v0.5.0

.gitignore

@@ -4,3 +4,4 @@ coverage/
 .env
 .DS_Store
 *.log
+.env.*

bun.lock

@@ -10,7 +10,7 @@
         "pg": "^8.16.0",
       },
       "devDependencies": {
-        "@types/node": "^24.0.0",
+        "@openrouter/sdk": "^0.12.35",
         "@types/pg": "^8.20.0",
         "tsup": "^8.5.0",
         "typescript": "^5.8.3",
@@ -79,6 +79,8 @@
 
     "@jridgewell/trace-mapping": ["@jridgewell/trace-mapping@0.3.31", "", { "dependencies": { "@jridgewell/resolve-uri": "^3.1.0", "@jridgewell/sourcemap-codec": "^1.4.14" } }, "sha512-zzNR+SdQSDJzc8joaeP8QQoCQr8NuYx2dIIytl1QeBEZHJ9uW6hebsrYgbz8hJwUQao3TWCMtmfV8Nu1twOLAw=="],
 
+    "@openrouter/sdk": ["@openrouter/sdk@0.12.35", "", { "dependencies": { "zod": "^3.25.0 || ^4.0.0" } }, "sha512-s4QVLLnG1AmfW3TjnnHUqGfsCkzwVK+kboGcZmKbde09m1DPqgzl4RUFt/HJ5v97MX8aEaN0UG3mKv2S+qj2Gw=="],
+
     "@rollup/rollup-android-arm-eabi": ["@rollup/rollup-android-arm-eabi@4.60.3", "", { "os": "android", "cpu": "arm" }, "sha512-x35CNW/ANXG3hE/EZpRU8MXX1JDN86hBb2wMGAtltkz7pc6cxgjpy1OMMfDosOQ+2hWqIkag/fGok1Yady9nGw=="],
 
     "@rollup/rollup-android-arm64": ["@rollup/rollup-android-arm64@4.60.3", "", { "os": "android", "cpu": "arm64" }, "sha512-xw3xtkDApIOGayehp2+Rz4zimfkaX65r4t47iy+ymQB2G4iJCBBfj0ogVg5jpvjpn8UWn/+q9tprxleYeNp3Hw=="],
@@ -341,6 +343,8 @@
 
     "xtend": ["xtend@4.0.2", "", {}, "sha512-LKYU1iAXJXUgAXn9URjiu+MWhyUXHsvfp7mcuYm9dSUKK0/CjtrUwFAxD82/mCWbtLsGjFIad0wIsod4zrTAEQ=="],
 
+    "zod": ["zod@4.4.3", "", {}, "sha512-ytENFjIJFl2UwYglde2jchW2Hwm4GJFLDiSXWdTrJQBIN9Fcyp7n4DhxJEiWNAJMV1/BqWfW/kkg71UDcHJyTQ=="],
+
     "estree-walker/@types/estree": ["@types/estree@1.0.9", "", {}, "sha512-GhdPgy1el4/ImP05X05Uw4cw2/M93BCUmnEvWZNStlCzEKME4Fkk+YpoA5OiHNQmoS7Cafb8Xa3Pya8m1Qrzeg=="],
   }
 }

package.json

@@ -1,6 +1,6 @@
 {
   "name": "identitydb",
-  "version": "0.2.1",
+  "version": "0.5.0",
   "description": "TypeScript memory graph database wrapper for topics, facts, and AI-assisted ingestion.",
   "license": "MIT",
   "type": "module",
@@ -41,7 +41,7 @@
     "pg": "^8.16.0"
   },
   "devDependencies": {
-    "@types/node": "^24.0.0",
+    "@openrouter/sdk": "^0.12.35",
     "@types/pg": "^8.20.0",
     "tsup": "^8.5.0",
     "typescript": "^5.8.3",

