
-* @param opts.hasTools - When `true`, appends a preamble instructing the LLM to
-*   speak a brief phrase before each tool call to fill silence.
-* @param opts.voice - When `true`, appends strict voice-specific output rules
-*   (no markdown, no bullet points, conversational tone, concise responses).
-* @returns The assembled system prompt string.
+* INTENTIONALLY BAD QUALITY. Pseudo-embedding hashes the text into a
+* 64-dim Float32Array of values in [-1, ~0.99], then L2-normalizes
+* the result. Because both stored and probe vectors are unit-length,
+* cosine similarity reduces to a plain dot product — that's what
+* `cosine()` computes. Used only for `aai dev` and tests — the goal
+* is proving tool wiring, not retrieval ranking.
 */
-function buildSystemPrompt(config, opts) {
-	const { hasTools } = opts;
-	const agentInstructions = config.systemPrompt && config.systemPrompt !== DEFAULT_SYSTEM_PROMPT ? `\n\nAgent-Specific Instructions:\n${config.systemPrompt}` : "";
-	const toolPreamble = hasTools ? "\n\nWhen you decide to use a tool, ALWAYS say a brief natural phrase BEFORE the tool call (e.g. \"Let me look that up\" or \"One moment while I check\"). This fills silence while the tool executes. Keep preambles to one short sentence." : "";
-	const guidance = opts.toolGuidance && opts.toolGuidance.length > 0 ? `\n\nBuilt-in Tool Usage:\n${opts.toolGuidance.join("\n")}` : "";
-	return DEFAULT_SYSTEM_PROMPT + `\n\nToday's date is ${getFormattedDate()}.` + agentInstructions + toolPreamble + guidance + (opts.voice ? VOICE_RULES : "");
+const stores = /* @__PURE__ */ new Map();
+function getStore(ns) {
+	let store = stores.get(ns);
+	if (!store) {
+		store = /* @__PURE__ */ new Map();
+		stores.set(ns, store);
+	}
+	return store;
+}
+const DIM = 64;
+function pseudoEmbed(text) {
+	const out = new Float32Array(DIM);
+	const h1 = createHash("sha256").update(text).digest();
+	const h2 = createHash("sha256").update(h1).digest();
+	for (let i = 0; i < 32; i++) out[i] = (h1[i] - 128) / 128;
+	for (let i = 0; i < 32; i++) out[i + 32] = (h2[i] - 128) / 128;
+	let norm = 0;
+	for (let i = 0; i < DIM; i++) norm += out[i] * out[i];
+	norm = Math.sqrt(norm) || 1;
+	for (let i = 0; i < DIM; i++) out[i] = out[i] / norm;
+	return out;
+}
+function cosine(a, b) {
+	let dot = 0;
+	for (let i = 0; i < DIM; i++) dot += a[i] * b[i];
+	return dot;
+}
+function matches(metadata, filter) {
+	for (const [key, want] of Object.entries(filter)) {
+		if (want !== null && typeof want === "object") throw new Error(`In-memory Vector: filter operator unsupported (${key}). Only top-level exact-match is supported.`);
+		if (metadata?.[key] !== want) return false;
+	}
+	return true;
+}
+function createMemoryVector(opts) {
+	const ns = opts.namespace;
+	return {
+		async upsert(id, text, metadata) {
+			getStore(ns).set(id, {
+				text,
+				metadata,
+				vec: pseudoEmbed(text)
+			});
+		},
+		async query(text, queryOpts) {
+			const topK = queryOpts?.topK ?? 5;
+			const filter = queryOpts?.filter;
+			const probe = pseudoEmbed(text);
+			const scored = [];
+			for (const [id, rec] of getStore(ns)) {
+				if (filter && !matches(rec.metadata, filter)) continue;
+				scored.push({
+					id,
+					score: cosine(probe, rec.vec),
