Skip to content

FEATURES.md

Latest commit

 

History

History
94 lines (64 loc) · 3.98 KB

FEATURES.md

File metadata and controls

94 lines (64 loc) · 3.98 KB

Detailed Feature List

This README contains a list of small features/logic flows which play a small, yet significate role in how the agents in the Social Media Agent work.

Each feature is nested under the graph it belongs to. If it does not belong to a graph, it will be under the Shared section.

Main Key

Generate Post

Key

Used URLs

Once a run reaches the humanNode, all of the URLs inside the relevantLinks and links state fields will be stored in the LangGraph store. This is then referenced after the verifyLinksSubGraph executes, to see if any of the URLs (relevantLinks and links) have already been used in previous posts.

If any of the URLs exist in the store, it will route the graph to the END node, and not generate a post.

This is implemented to ensure duplicated content is not generated.

Skipping Saved URLs check

If you are okay with duplicated content being generated, you can set the SKIP_USED_URLS_CHECK environment variable to true, or pass the skipUsedUrlsCheck configurable field (variable) to the graph.

This will prevent the graph from saving any URLs, or reading URLs, ensuring the graph will never prevent a link from being used due to the URLs already being used in previous posts.

import { Client } from "@langchain/langgraph-sdk";
import { SKIP_USED_URLS_CHECK } from "src/agents/generate-post/constants";

const client = new Client({
  apiUrl: process.env.LANGGRAPH_API_URL,
});

const { thread_id } = await client.threads.create();
const res = await client.runs.create(thread_id, "generate_post", {
  input: {
    links: ["https://www.example.com"],
  },
  config: {
    configurable: {
      // Pass this to the graph to skip used URLs check, or set the environment variable
      [SKIP_USED_URLS_CHECK]: true,
    },
  },
});

Skip Content Verification

The generate_post graph by default always attempts to "verify" the content from the link provided is relevant to your business context. If you want to bypass this step, and assume all input links are relevant, you can set the SKIP_CONTENT_RELEVANCY_CHECK environment variable to true, or pass the skipContentRelevancyCheck configurable field (variable) to the graph.

import { Client } from "@langchain/langgraph-sdk";
import { SKIP_CONTENT_RELEVANCY_CHECK } from "src/agents/generate-post/constants";

const client = new Client({
  apiUrl: process.env.LANGGRAPH_API_URL,
});

const { thread_id } = await client.threads.create();
const res = await client.runs.create(thread_id, "generate_post", {
  input: {
    links: ["https://www.example.com"],
  },
  config: {
    configurable: {
      // Pass this to the graph to skip content verification, or set the environment variable
      [SKIP_CONTENT_RELEVANCY_CHECK]: true,
    },
  },
});

Shared

Key

Exclude URLs

Inside each of the verify links subgraphs (verify-general, verify-github, verify-youtube, verify-tweet) there is a check (typically named shouldExclude<type>Content) which checks if the URL which was passed should be excluded, and thus, a post should not be generated.

Each of these util functions can be found inside the should-exclude.ts file.

They will first check if the USE_LANGCHAIN_PROMPTS env variable is set to true. If it is not, then they will return false and the graph will continue as normal. If it is set to true, it will then check the URL against a list/single string to see if it should be excluded. This is because the exclusion logic for these functions is specific to LangChain, and we do not want it running (by default) for non-LangChain users.

If the input matches the exclusion string(s), it will return true to not generate a post for that URL.