[Feature request] Force AI to use existing tags (instead of creating them)

apps/workers/openaiWorker.ts

@@ -1,5 +1,6 @@
 import { and, Column, eq, inArray, sql } from "drizzle-orm";
 import { DequeuedJob, Runner } from "liteque";
+import { buildImpersonatingTRPCClient } from "trpc";
 import { z } from "zod";
 
 import type { InferenceClient } from "@hoarder/shared/inference";
@@ -200,7 +201,47 @@ async function fetchCustomPrompts(
     },
   });
 
-  return prompts.map((p) => p.text);
+  let promptTexts = prompts.map((p) => p.text);
+  if (containsTagsPlaceholder(prompts)) {
+    promptTexts = await replaceTagsPlaceholders(promptTexts, userId);
+  }
+
+  return promptTexts;
+}
+
+async function replaceTagsPlaceholders(
+  prompts: string[],
+  userId: string,
+): Promise<string[]> {
+  const api = await buildImpersonatingTRPCClient(userId);
+  const tags = (await api.tags.list()).tags;
+  const tagsString = `[${tags.map((tag) => tag.name).join(",")}]`;
+  const aiTagsString = `[${tags
+    .filter((tag) => tag.numBookmarksByAttachedType.human ?? 0 == 0)
+    .map((tag) => tag.name)
+    .join(",")}]`;
+  const userTagsString = `[${tags
+    .filter((tag) => tag.numBookmarksByAttachedType.human ?? 0 > 0)
+    .map((tag) => tag.name)
+    .join(",")}]`;
+
+  return prompts.map((p) =>
+    p
+      .replaceAll("$tags", tagsString)
+      .replaceAll("$aiTags", aiTagsString)
+      .replaceAll("$userTags", userTagsString),
+  );
+}
+
+function containsTagsPlaceholder(prompts: { text: string }[]): boolean {
+  return (
+    prompts.filter(
+      (p) =>
+        p.text.includes("$tags") ||
+        p.text.includes("$aiTags") ||
+        p.text.includes("$userTags"),
+    ).length > 0
+  );
 }
 
 async function inferTagsFromPDF(

docs/docs/03-configuration.md

@@ -60,6 +60,11 @@ Either `OPENAI_API_KEY` or `OLLAMA_BASE_URL` need to be set for automatic taggin
 | INFERENCE_LANG            | No       | english     | The language in which the tags will be generated.                                                                                                                                                                                                                                                                                                                                     |
 | INFERENCE_JOB_TIMEOUT_SEC | No       | 30          | How long to wait for the inference job to finish before timing out. If you're running ollama without powerful GPUs, you might want to increase the timeout a bit.                                                                                                                                                                                                                     |
 
+:::info
+- You can append additional instructions to the prompt used for automatic tagging, in the `AI Settings` (in the `User Settings` screen)
+- You can use the placeholders `$tags`, `$aiTags`, `$userTags` in the prompt. These placeholders will be replaced with all tags, ai generated tags or human created tags when automatic tagging is performed (e.g. `[hoarder, computer, ai]`)
+:::
+
 ## Crawler Configs
 
 | Name                               | Required | Default | Description                                                                                                                                                                                                                                                                                                                                                                        |