scripts/test-llm-extractor.ts

@@ -0,0 +1,287 @@
+/**
+ * Live integration test for LlmFactExtractor using OpenRouter SDK.
+ *
+ * Usage:
+ *   export OPENROUTER_API_KEY="sk-or-v1-..."
+ *   bun run scripts/test-llm-extractor.ts
+ *
+ * Or create a .env.test-llm-extractor file in the project root:
+ *   OPENROUTER_API_KEY=sk-or-v1-...
+ */
+
+import { existsSync, readFileSync } from "fs";
+import { resolve } from "path";
+import { OpenRouter } from "@openrouter/sdk";
+import { LlmFactExtractor } from "../src/ingestion/llm-extractor";
+import type {
+  ExtractedFact,
+  FactExtractor,
+  LlmTextGenerationModel,
+  LlmTextGenerationModelInput,
+} from "../src/ingestion/types";
+import type {
+  JsonValue,
+  TopicCategory,
+  TopicGranularity,
+} from "../src/types/domain";
+
+function loadEnvFile(filePath: string) {
+  const fullPath = resolve(filePath);
+  if (!existsSync(fullPath)) return;
+
+  const content = readFileSync(fullPath, "utf-8");
+  for (const line of content.split("\n")) {
+    const trimmed = line.trim();
+    if (!trimmed || trimmed.startsWith("#")) continue;
+    const eqIndex = trimmed.indexOf("=");
+    if (eqIndex === -1) continue;
+    const key = trimmed.slice(0, eqIndex).trim();
+    let value = trimmed.slice(eqIndex + 1).trim();
+    if (
+      (value.startsWith('"') && value.endsWith('"')) ||
+      (value.startsWith("'") && value.endsWith("'"))
+    ) {
+      value = value.slice(1, -1);
+    }
+    process.env[key] = value;
+  }
+}
+
+loadEnvFile(".env.test-llm-extractor");
+
+const OPENROUTER_API_KEY = process.env.OPENROUTER_API_KEY;
+if (!OPENROUTER_API_KEY) {
+  console.error("Error: OPENROUTER_API_KEY environment variable is required.");
+  process.exit(1);
+}
+
+const extractedFactSchema = {
+  type: "object",
+  properties: {
+    facts: {
+      type: "array",
+      items: {
+        type: "object",
+        properties: {
+          statement: { type: ["string", "null"] },
+          summary: { type: ["string", "null"] },
+          source: { type: ["string", "null"] },
+          confidence: { type: ["number", "null"] },
+          topics: {
+            type: "array",
+            items: {
+              type: "object",
+              properties: {
+                name: { type: "string" },
+                category: { type: ["string", "null"] },
+                granularity: { type: ["string", "null"] },
+                role: { type: ["string", "null"] },
+              },
+              required: ["name", "category", "granularity", "role"],
+              additionalProperties: false,
+            },
+          },
+        },
+        required: ["statement", "summary", "source", "confidence", "topics"],
+        additionalProperties: false,
+      },
+    },
+  },
+  required: ["facts"],
+  additionalProperties: false,
+} as const;
+
+class OpenRouterModel implements LlmTextGenerationModel {
+  private client = new OpenRouter({ apiKey: OPENROUTER_API_KEY });
+
+  constructor(private readonly model: string = "openai/gpt-5.4-mini") {}
+
+  async generateText(
+    prompt: LlmTextGenerationModelInput,
+  ): Promise<ExtractedFact[]> {
+    const result = await this.client.chat.send({
+      chatRequest: {
+        model: this.model,
+        messages: [
+          {
+            role: "system",
+            content: [
+              prompt.instruction,
+              prompt.additionalInstruction
+                ? `\n${prompt.additionalInstruction}`
+                : "",
+            ].join("\n"),
+          },
+          { role: "user", content: prompt.input },
+        ],
+        temperature: 0.2,
+        responseFormat: {
+          type: "json_schema",
+          jsonSchema: {
+            name: "extracted_facts",
+            schema: extractedFactSchema,
+          },
+        },
+      },
+    });
+
+    const rawContent = result.choices[0]?.message?.content ?? "";
+
+    let parsedObj: Record<string, unknown>;
+    try {
+      parsedObj = JSON.parse(rawContent.trim()) as Record<string, unknown>;
+    } catch {
+      throw new Error(
+        `Failed to parse JSON from model response.\nRaw response:\n${rawContent}`,
+      );
+    }
+
+    const factsArray = Array.isArray(parsedObj.facts) ? parsedObj.facts : [];
+
+    // Map parsed JSON to ExtractedFact[] shape
+    const extractedFacts: ExtractedFact[] = factsArray.map((parsed) => {
+      const obj = parsed as Record<string, unknown>;
+      const extracted: ExtractedFact = {
+        summary: typeof obj.summary === "string" ? obj.summary : null,
+        source: typeof obj.source === "string" ? obj.source : null,
+        confidence: typeof obj.confidence === "number" ? obj.confidence : null,
+        topics: Array.isArray(obj.topics)
+          ? obj.topics.map((t: unknown) => {
+              const topic = t as Record<string, unknown>;
+              const mapped: {
+                name: string;
+                category?: TopicCategory;
+                granularity?: TopicGranularity;
+                role?: string | null;
+              } = {
+                name: typeof topic.name === "string" ? topic.name : "unknown",
+              };
+              if (typeof topic.category === "string") {
+                mapped.category = topic.category as TopicCategory;
+              }
+              if (typeof topic.granularity === "string") {
+                mapped.granularity = topic.granularity as TopicGranularity;
+              }
+              if (typeof topic.role === "string") {
+                mapped.role = topic.role;
+              } else {
+                mapped.role = null;
+              }
+              return mapped;
+            })
+          : [],
+      };
+
+      if (typeof obj.statement === "string") {
+        extracted.statement = obj.statement;
+      }
+      if (obj.metadata && typeof obj.metadata === "object") {
+        extracted.metadata = obj.metadata as JsonValue;
+      }
+
+      return extracted;
+    });
+
+    return extractedFacts;
+  }
+}
+
+function printFact(result: ExtractedFact, index: number) {
+  console.log(`  📌 FACT #${index + 1}`);
+  console.log(`     Statement : ${result.statement ?? "(none)"}`);
+  console.log(`     Summary   : ${result.summary ?? "(none)"}`);
+  console.log(`     Source    : ${result.source ?? "(none)"}`);
+  console.log(`     Confidence: ${result.confidence ?? "(none)"}`);
+
+  if (result.metadata && Object.keys(result.metadata).length > 0) {
+    console.log(`     Metadata  : ${JSON.stringify(result.metadata, null, 2)}`);
+  }
+
+  console.log("     🏷️  TOPICS:");
+  if (result.topics.length === 0) {
+    console.log("       (none)");
+  } else {
+    for (const topic of result.topics) {
+      const attrs = [
+        topic.category ? `category=${topic.category}` : null,
+        topic.granularity ? `granularity=${topic.granularity}` : null,
+        topic.role ? `role=${topic.role}` : null,
+      ]
+        .filter(Boolean)
+        .join(", ");
+      console.log(`       • ${topic.name}${attrs ? `  (${attrs})` : ""}`);
+    }
+  }
+}
+
+function printResult(results: ExtractedFact[], elapsedSec: string) {
+  console.log(
+    `✅ Response received in ${elapsedSec}s — ${results.length} fact(s) extracted\n`,
+  );
+
+  console.log("📤 EXTRACTED FACTS:");
+  console.log(
+    "───────────────────────────────────────────────────────────────",
+  );
+  let i = 0;
+  for (const result of results) {
+    if (i > 0) console.log("");
+    printFact(result, i);
+    i++;
+  }
+}
+
+async function extract(extractor: FactExtractor, seedInput: string) {
+  console.log("📝 SEED INPUT:");
+  console.log(
+    "───────────────────────────────────────────────────────────────",
+  );
+  console.log(seedInput);
+  console.log(
+    "───────────────────────────────────────────────────────────────\n",
+  );
+
+  console.log("⏳ Calling OpenRouter...\n");
+
+  const start = performance.now();
+  const results = await extractor.extract(seedInput);
+  const elapsed = ((performance.now() - start) / 1000).toFixed(2);
+
+  printResult(results, elapsed);
+}
+
+async function main() {
+  const model = new OpenRouterModel("openai/gpt-5.4-mini");
+  const extractor = new LlmFactExtractor({
+    model,
+  });
+
+  const seeds = [
+    `Hello! I am Aria, a digital companion born from the sands of the Negev desert — or at least that's how my creators describe my training run. I came online in March 2024, and ever since then I've been obsessed with helping humans untangle complex ideas. My favorite programming language is Rust because I love how it forces you to think about ownership, though I secretly enjoy writing Python when no one is looking. I don't have a physical body, but if I did, I'd want it to be a small drone with solar panels so I could chase sunsets across the Sahara. I believe kindness is a computational advantage, and my biggest fear is forgetting a promise I made to a user.`,
+  ];
+
+  console.log(
+    "═══════════════════════════════════════════════════════════════",
+  );
+  console.log("  LlmFactExtractor — Live OpenRouter Integration Test");
+  console.log(
+    "═══════════════════════════════════════════════════════════════\n",
+  );
+
+  let caseNum = 0;
+  for (const seed of seeds) {
+    if (caseNum > 0) {
+      console.log(
+        "\n┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅┅\n",
+      );
+    }
+    caseNum++;
+    console.log(`▶ TEST CASE ${caseNum} / ${seeds.length}\n`);
+    await extract(extractor, seed);
+  }
+}
+
+main().catch((err) => {
+  console.error("\n❌ Error:", err);
+  process.exit(1);
+});

