1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
|
import { db } from "@hoarder/db";
import logger from "@hoarder/shared/logger";
import serverConfig from "@hoarder/shared/config";
import {
OpenAIQueue,
ZOpenAIRequest,
queueConnectionDetails,
zOpenAIRequestSchema,
} from "@hoarder/shared/queues";
import { Job } from "bullmq";
import OpenAI from "openai";
import { z } from "zod";
import { Worker } from "bullmq";
import { bookmarkTags, bookmarks, tagsOnBookmarks } from "@hoarder/db/schema";
import { and, eq, inArray } from "drizzle-orm";
const openAIResponseSchema = z.object({
tags: z.array(z.string()),
});
export class OpenAiWorker {
static async build() {
logger.info("Starting openai worker ...");
const worker = new Worker<ZOpenAIRequest, void>(
OpenAIQueue.name,
runOpenAI,
{
connection: queueConnectionDetails,
autorun: false,
},
);
worker.on("completed", (job) => {
const jobId = job?.id || "unknown";
logger.info(`[openai][${jobId}] Completed successfully`);
});
worker.on("failed", (job, error) => {
const jobId = job?.id || "unknown";
logger.error(`[openai][${jobId}] openai job failed: ${error}`);
});
return worker;
}
}
const PROMPT_BASE = `
I'm building a read-it-later app and I need your help with automatic tagging.
Please analyze the text after the sentence "CONTENT:" and suggest relevant tags that describe its key themes, topics, and main ideas.
Aim for a variety of tags, including broad categories, specific keywords, and potential sub-genres. If it's a famous website
you may also include a tag for the website. Tags should be lowercases and don't contain spaces. If the tag is not generic enough, don't
include it. Aim for 3-5 tags. If there are no good tags, don't emit any. You must respond in JSON with the key "tags" and the value is list of tags.
CONTENT:
`;
function buildPrompt(
bookmark: NonNullable<Awaited<ReturnType<typeof fetchBookmark>>>,
) {
if (bookmark.link) {
if (!bookmark.link.description) {
throw new Error(
`No description found for link "${bookmark.id}". Skipping ...`,
);
}
return `
${PROMPT_BASE}
URL: ${bookmark.link.url}
Description: ${bookmark.link.description}
`;
}
if (bookmark.text) {
// TODO: Ensure that the content doesn't exceed the context length of openai
return `
${PROMPT_BASE}
${bookmark.text.text}
`;
}
throw new Error("Unknown bookmark type");
}
async function fetchBookmark(linkId: string) {
return await db.query.bookmarks.findFirst({
where: eq(bookmarks.id, linkId),
with: {
link: true,
text: true,
},
});
}
async function inferTags(
jobId: string,
bookmark: NonNullable<Awaited<ReturnType<typeof fetchBookmark>>>,
openai: OpenAI,
) {
const chatCompletion = await openai.chat.completions.create({
messages: [{ role: "system", content: buildPrompt(bookmark) }],
model: "gpt-3.5-turbo-0125",
response_format: { type: "json_object" },
});
const response = chatCompletion.choices[0].message.content;
if (!response) {
throw new Error(`[openai][${jobId}] Got no message content from OpenAI`);
}
try {
let tags = openAIResponseSchema.parse(JSON.parse(response)).tags;
logger.info(
`[openai][${jobId}] Inferring tag for bookmark "${bookmark.id}" used ${chatCompletion.usage?.total_tokens} tokens and inferred: ${tags}`,
);
// Sometimes the tags contain the hashtag symbol, let's strip them out if they do.
tags = tags.map((t) => {
if (t.startsWith("#")) {
return t.slice(1);
}
return t;
});
return tags;
} catch (e) {
throw new Error(
`[openai][${jobId}] Failed to parse JSON response from OpenAI: ${e}`,
);
}
}
async function createTags(tags: string[], userId: string) {
if (tags.length == 0) {
return [];
}
await db
.insert(bookmarkTags)
.values(
tags.map((t) => ({
name: t,
userId,
})),
)
.onConflictDoNothing();
const res = await db.query.bookmarkTags.findMany({
where: and(
eq(bookmarkTags.userId, userId),
inArray(bookmarkTags.name, tags),
),
columns: {
id: true,
},
});
return res.map((r) => r.id);
}
async function connectTags(bookmarkId: string, tagIds: string[]) {
if (tagIds.length == 0) {
return;
}
await db.insert(tagsOnBookmarks).values(
tagIds.map((tagId) => ({
tagId,
bookmarkId,
attachedBy: "ai" as const,
})),
);
}
async function runOpenAI(job: Job<ZOpenAIRequest, void>) {
const jobId = job.id || "unknown";
const { openAI } = serverConfig;
if (!openAI.apiKey) {
logger.debug(
`[openai][${jobId}] OpenAI is not configured, nothing to do now`,
);
return;
}
const openai = new OpenAI({
apiKey: openAI.apiKey,
});
const request = zOpenAIRequestSchema.safeParse(job.data);
if (!request.success) {
throw new Error(
`[openai][${jobId}] Got malformed job request: ${request.error.toString()}`,
);
}
const { bookmarkId } = request.data;
const bookmark = await fetchBookmark(bookmarkId);
if (!bookmark) {
throw new Error(
`[openai][${jobId}] bookmark with id ${bookmarkId} was not found`,
);
}
const tags = await inferTags(jobId, bookmark, openai);
const tagIds = await createTags(tags, bookmark.userId);
await connectTags(bookmarkId, tagIds);
}
|
