init

ts/packages/flow/src/embeddings/ollama.ts

@@ -0,0 +1,76 @@
+/**
+ * Ollama embeddings service.
+ *
+ * Simple HTTP POST to a local Ollama instance to generate embeddings.
+ * Extends EmbeddingsService from @trustgraph/base so it plugs into the
+ * flow processor framework (consumer/producer wiring is handled by the base class).
+ *
+ * Python reference: trustgraph-flow/trustgraph/embeddings/ollama/processor.py
+ */
+
+import {
+  EmbeddingsService,
+  type ProcessorConfig,
+} from "@trustgraph/base";
+
+export interface OllamaEmbeddingsConfig extends ProcessorConfig {
+  model?: string;
+  ollamaHost?: string;
+}
+
+interface OllamaEmbedResponse {
+  embeddings: number[][];
+}
+
+export class OllamaEmbeddingsProcessor extends EmbeddingsService {
+  private defaultModel: string;
+  private ollamaHost: string;
+
+  constructor(config: OllamaEmbeddingsConfig) {
+    super(config);
+
+    this.defaultModel = config.model ?? "mxbai-embed-large";
+    this.ollamaHost =
+      config.ollamaHost ??
+      process.env.OLLAMA_HOST ??
+      "http://localhost:11434";
+
+    console.log(
+      `[OllamaEmbeddings] Initialized (host=${this.ollamaHost}, model=${this.defaultModel})`,
+    );
+  }
+
+  async onEmbeddings(texts: string[], model?: string): Promise<number[][]> {
+    if (!texts || texts.length === 0) {
+      return [];
+    }
+
+    const useModel = model ?? this.defaultModel;
+
+    const url = `${this.ollamaHost}/api/embed`;
+
+    const response = await fetch(url, {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({
+        model: useModel,
+        input: texts,
+      }),
+    });
+
+    if (!response.ok) {
+      const body = await response.text();
+      throw new Error(
+        `Ollama embeddings request failed (${response.status}): ${body}`,
+      );
+    }
+
+    const data = (await response.json()) as OllamaEmbedResponse;
+
+    return data.embeddings;
+  }
+}
+
+export async function run(): Promise<void> {
+  await OllamaEmbeddingsProcessor.launch("embeddings");
+}