+					text: rec.text,
+					...rec.metadata !== void 0 ? { metadata: rec.metadata } : {}
+				});
+			}
+			scored.sort((a, b) => b.score - a.score);
+			return scored.slice(0, topK);
+		},
+		async delete(ids) {
+			const store = getStore(ns);
+			const list = Array.isArray(ids) ? ids : [ids];
+			for (const id of list) store.delete(id);
+		}
+	};
 }
 //#endregion
 //#region host/providers/stt/assemblyai.ts
@@ -762,7 +814,7 @@ function openSoniox(opts = {}) {
 		async open(openOpts) {
 			const apiKey = openOpts.apiKey || process.env.SONIOX_API_KEY;
 			if (!apiKey) throw makeSttError("stt_auth_failed", "Soniox STT: missing API key. Set SONIOX_API_KEY in the agent env.");
-			const ws = new WsWebSocket(SONIOX_WS_URL);
+			const ws = new WebSocket(SONIOX_WS_URL);
 			const emitter = createNanoEvents();
 			let closed = false;
 			const finalBuf = { value: "" };
@@ -800,7 +852,7 @@ function openSoniox(opts = {}) {
 			else openOpts.signal.addEventListener("abort", () => void close(), { once: true });
 			return {
 				sendAudio(pcm) {
-					if (closed || ws.readyState !== WsWebSocket.OPEN) return;
+					if (closed || ws.readyState !== WebSocket.OPEN) return;
 					ws.send(new Uint8Array(pcm.buffer, pcm.byteOffset, pcm.byteLength), { binary: true });
 				},
 				on(event, fn) {
@@ -1112,7 +1164,7 @@ function openRime(opts) {
 			const url = `wss://users-ws.rime.ai/ws2?speaker=${encodeURIComponent(voice)}&modelId=${encodeURIComponent(model)}&audioFormat=pcm&samplingRate=${sampleRate}&lang=${encodeURIComponent(lang)}`;
 			let ws;
 			try {
-				ws = new WsWebSocket(url, { headers: { Authorization: `Bearer ${apiKey}` } });
+				ws = new WebSocket(url, { headers: { Authorization: `Bearer ${apiKey}` } });
 			} catch (cause) {
 				throw makeTtsError("tts_connect_failed", `Rime TTS: failed to create WebSocket: ${cause instanceof Error ? cause.message : String(cause)}`);
 			}
@@ -1172,19 +1224,19 @@ function openRime(opts) {
 			return {
 				sendText(text) {
 					if (closed || text.length === 0) return;
-					if (ws.readyState !== WsWebSocket.OPEN) return;
+					if (ws.readyState !== WebSocket.OPEN) return;
 					doneEmitted = false;
 					ws.send(JSON.stringify({ text }));
 				},
 				flush() {
 					if (closed) return;
-					if (ws.readyState !== WsWebSocket.OPEN) return;
+					if (ws.readyState !== WebSocket.OPEN) return;
 					ws.send(JSON.stringify({ text: "." }));
 					armFirstAudioTimer();
 				},
 				cancel() {
 					if (closed) return;
-					if (ws.readyState === WsWebSocket.OPEN) ws.send(JSON.stringify({ operation: "clear" }));
+					if (ws.readyState === WebSocket.OPEN) ws.send(JSON.stringify({ operation: "clear" }));
 					emitDoneOnce();
 				},
 				on(event, fn) {
@@ -1259,12 +1311,259 @@ function resolveLlm(descriptor, env) {
 		default: throw new Error(`Unknown LLM provider kind: "${descriptor.kind}". Supported: ${ANTHROPIC_KIND}, ${OPENAI_KIND}, ${GOOGLE_KIND}, ${MISTRAL_KIND}, xai, ${GROQ_KIND}.`);
 	}
 }
+const requireFromHere = createRequire(import.meta.url);
+/**
+* Lazy-load a package via createRequire so the package is a true optional
+* peer dependency — if it's not installed the error surfaces only when the
+* provider is actually used, not at module load time.