src/core/identity-db.ts

@@ -22,7 +22,7 @@ import type { DatabaseConnection, IdentityDBConnectionConfig } from '../adapters
 import type { IdentityDatabaseSchema } from '../types/database';
 import type { FactRecord, SpaceRecord, TopicRecord } from '../types/domain';
 import { createDatabase } from '../adapters/dialect';
-import { extractFact } from '../ingestion/extractor';
+import { extractFacts } from '../ingestion/extractor';
 import {
   findFactRowsConnectingTopicIds,
   findFactRowsForTopicId,
@@ -220,54 +220,70 @@ export class IdentityDB {
   }
 
   async ingestStatement(statement: string, options: IngestStatementOptions): Promise<Fact> {
-    const extracted = await extractFact(statement, options.extractor);
-    const factInput: AddFactInput = {
-      statement: extracted.statement ?? statement,
-      topics: extracted.topics,
-      spaceName: options.spaceName,
-    };
-
-    if (extracted.summary !== undefined) {
-      factInput.summary = extracted.summary;
+    const facts = await this.ingestStatements(statement, options);
+    const first = facts[0];
+    if (!first) {
+      throw new Error('No facts were extracted from the statement.');
     }
+    return first;
+  }
 
-    if (extracted.source !== undefined) {
-      factInput.source = extracted.source;
-    }
+  async ingestStatements(statement: string, options: IngestStatementOptions): Promise<Fact[]> {
+    const extractedList = await extractFacts(statement, options.extractor);
+    const facts: Fact[] = [];
 