+*/
+function loadProviderPackage(name, label) {
+	try {
+		return requireFromHere(name);
+	} catch (err) {
+		if (err instanceof Error && (err.code === "MODULE_NOT_FOUND" || err.code === "ERR_MODULE_NOT_FOUND") && err.message.includes(name)) throw new Error(`${label}: package \`${name}\` is not installed. Run \`pnpm add ${name}\`.`, { cause: err });
+		throw err;
+	}
+}
 function requireKey(env, name, label) {
 	const key = resolveApiKey(name, env);
 	if (!key) throw new Error(`${label} LLM: missing API key. Set ${name} in the agent env.`);
 	return key;
 }
 //#endregion
+//#region host/pinecone-vector.ts
+function createPineconeVector(opts) {
+	const { Pinecone } = loadProviderPackage("@pinecone-database/pinecone", "Pinecone Vector");
+	const client = new Pinecone({ apiKey: opts.apiKey });
+	const ns = () => client.index(opts.index).namespace(opts.namespace);
+	return {
+		async upsert(id, text, metadata) {
+			const record = {
+				_id: id,
+				text,
+				...metadata ?? {}
+			};
+			await ns().upsertRecords([record]);
+		},
+		async query(text, queryOpts) {
+			const topK = queryOpts?.topK ?? 5;
+			const req = {
+				query: {
+					inputs: { text },
+					topK,
+					...queryOpts?.filter !== void 0 ? { filter: queryOpts.filter } : {}
+				},
+				fields: ["*"]
+			};
+			return (await ns().searchRecords(req)).result.hits.map((hit) => {
+				const { text: hitText, ...rest } = hit.fields;
+				const metadata = Object.keys(rest).length > 0 ? rest : void 0;
+				return {
+					id: hit._id,
+					score: hit._score,
+					text: typeof hitText === "string" ? hitText : "",
+					...metadata !== void 0 ? { metadata } : {}
+				};
+			});
+		},
+		async delete(ids) {
+			const list = Array.isArray(ids) ? ids : [ids];
+			await ns().deleteMany(list);
+		}
+	};
+}
+//#endregion
+//#region host/unstorage-kv.ts
+/**
+* Key-value store backed by unstorage.
+*
+* Works with any unstorage driver (memory, fs, S3/R2, etc.).
+*/
+/**
+* Create a KV store backed by any unstorage driver.
+*
+* @param options - See {@link UnstorageKvOptions}.
+* @returns A {@link Kv} instance.
+*
+* @example
+* ```ts
+* import { createStorage } from "unstorage";
+* import { createUnstorageKv } from "@alexkroman1/aai/unstorage-kv";
+*
+* const kv = createUnstorageKv({ storage: createStorage() });
+* await kv.set("greeting", "hello");
+* const value = await kv.get<string>("greeting"); // "hello"
+* ```
+*/
+function createUnstorageKv(options) {
+	const store = options.prefix ? prefixStorage(options.storage, options.prefix) : options.storage;
+	return {
+		async get(key) {
+			return await store.getItem(key) ?? null;
+		},
+		async set(key, value, setOptions) {
+			if (JSON.stringify(value).length > 65536) throw new Error(`Value exceeds max size of ${MAX_VALUE_SIZE} bytes`);
+			const storable = value;
+			if (setOptions?.expireIn && setOptions.expireIn > 0) await store.setItem(key, storable, { ttl: Math.ceil(setOptions.expireIn / 1e3) });
+			else await store.setItem(key, storable);
+		},
+		async delete(keys) {
+			const keyArray = Array.isArray(keys) ? keys : [keys];
+			await Promise.all(keyArray.map((k) => store.removeItem(k)));
+		},
+		close() {
+			store.dispose();
+		}
+	};
+}
+//#endregion
+//#region host/providers/resolve-kv.ts
+/**
+* Descriptor → concrete `Kv` resolver. Mirror of `resolveLlm` /
+* `resolveVector`. Always wraps the produced unstorage Storage in
+* `createUnstorageKv` with the provided per-tenant prefix so namespace
+* isolation is enforced regardless of backend choice.
+*/
+/**
+* Load a CJS unstorage driver factory. The CJS variants use
+* `module.exports = defineDriver(...)` so the require result is the
+* factory itself (not an object with `.default`).
+*
+* Delegates to loadProviderPackage (lazy-load via createRequire so the
+* driver is a true optional peer dep).
+*/
+function loadDriver(modulePath, label) {
+	return loadProviderPackage(modulePath, `${label} KV: driver`);
+}
+/**
+* Build a lazy unstorage Driver that defers loading the real driver
+* factory until the first I/O operation. This is necessary for drivers
+* whose peer dependencies (e.g. `ioredis`) may not be installed on the
+* host at startup — the missing package will only surface when the agent
+* actually performs KV operations, not at session creation time.
+*/
+function makeLazyDriver(modulePath, label, opts) {
+	let resolved = null;
+	const get = () => {
+		if (!resolved) resolved = loadDriver(modulePath, label)(opts);
+		return resolved;
+	};
+	return {
+		name: label.toLowerCase(),
+		hasItem: (key, txOpts) => get().hasItem(key, txOpts),
+		getItem: (key, txOpts) => get().getItem(key, txOpts),
+		getItemRaw: (key, txOpts) => get().getItemRaw?.(key, txOpts) ?? null,
+		setItem: (key, value, txOpts) => get().setItem?.(key, value, txOpts),
+		setItemRaw: (key, value, txOpts) => get().setItemRaw?.(key, value, txOpts),
+		removeItem: (key, txOpts) => get().removeItem?.(key, txOpts),
+		getKeys: (base, txOpts) => get().getKeys(base, txOpts),
+		clear: (base, txOpts) => get().clear?.(base, txOpts),
+		dispose: () => resolved ? resolved.dispose?.() : void 0
+	};
+}
+/** Resolve a {@link KvProvider} descriptor into a {@link Kv}. */
+function resolveKv(descriptor, env, prefix) {
+	switch (descriptor.kind) {
+		case MEMORY_KV_KIND: return createUnstorageKv({
+			storage: createStorage(),
+			prefix
+		});
+		case "fs": {
+			const opts = descriptor.options;
+			return createUnstorageKv({
+				storage: createStorage({ driver: loadDriver("unstorage/drivers/fs", "fs")({ base: opts.base }) }),
+				prefix
+			});
+		}
+		case "s3": {
+			const opts = descriptor.options;
+			const accessKeyId = resolveApiKey("AWS_ACCESS_KEY_ID", env);
+			const secretAccessKey = resolveApiKey("AWS_SECRET_ACCESS_KEY", env);
+			if (!(accessKeyId && secretAccessKey)) throw new Error("S3 KV: missing AWS credentials. Set AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY in the agent env.");
+			return createUnstorageKv({
+				storage: createStorage({ driver: loadDriver("unstorage/drivers/s3", "S3")({
+					bucket: opts.bucket,
+					...opts.endpoint !== void 0 ? { endpoint: opts.endpoint } : {},
+					region: opts.region ?? "auto",
+					accessKeyId,
+					secretAccessKey
+				}) }),
+				prefix
+			});
+		}
+		case REDIS_KV_KIND: {
+			const opts = descriptor.options;
+			const url = resolveApiKey("REDIS_URL", env);
+			if (!url) throw new Error("Redis KV: missing connection URL. Set REDIS_URL in the agent env.");
+			return createUnstorageKv({
+				storage: createStorage({ driver: makeLazyDriver("unstorage/drivers/redis", "Redis", {
+					url,
+					...opts.tls !== void 0 ? { tls: opts.tls } : {}
+				}) }),
+				prefix
+			});
+		}
+		default: throw new Error(`Unknown KV provider kind: "${descriptor.kind}". Supported: ${MEMORY_KV_KIND}, fs, s3, ${REDIS_KV_KIND}.`);
+	}
+}
+//#endregion
+//#region host/providers/resolve-vector.ts
+/**
+* Descriptor → concrete `Vector` resolver. Mirror of `resolveLlm`.
+*
+* Pulls API keys from the agent env so descriptors stay
+* secret-free. Lazy-loads provider SDKs via `createRequire` so
+* unused providers never enter the bundle.
+*/
+/** Resolve a {@link VectorProvider} descriptor into a {@link Vector}. */
+function resolveVector(descriptor, env, namespace) {
+	switch (descriptor.kind) {
+		case IN_MEMORY_VECTOR_KIND: return createMemoryVector({ namespace });
+		case PINECONE_VECTOR_KIND: {
+			const apiKey = resolveApiKey("PINECONE_API_KEY", env);
+			if (!apiKey) throw new Error("Pinecone Vector: missing API key. Set PINECONE_API_KEY in the agent env.");
+			const opts = descriptor.options;
+			return createPineconeVector({
+				apiKey,
+				index: opts.index,
+				namespace
+			});
+		}
+		default: throw new Error(`Unknown Vector provider kind: "${descriptor.kind}". Supported: ${IN_MEMORY_VECTOR_KIND}, ${PINECONE_VECTOR_KIND}.`);
+	}
+}
+//#endregion
+//#region sdk/system-prompt.ts
+function getFormattedDate() {
+	return (/* @__PURE__ */ new Date()).toLocaleDateString("en-US", {
+		weekday: "long",
+		year: "numeric",
+		month: "long",
+		day: "numeric"
+	});
+}
+const VOICE_RULES = "\n\nCRITICAL OUTPUT RULES — you MUST follow these for EVERY response:\nYour response will be spoken aloud by a TTS system and displayed as plain text.\n- NEVER use markdown: no **, no *, no _, no #, no `, no [](), no ---\n- NEVER use bullet points (-, *, •) or numbered lists (1., 2.)\n- NEVER use code blocks or inline code\n- NEVER mention tools, search, APIs, or technical failures to the user. If a tool returns no results, just answer naturally without explaining why.\n- Write exactly as you would say it out loud to a friend\n- Use short conversational sentences. To list things, say \"First,\" \"Next,\" \"Finally,\"\n- Keep responses concise — 1 to 3 sentences max";
+/**
+* Build the system prompt sent to the LLM from the agent configuration.