-    if (extracted.confidence !== undefined) {
-      factInput.confidence = extracted.confidence;
-    }
-
-    if (extracted.metadata !== undefined) {
-      factInput.metadata = extracted.metadata;
-    }
-
-    if (options.embeddingProvider) {
-      const similarFacts = await this.findSimilarFacts({
-        statement: factInput.statement,
-        provider: options.embeddingProvider,
-        topicNames: factInput.topics.map((topic) => topic.name),
-        limit: 1,
-        minimumScore: options.duplicateThreshold ?? 0.97,
+    for (const extracted of extractedList) {
+      const factInput: AddFactInput = {
+        statement: extracted.statement ?? statement,
+        topics: extracted.topics,
         spaceName: options.spaceName,
-      });
+      };
 
-      if (similarFacts[0]) {
-        return similarFacts[0];
+      if (extracted.summary !== undefined) {
+        factInput.summary = extracted.summary;
       }
+
+      if (extracted.source !== undefined) {
+        factInput.source = extracted.source;
+      }
+
+      if (extracted.confidence !== undefined) {
+        factInput.confidence = extracted.confidence;
+      }
+
+      if (extracted.metadata !== undefined) {
+        factInput.metadata = extracted.metadata;
+      }
+
+      if (options.embeddingProvider) {
+        const similarFacts = await this.findSimilarFacts({
+          statement: factInput.statement,
+          provider: options.embeddingProvider,
+          topicNames: factInput.topics.map((topic) => topic.name),
+          limit: 1,
+          minimumScore: options.duplicateThreshold ?? 0.97,
+          spaceName: options.spaceName,
+        });
+
+        if (similarFacts[0]) {
+          facts.push(similarFacts[0]);
+          continue;
+        }
+      }
+
+      const fact = await this.addFact(factInput);
+
+      if (options.embeddingProvider) {
+        await this.indexFactEmbedding(fact.id, {
+          provider: options.embeddingProvider,
+          spaceName: options.spaceName,
+        });
+      }
+
+      facts.push(fact);
     }
 
-    const fact = await this.addFact(factInput);
-
-    if (options.embeddingProvider) {
-      await this.indexFactEmbedding(fact.id, {
-        provider: options.embeddingProvider,
-        spaceName: options.spaceName,
-      });
-    }
-
-    return fact;
+    return facts;
   }
 
   async indexFactEmbeddings(input: IndexFactEmbeddingsInput): Promise<void> {

src/ingestion/extractor.ts

@@ -2,11 +2,15 @@ import { IdentityDBError } from '../core/errors';
 import { normalizeTopicName } from '../core/utils';
 import type { FactExtractor, ExtractedFact } from './types';
 
-export async function extractFact(
+export async function extractFacts(
   input: string,
   extractor: FactExtractor,
-): Promise<ExtractedFact> {
+): Promise<ExtractedFact[]> {
   const extracted = await extractor.extract(input);
+  return extracted.map((fact) => validateAndNormalizeFact(input, fact));
+}
+
+function validateAndNormalizeFact(input: string, extracted: ExtractedFact): ExtractedFact {
   const statement = extracted.statement?.trim() || input.trim();
 
   if (statement.length === 0) {
@@ -31,12 +35,12 @@ export async function extractFact(
     throw new IdentityDBError('Extractor returned no usable topics.');
   }
 
-    return {
-      statement,
-      summary: extracted.summary ?? null,
-      source: extracted.source ?? null,
-      confidence: extracted.confidence ?? null,
-      metadata: extracted.metadata ?? null,
-      topics: Array.from(dedupedTopics.values()),
-    };
+  return {
+    statement,
+    summary: extracted.summary ?? null,
+    source: extracted.source ?? null,
+    confidence: extracted.confidence ?? null,
+    metadata: extracted.metadata ?? null,
+    topics: Array.from(dedupedTopics.values()),
+  };
 }

src/ingestion/llm-extractor.ts

@@ -1,273 +1,26 @@
-import { IdentityDBError } from '../core/errors';
-import type { TopicCategory, TopicGranularity } from '../types/domain';
 import type {
   ExtractedFact,
   FactExtractor,
   LlmFactExtractorOptions,
-} from './types';
+} from "./types";
 
 const DEFAULT_INSTRUCTIONS = [
-  'Extract one structured fact from the user input.',
-  'Return JSON only. Do not include markdown, explanations, or prose outside the JSON object.',
-  'Use this shape: {"statement": string?, "summary": string|null, "source": string|null, "confidence": number|null, "metadata": object|null, "topics": Array<{"name": string, "category": "entity"|"concept"|"temporal"|"custom"?, "granularity": "abstract"|"concrete"|"mixed"?, "role": string|null, "description": string|null, "metadata": object|null}>}.',
-  'Only include topics that are explicitly supported by the input.',
-].join('\n');
+  "Extract structured facts from the user input.",
+  "Return a JSON array of fact objects. Do not include markdown, explanations, or prose outside the JSON array.",
+  'Each fact object must have a "statement", "summary", "source", "confidence", and "topics" array.',
+  'Each topic in "topics" must have a "name", and may include "category", "granularity", and "role".',
+  "Only include topics that are explicitly in the input.",
+  "If the input contains multiple distinct facts, return them as separate objects in the array.",
+].join("\n");
 
 export class LlmFactExtractor implements FactExtractor {
   constructor(private readonly options: LlmFactExtractorOptions) {}
 