+*
+* Assembles the default system prompt, today's date, agent-specific instructions,
+* and optional sections for tool usage preamble and voice output rules.
+*
+* @param config - The serializable agent configuration (name, systemPrompt, etc.).
+* @param opts.hasTools - When `true`, appends a preamble instructing the LLM to
+*   speak a brief phrase before each tool call to fill silence.
+* @param opts.voice - When `true`, appends strict voice-specific output rules
+*   (no markdown, no bullet points, conversational tone, concise responses).
+* @returns The assembled system prompt string.
+*/
+function buildSystemPrompt(config, opts) {
+	const { hasTools } = opts;
+	const agentInstructions = config.systemPrompt && config.systemPrompt !== DEFAULT_SYSTEM_PROMPT ? `\n\nAgent-Specific Instructions:\n${config.systemPrompt}` : "";
+	const toolPreamble = hasTools ? "\n\nWhen you decide to use a tool, ALWAYS say a brief natural phrase BEFORE the tool call (e.g. \"Let me look that up\" or \"One moment while I check\"). This fills silence while the tool executes. Keep preambles to one short sentence." : "";
+	const guidance = opts.toolGuidance && opts.toolGuidance.length > 0 ? `\n\nBuilt-in Tool Usage:\n${opts.toolGuidance.join("\n")}` : "";
+	return DEFAULT_SYSTEM_PROMPT + `\n\nToday's date is ${getFormattedDate()}.` + agentInstructions + toolPreamble + guidance + (opts.voice ? VOICE_RULES : "");
+}
+//#endregion
 //#region host/runtime-config.ts
 /**
 * Runtime dependencies injected into the session pipeline.
@@ -1306,7 +1605,7 @@ const jsonLogger = {
 };
 /** Default S2S endpoint configuration. */
 const DEFAULT_S2S_CONFIG = {
-	wssUrl: "wss://agents.assemblyai.com/v1/voice",
+	wssUrl: "wss://agents.assemblyai.com/v1/ws",
 	inputSampleRate: DEFAULT_STT_SAMPLE_RATE,
 	outputSampleRate: DEFAULT_TTS_SAMPLE_RATE
 };
@@ -1547,7 +1846,7 @@ function createSessionCore(opts) {
 */
 const yieldTick = () => new Promise((r) => setTimeout(r, 0));
 function buildToolContext(opts) {
-	const { env, state, kv, messages, sessionId } = opts;
+	const { env, state, kv, vector, messages, sessionId } = opts;
 	return {
 		env,
 		state: state ?? {},
@@ -1555,6 +1854,10 @@ function buildToolContext(opts) {
 			if (!kv) throw new Error("KV not available");
 			return kv;
 		},
+		get vector() {
+			if (!vector) throw new Error("Vector not available");
+			return vector;
+		},
 		messages: messages ?? [],
 		sessionId: sessionId ?? "",
 		send(event, data) {
@@ -2022,7 +2325,7 @@ function createPipelineTransport(opts) {
 //#region host/s2s.ts
 const uint8ToBase64 = (bytes) => Buffer.from(bytes).toString("base64");
 const base64ToUint8 = (base64) => new Uint8Array(Buffer.from(base64, "base64"));
-const defaultCreateS2sWebSocket = (url, opts) => new WsWebSocket(url, { headers: opts.headers });
+const defaultCreateS2sWebSocket = (url, opts) => new WebSocket(url, { headers: opts.headers });
 const S2sMessageSchema = z.discriminatedUnion("type", [
 	z.object({
 		type: z.literal("session.ready"),
@@ -2462,50 +2765,6 @@ function createS2sTransport(opts) {
 	};
 }
 //#endregion
-//#region host/unstorage-kv.ts
-/**
-* Key-value store backed by unstorage.