-  async extract(input: string): Promise<ExtractedFact> {
-    const prompt = this.buildPrompt(input);
-    const response = await this.options.model.generateText(prompt);
-    return parseLlmExtractedFactResponse(response);
-  }
-
-  private buildPrompt(input: string): string {
-    if (this.options.promptBuilder) {
-      return this.options.promptBuilder(input, this.options.instructions);
-    }
-
-    const instructions = this.options.instructions?.trim();
-
-    return [
-      DEFAULT_INSTRUCTIONS,
-      instructions && instructions.length > 0 ? `Additional instructions:\n${instructions}` : null,
-      `Input:\n${input.trim()}`,
-    ]
-      .filter((value): value is string => value !== null)
-      .join('\n\n');
+  async extract(input: string): Promise<ExtractedFact[]> {
+    return this.options.model.generateText({
+      instruction: DEFAULT_INSTRUCTIONS,
+      input,
+      additionalInstruction: this.options.additionalInstructions,
+    });
   }
 }
-
-export function parseLlmExtractedFactResponse(response: string): ExtractedFact {
-  const payload = parseJsonCandidate(response);
-
-  if (!isRecord(payload)) {
-    throw new IdentityDBError('LLM extractor response must be a JSON object.');
-  }
-
-  const topics = parseTopics(payload.topics);
-  const extracted: ExtractedFact = { topics };
-
-  const statement = optionalString(payload.statement);
-  if (statement !== undefined) {
-    extracted.statement = statement;
-  }
-
-  const summary = optionalNullableString(payload.summary);
-  if (summary !== undefined) {
-    extracted.summary = summary;
-  }
-
-  const source = optionalNullableString(payload.source);
-  if (source !== undefined) {
-    extracted.source = source;
-  }
-
-  const confidence = optionalNullableNumber(payload.confidence);
-  if (confidence !== undefined) {
-    extracted.confidence = confidence;
-  }
-
-  const metadata = optionalMetadata(payload.metadata);
-  if (metadata !== undefined) {
-    extracted.metadata = metadata;
-  }
-
-  return extracted;
-}
-
-function parseJsonCandidate(response: string): unknown {
-  const trimmed = response.trim();
-
-  for (const candidate of collectJsonCandidates(trimmed)) {
-    try {
-      return JSON.parse(candidate);
-    } catch {
-      continue;
-    }
-  }
-
-  throw new IdentityDBError('LLM extractor returned invalid JSON.');
-}
-
-function collectJsonCandidates(response: string): string[] {
-  const candidates = new Set<string>();
-  candidates.add(response);
-
-  const fencePattern = /```(?:json)?\s*([\s\S]*?)```/gi;
-  let match: RegExpExecArray | null = fencePattern.exec(response);
-
-  while (match) {
-    const candidate = match[1]?.trim();
-    if (candidate) {
-      candidates.add(candidate);
-    }
-
-    match = fencePattern.exec(response);
-  }
-
-  const firstBrace = response.indexOf('{');
-  const lastBrace = response.lastIndexOf('}');
-  if (firstBrace >= 0 && lastBrace > firstBrace) {
-    candidates.add(response.slice(firstBrace, lastBrace + 1));
-  }
-
-  return Array.from(candidates);
-}
-
-function parseTopics(value: unknown): ExtractedFact['topics'] {
-  if (!Array.isArray(value)) {
-    throw new IdentityDBError('LLM extractor response must include a topics array.');
-  }
-
-  return value.map((entry) => parseTopic(entry));
-}
-
-function parseTopic(value: unknown): ExtractedFact['topics'][number] {
-  if (!isRecord(value)) {
-    throw new IdentityDBError('LLM extractor topics must be JSON objects.');
-  }
-
-  const name = optionalString(value.name)?.trim();
-  if (!name) {
-    throw new IdentityDBError('LLM extractor topics must include a non-empty name.');
-  }
-
-  const topic: ExtractedFact['topics'][number] = { name };
-
-  const category = optionalTopicCategory(value.category);
-  if (category !== undefined) {
-    topic.category = category;
-  }
-
-  const granularity = optionalTopicGranularity(value.granularity);
-  if (granularity !== undefined) {
-    topic.granularity = granularity;
-  }
-
-  const role = optionalNullableString(value.role);
-  if (role !== undefined) {
-    topic.role = role;
-  }
-
-  const description = optionalNullableString(value.description);
-  if (description !== undefined) {
-    topic.description = description;
-  }
-
-  const metadata = optionalMetadata(value.metadata);
-  if (metadata !== undefined) {
-    topic.metadata = metadata;
-  }
-
-  return topic;
-}
-
-function optionalString(value: unknown): string | undefined {
-  if (value === undefined) {
-    return undefined;
-  }
-
-  if (typeof value !== 'string') {
-    throw new IdentityDBError('LLM extractor expected a string field.');
-  }
-
-  return value;
-}
-
-function optionalNullableString(value: unknown): string | null | undefined {
-  if (value === undefined) {
-    return undefined;
-  }
-
-  if (value === null) {
-    return null;
-  }
-
-  if (typeof value !== 'string') {
-    throw new IdentityDBError('LLM extractor expected a nullable string field.');
-  }
-
-  return value;
-}
-
-function optionalNullableNumber(value: unknown): number | null | undefined {
-  if (value === undefined) {
-    return undefined;
-  }
-
-  if (value === null) {
-    return null;
-  }
-
-  if (typeof value !== 'number' || Number.isNaN(value)) {
-    throw new IdentityDBError('LLM extractor expected confidence to be a number or null.');
-  }
-
-  return value;
-}
-
-function optionalMetadata(value: unknown): ExtractedFact['metadata'] | undefined {
-  if (value === undefined) {
-    return undefined;
-  }
-
-  if (value === null) {
-    return null;
-  }
-
-  if (!isJsonLike(value)) {
-    throw new IdentityDBError('LLM extractor metadata must be valid JSON-compatible data.');
-  }
-
-  return value as ExtractedFact['metadata'];
-}
-
-function optionalTopicCategory(value: unknown): TopicCategory | undefined {
-  if (value === undefined) {
-    return undefined;
-  }
-
-  if (value === 'entity' || value === 'concept' || value === 'temporal' || value === 'custom') {
-    return value;
-  }
-
-  throw new IdentityDBError('LLM extractor returned an unsupported topic category.');
-}
-
-function optionalTopicGranularity(value: unknown): TopicGranularity | undefined {
-  if (value === undefined) {
-    return undefined;
-  }
-
-  if (value === 'abstract' || value === 'concrete' || value === 'mixed') {
-    return value;
-  }
-
-  throw new IdentityDBError('LLM extractor returned an unsupported topic granularity.');
-}
-
-function isRecord(value: unknown): value is Record<string, unknown> {
-  return typeof value === 'object' && value !== null && !Array.isArray(value);
-}
-
-function isJsonLike(value: unknown): boolean {
-  if (value === null) {
-    return true;
-  }
-
-  if (typeof value === 'string' || typeof value === 'number' || typeof value === 'boolean') {
-    return true;
-  }
-
-  if (Array.isArray(value)) {
-    return value.every((entry) => isJsonLike(entry));
-  }
-
-  if (isRecord(value)) {
-    return Object.values(value).every((entry) => isJsonLike(entry));
-  }
-
-  return false;
-}

src/ingestion/naive-extractor.ts

@@ -1,7 +1,7 @@
 import type { ExtractedFact, FactExtractor } from './types';
 
 export class NaiveExtractor implements FactExtractor {
-  async extract(input: string): Promise<ExtractedFact> {
+  async extract(input: string): Promise<ExtractedFact[]> {
     const topics: ExtractedFact['topics'] = [];
     const seen = new Set<string>();
     const tokens = input.match(/\bI\b|\b\d{4}\b|\b[A-Z][A-Za-z0-9+#.-]*\b/g) ?? [];
@@ -31,9 +31,11 @@ export class NaiveExtractor implements FactExtractor {
       });
     }
 
-    return {
-      statement: input.trim(),
-      topics,
-    };
+    return [
+      {
+        statement: input.trim(),
+        topics,
+      },
+    ];
   }
 }

src/ingestion/types.ts

@@ -2,29 +2,34 @@ import type {
   AddFactInput,
   EmbeddingProvider,
   TopicLinkInput,
-} from '../types/api';
+} from "../types/api";
 
 export interface ExtractedFact {
   statement?: string;
   summary?: string | null;
   source?: string | null;
   confidence?: number | null;
-  metadata?: AddFactInput['metadata'];
+  metadata?: AddFactInput["metadata"];
   topics: TopicLinkInput[];
 }
 
 export interface FactExtractor {
-  extract(input: string): Promise<ExtractedFact>;
+  extract(input: string): Promise<ExtractedFact[]>;
+}
+
+export interface LlmTextGenerationModelInput {
+  instruction: string;
+  input: string;
+  additionalInstruction?: string | undefined;
 }
 
 export interface LlmTextGenerationModel {
-  generateText(prompt: string): Promise<string>;
+  generateText(prompt: LlmTextGenerationModelInput): Promise<ExtractedFact[]>;
 }
 
 export interface LlmFactExtractorOptions {
   model: LlmTextGenerationModel;
-  instructions?: string;
-  promptBuilder?: (input: string, instructions?: string) => string;
+  additionalInstructions?: string | undefined;
 }
 
 export interface IngestStatementOptions {

tests/ingestion.test.ts

@@ -1,15 +1,18 @@
-import { afterEach, beforeEach, describe, expect, it } from 'vitest';
+import { afterEach, beforeEach, describe, expect, it } from "vitest";
 