-*
-* Works with any unstorage driver (memory, fs, S3/R2, etc.).
-*/
-/**
-* Create a KV store backed by any unstorage driver.
-*
-* @param options - See {@link UnstorageKvOptions}.
-* @returns A {@link Kv} instance.
-*
-* @example
-* ```ts
-* import { createStorage } from "unstorage";
-* import { createUnstorageKv } from "@alexkroman1/aai/unstorage-kv";
-*
-* const kv = createUnstorageKv({ storage: createStorage() });
-* await kv.set("greeting", "hello");
-* const value = await kv.get<string>("greeting"); // "hello"
-* ```
-*/
-function createUnstorageKv(options) {
-	const store = options.prefix ? prefixStorage(options.storage, options.prefix) : options.storage;
-	return {
-		async get(key) {
-			return await store.getItem(key) ?? null;
-		},
-		async set(key, value, setOptions) {
-			if (JSON.stringify(value).length > 65536) throw new Error(`Value exceeds max size of ${MAX_VALUE_SIZE} bytes`);
-			const storable = value;
-			if (setOptions?.expireIn && setOptions.expireIn > 0) await store.setItem(key, storable, { ttl: Math.ceil(setOptions.expireIn / 1e3) });
-			else await store.setItem(key, storable);
-		},
-		async delete(keys) {
-			const keyArray = Array.isArray(keys) ? keys : [keys];
-			await Promise.all(keyArray.map((k) => store.removeItem(k)));
-		},
-		close() {
-			store.dispose();
-		}
-	};
-}
-//#endregion
 //#region host/ws-handler.ts
 /**
 * WebSocket session lifecycle handler.
@@ -2741,6 +3000,10 @@ function resolveLlmIfDescriptor(value, env) {
 function createLocalKv() {
 	return createUnstorageKv({ storage: createStorage() });
 }
+/** Create an in-memory Vector store (default for self-hosted). */
+function createLocalVector(slug) {
+	return createMemoryVector({ namespace: slug });
+}
 /**
 * Create an agent runtime — the execution engine for a voice agent.
 *
@@ -2754,8 +3017,11 @@ function createLocalKv() {
 * @public
 */
 function createRuntime(opts) {
-	const { agent, env, kv = createLocalKv(), createWebSocket, logger = consoleLogger, s2sConfig = DEFAULT_S2S_CONFIG, sessionStartTimeoutMs, shutdownTimeoutMs = DEFAULT_SHUTDOWN_TIMEOUT_MS } = opts;
+	const { agent, env, kv = createLocalKv(), vector, createWebSocket, logger = consoleLogger, s2sConfig = DEFAULT_S2S_CONFIG, sessionStartTimeoutMs, shutdownTimeoutMs = DEFAULT_SHUTDOWN_TIMEOUT_MS } = opts;
 	const mode = assertProviderTriple(opts.stt, opts.llm, opts.tts);
+	const slug = agent.name ?? "local";
+	const resolvedKv = agent.kv ? resolveKv(agent.kv, env, "") : kv;
+	const resolvedVector = agent.vector ? resolveVector(agent.vector, env, slug) : vector ?? createLocalVector(slug);
 	const agentConfig = toAgentConfig(agent);
 	const sessions = /* @__PURE__ */ new Map();
 	const sinkMap = /* @__PURE__ */ new Map();
@@ -2775,7 +3041,8 @@ function createRuntime(opts) {
 					tool,
 					env: frozenEnv,
 					sessionId: sessionId ?? "",
-					kv,
+					kv: resolvedKv,
+					vector: resolvedVector,
 					messages,
 					logger
 				});
@@ -2807,7 +3074,8 @@ function createRuntime(opts) {
 				env: frozenEnv,
 				state: getState(sessionId ?? ""),
 				sessionId: sessionId ?? "",
-				kv,
+				kv: resolvedKv,
+				vector: resolvedVector,
 				messages,
 				logger,
 				send: sink ? (event, data) => sink.event({
@@ -2982,6 +3250,47 @@ async function serveStatic(dir, req, res) {
 		return false;
 	}
 }
+function handleVectorPost(vector, req, res) {
+	let body = "";
+	req.on("data", (chunk) => {
+		body += chunk;
+	});
+	req.on("end", async () => {
+		try {
+			const json = JSON.parse(body);