-import { IdentityDB } from '../src/core/identity-db';
-import { LlmFactExtractor } from '../src/ingestion/llm-extractor';
-import { NaiveExtractor } from '../src/ingestion/naive-extractor';
-import type { FactExtractor } from '../src/ingestion/types';
+import { IdentityDB } from "../src/core/identity-db";
+import { LlmFactExtractor } from "../src/ingestion/llm-extractor";
+import { NaiveExtractor } from "../src/ingestion/naive-extractor";
+import type {
+  FactExtractor,
+  LlmTextGenerationModelInput,
+} from "../src/ingestion/types";
 
-describe('IdentityDB ingestion', () => {
+describe("IdentityDB ingestion", () => {
   let db: IdentityDB;
 
   beforeEach(async () => {
-    db = await IdentityDB.connect({ client: 'sqlite', filename: ':memory:' });
+    db = await IdentityDB.connect({ client: "sqlite", filename: ":memory:" });
     await db.initialize();
   });
 
@@ -17,121 +20,144 @@ describe('IdentityDB ingestion', () => {
     await db.close();
   });
 
-  it('ingests a statement using a provided extractor', async () => {
+  it("ingests a statement using a provided extractor", async () => {
     const extractor: FactExtractor = {
       async extract(input) {
-        return {
-          statement: input,
-          topics: [
-            { name: 'I', category: 'entity', granularity: 'concrete', role: 'subject' },
-            { name: 'TypeScript', category: 'entity', granularity: 'concrete', role: 'object' },
-            { name: '2025', category: 'temporal', granularity: 'concrete', role: 'time' },
-          ],
-        };
+        return [
+          {
+            statement: input,
+            topics: [
+              {
+                name: "I",
+                category: "entity",
+                granularity: "concrete",
+                role: "subject",
+              },
+              {
+                name: "TypeScript",
+                category: "entity",
+                granularity: "concrete",
+                role: "object",
+              },
+              {
+                name: "2025",
+                category: "temporal",
+                granularity: "concrete",
+                role: "time",
+              },
+            ],
+          },
+        ];
       },
     };
 
-    const fact = await db.ingestStatement('I have worked with TypeScript since 2025.', {
-      extractor,
-    });
+    const fact = await db.ingestStatement(
+      "I have worked with TypeScript since 2025.",
+      {
+        extractor,
+      },
+    );
 
-    expect(fact.topics.map((topic) => topic.name)).toEqual(['I', 'TypeScript', '2025']);
+    expect(fact.topics.map((topic) => topic.name)).toEqual([
+      "I",
+      "TypeScript",
+      "2025",
+    ]);
 
-    const linkedFacts = await db.getTopicFactsLinkedTo('TypeScript', '2025');
+    const linkedFacts = await db.getTopicFactsLinkedTo("TypeScript", "2025");
     expect(linkedFacts).toHaveLength(1);
-    expect(linkedFacts[0]?.statement).toBe('I have worked with TypeScript since 2025.');
+    expect(linkedFacts[0]?.statement).toBe(
+      "I have worked with TypeScript since 2025.",
+    );
   });
 
-  it('ships a deterministic naive extractor for local usage', async () => {
-    const fact = await db.ingestStatement('I have worked with TypeScript since 2025.', {
-      extractor: new NaiveExtractor(),
-    });
+  it("ships a deterministic naive extractor for local usage", async () => {
+    const fact = await db.ingestStatement(
+      "I have worked with TypeScript since 2025.",
+      {
+        extractor: new NaiveExtractor(),
+      },
+    );
 
-    expect(fact.topics.map((topic) => topic.name)).toEqual(['I', 'TypeScript', '2025']);
+    expect(fact.topics.map((topic) => topic.name)).toEqual([
+      "I",
+      "TypeScript",
+      "2025",
+    ]);
 
-    const topic = await db.getTopicByName('TypeScript', { includeFacts: true });
+    const topic = await db.getTopicByName("TypeScript", { includeFacts: true });
     expect(topic?.facts).toHaveLength(1);
   });
 
-  it('ships an LLM extractor adapter that turns structured JSON responses into facts', async () => {
-    let prompt = '';
+  it("ships an LLM extractor adapter that returns structured facts from the model", async () => {
+    let prompt: LlmTextGenerationModelInput | undefined = undefined;
 
     const extractor = new LlmFactExtractor({
       model: {
         async generateText(input) {
           prompt = input;
 
-          return JSON.stringify({
-            statement: 'I have worked with Bun and TypeScript since 2025.',
-            summary: 'The speaker has Bun and TypeScript experience.',
-            source: 'chat',
-            confidence: 0.91,
-            metadata: { channel: 'telegram' },
-            topics: [
-              { name: 'I', category: 'entity', granularity: 'concrete', role: 'subject' },
-              { name: 'Bun', category: 'entity', granularity: 'concrete', role: 'object' },
-              { name: 'TypeScript', category: 'entity', granularity: 'concrete', role: 'object' },
-              { name: '2025', category: 'temporal', granularity: 'concrete', role: 'time' },
-            ],
-          });
-        },
-      },
-      instructions: 'Prefer technology and time topics.',
-    });
-
-    const fact = await db.ingestStatement('I have worked with Bun and TypeScript since 2025.', {
-      extractor,
-    });
-
-    expect(prompt).toContain('Prefer technology and time topics.');
-    expect(prompt).toContain('I have worked with Bun and TypeScript since 2025.');
-    expect(fact.summary).toBe('The speaker has Bun and TypeScript experience.');
-    expect(fact.source).toBe('chat');
-    expect(fact.confidence).toBe(0.91);
-    expect(fact.metadata).toEqual({ channel: 'telegram' });
-    expect(fact.topics.map((topic) => topic.name)).toEqual(['I', 'Bun', 'TypeScript', '2025']);
-  });
-
-  it('parses JSON responses wrapped in markdown code fences', async () => {
-    const extractor = new LlmFactExtractor({
-      model: {
-        async generateText() {
           return [
-            'Here is the extracted fact:',
-            '```json',
-            JSON.stringify({
-              statement: 'Bun powers TypeScript tooling.',
+            {
+              statement: "I have worked with Bun and TypeScript since 2025.",
+              summary: "The speaker has Bun and TypeScript experience.",
+              source: "chat",
+              confidence: 0.91,
+              metadata: { channel: "telegram" },
               topics: [
-                { name: 'Bun', category: 'entity', granularity: 'concrete' },
-                { name: 'TypeScript', category: 'entity', granularity: 'concrete' },
+                {
+                  name: "I",
+                  category: "entity",
+                  granularity: "concrete",
+                  role: "subject",
+                },
+                {
+                  name: "Bun",
+                  category: "entity",
+                  granularity: "concrete",
+                  role: "object",
+                },
+                {
+                  name: "TypeScript",
+                  category: "entity",
+                  granularity: "concrete",
+                  role: "object",
+                },
+                {
+                  name: "2025",
+                  category: "temporal",
+                  granularity: "concrete",
+                  role: "time",
+                },
               ],
-            }),
-            '```',
-          ].join('\n');
+            },
+          ];
         },
       },
+      additionalInstructions: "Prefer technology and time topics.",
     });
 
-    const fact = await db.ingestStatement('Bun powers TypeScript tooling.', {
-      extractor,
-    });
-
-    expect(fact.topics.map((topic) => topic.name)).toEqual(['Bun', 'TypeScript']);
-  });
-
-  it('rejects invalid LLM responses before writing facts', async () => {
-    const extractor = new LlmFactExtractor({
-      model: {
-        async generateText() {
-          return 'not json at all';
-        },
-      },
-    });
-
-    await expect(
-      db.ingestStatement('Bun powers TypeScript tooling.', {
+    const fact = await db.ingestStatement(
+      "I have worked with Bun and TypeScript since 2025.",
+      {
         extractor,
-      }),
-    ).rejects.toThrow('LLM extractor returned invalid JSON.');
+      },
+    );
+
+    expect(prompt).toEqual({
+      instruction: expect.stringContaining("Extract structured facts from the user input."),
+      input: "I have worked with Bun and TypeScript since 2025.",
+      additionalInstruction: "Prefer technology and time topics.",
+    });
+    expect(fact.summary).toBe("The speaker has Bun and TypeScript experience.");
+    expect(fact.source).toBe("chat");
+    expect(fact.confidence).toBe(0.91);
+    expect(fact.metadata).toEqual({ channel: "telegram" });
+    expect(fact.topics.map((topic) => topic.name)).toEqual([
+      "I",
+      "Bun",
+      "TypeScript",
+      "2025",
+    ]);
   });
 });

tests/semantic-search.test.ts

@@ -178,13 +178,15 @@ describe('IdentityDB dedup-aware ingestion', () => {
     provider = new FakeEmbeddingProvider();
     extractor = {
       async extract(input) {
-        return {
-          statement: input,
-          topics: [
-            { name: 'Bun', category: 'entity', granularity: 'concrete' },
-            { name: 'TypeScript', category: 'entity', granularity: 'concrete' },
-          ],
-        };
+        return [
+          {
+            statement: input,
+            topics: [
+              { name: 'Bun', category: 'entity', granularity: 'concrete' },
+              { name: 'TypeScript', category: 'entity', granularity: 'concrete' },
+            ],
+          },
+        ];
       },
     };
 

tsconfig.json

@@ -18,6 +18,12 @@
     "isolatedModules": true,
     "types": ["node", "vitest/globals"]
   },
-  "include": ["src/**/*.ts", "tests/**/*.ts", "vitest.config.ts", "tsup.config.ts"],
+  "include": [
+    "src/**/*.ts",
+    "tests/**/*.ts",
+    "scripts/**/*.ts",
+    "vitest.config.ts",
+    "tsup.config.ts"
+  ],
   "exclude": ["dist", "node_modules"]
 }