+			const parsed = VectorRequestSchema.safeParse(json);
+			if (!parsed.success) {
+				res.statusCode = 400;
+				res.end(JSON.stringify({ error: parsed.error.message }));
+				return;
+			}
+			const op = parsed.data;
+			let result;
+			switch (op.op) {
+				case "upsert":
+					await vector.upsert(op.id, op.text, op.metadata);
+					result = "OK";
+					break;
+				case "query":
+					result = await vector.query(op.text, {
+						...op.topK !== void 0 ? { topK: op.topK } : {},
+						...op.filter !== void 0 ? { filter: op.filter } : {}
+					});
+					break;
+				case "delete":
+					await vector.delete(op.ids);
+					result = "OK";
+					break;
+				default: break;
+			}
+			res.statusCode = 200;
+			res.end(JSON.stringify({ result }));
+		} catch (err) {
+			res.statusCode = 500;
+			res.end(JSON.stringify({ error: err instanceof Error ? err.message : String(err) }));
+		}
+	});
+}
 function handleKvGet(kv, req, res) {
 	const key = new URL(req.url ?? "/", "http://localhost").searchParams.get("key");
 	if (!key) {
@@ -3008,7 +3317,7 @@ function handleKvGet(kv, req, res) {
 * @internal Used by aai-cli dev server.
 */
 function createServer(options) {
-	const { runtime, clientHtml, clientDir, logger = consoleLogger, kv } = options;
+	const { runtime, clientHtml, clientDir, logger = consoleLogger, kv, vector } = options;
 	const name = options.name ?? "agent";
 	if (clientHtml && clientDir) throw new Error("clientHtml and clientDir are mutually exclusive");
 	const escapedName = escapeHtml(name);
@@ -3031,6 +3340,10 @@ function createServer(options) {
 			handleKvGet(kv, req, res);
 			return;
 		}
+		if (vector && method === "POST" && url === "/vector") {
+			handleVectorPost(vector, req, res);
+			return;
+		}
 		handleRequest(req, res, url, method);
 	});
 	async function handleRequest(req, res, url, method) {
@@ -3089,4 +3402,4 @@ function createServer(options) {
 	};
 }
 //#endregion
-export { DEFAULT_S2S_CONFIG, _internals, consoleLogger, createPipelineTransport, createRuntime, createS2sTransport, createServer, createSessionCore, createUnstorageKv, executeInIsolate, executeToolCall, jsonLogger, resolveAllBuiltins, wireSessionSocket };
+export { DEFAULT_S2S_CONFIG, _internals, consoleLogger, createMemoryVector, createPineconeVector, createPipelineTransport, createRuntime, createS2sTransport, createServer, createSessionCore, createUnstorageKv, executeInIsolate, executeToolCall, jsonLogger, resolveAllBuiltins, resolveKv, resolveVector, wireSessionSocket };

package/dist/host/runtime.d.ts

@@ -12,6 +12,7 @@ import type { ClientSink } from "../sdk/protocol.ts";
 import { type ReadyConfig } from "../sdk/protocol.ts";
 import { type LlmProvider, type SttOpener, type SttProvider, type TtsOpener, type TtsProvider } from "../sdk/providers.ts";
 import type { AgentDef } from "../sdk/types.ts";
+import type { Vector } from "../sdk/vector.ts";
 import type { Logger, S2SConfig } from "./runtime-config.ts";
 import type { CreateS2sWebSocket } from "./s2s.ts";
 import { type SessionCore } from "./session-core.ts";
@@ -50,6 +51,11 @@ export type RuntimeOptions = {
     agent: AgentDef<any>;
     env: Record<string, string>;
     kv?: Kv | undefined;
+    /**
+     * Vector store. If omitted, an in-memory store is created. The
+     * runtime overrides this with `agent.vector` if set.
+     */
+    vector?: Vector | undefined;
     /** Custom WebSocket factory for the S2S connection (useful for testing). */
     createWebSocket?: CreateS2sWebSocket | undefined;
     logger?: Logger | undefined;