Create a detailed prompt for this Create a json n8n to search for latest tech news in…
- User
Create a detailed prompt for this Create a json n8n to search for latest tech news in hackernews , substack and devnews and rank them and store top 5 it in the excel sheet and then write the blog and store it in excel sheet for linkedin post and post in the linkedin 1 nd marked it as posted and create another automation to post the unposted based on rank and the sheet should be based on the topic.to post it in
Assistant
Okay, this is a fantastic and complex n8n automation challenge! We'll break it down into two main n8n workflows, defining their triggers, steps, and required integrations.
The goal is to create a detailed, actionable prompt for an n8n expert (or an AI capable of generating n8n workflows) to build the required JSON workflows.
Detailed n8n Workflow Prompt: Automated Tech News Aggregation, Blogging & LinkedIn Posting
Objective: Create two interconnected n8n workflows to automate the discovery, ranking, content generation, and scheduled posting of top tech news on LinkedIn, with robust Excel sheet management for tracking and historical data.
Workflow 1: Daily Tech News Aggregation, Blog Generation & Primary Posting
This workflow will run daily, identify the top tech news, generate a LinkedIn-optimized blog post, record everything in an Excel sheet, and attempt the first LinkedIn post.
Workflow Name:
TechNews_DailyAggregator_BloggerTrigger:
ScheduleNode- Mode: Every Day
- Time: e.g., 9:00 AM (or preferred daily run time)
Steps & Logic:
-
Data Acquisition - Fetch Latest Tech News:
- Purpose: Gather recent articles from specified tech news sources.
- Node Type:
HTTP Requestfor APIs,RSS Feed Readfor RSS feeds. - Hacker News:
- Endpoint 1:
https://hacker-news.firebaseio.com/v0/topstories.json(Get top story IDs) - Endpoint 2: For each ID from Endpoint 1, fetch details:
https://hacker-news.firebaseio.com/v0/item/{id}.json - Data Points to Extract:
title,url,score(upvotes),time(timestamp),id. - Source Identifier: Add a static field
source: "HackerNews".
- Endpoint 1:
- Substack:
- Approach: List of popular tech-focused Substack RSS feeds. (Assume a predefined list provided by user, e.g., "The Rundown AI", "Dense Discovery", "Notion Everything").
- Node:
RSS Feed Read(multiple instances or loop through a list of URLs). - Data Points to Extract:
title,link(URL),pubDate(published date),contentSnippet(short description). - Source Identifier: Add a static field
source: "Substack".
- Dev.to:
- Endpoint:
https://dev.to/api/articles?state=fresh&per_page=50(orfeedfor highly recent ones,state=freshfor a broader recent view). - Data Points to Extract:
title,url,published_at(published date),positive_reactions_count(likes),description. - Source Identifier: Add a static field
source: "Dev.to".
- Endpoint:
-
Data Normalization & Deduplication:
- Purpose: Consolidate disparate data structures into a common format and remove duplicates.
- Node Type:
CodeorItem Lists(for merging and filtering). - Common Fields:
title(string)url(string, primary key for deduplication)source(string, e.g., "HackerNews", "Substack", "Dev.to")raw_score(number, original score from source)published_timestamp(number, epoch time for consistent date comparison)description(string, can be null)
- Deduplication Logic: Use the
urlfield. If multiple entries have the same URL, keep the one with the highestraw_scoreor the most comprehensive data.
-
Ranking & Filtering:
- Purpose: Assign a normalized score to each article and select the top 5 most relevant tech news items.
- Node Type:
Code(for scoring),Sort(for ranking),Item Lists(for slicing). - Scoring Logic (within a
Codenode):- Recency Score:
- Calculate
time_difference_hours = (current_timestamp - published_timestamp) / 3600. - Normalize recency:
recency_score = max(0, 100 - (time_difference_hours / X))(whereXis a decay factor, e.g., 2 for slower decay, meaning older articles lose score faster). Max score for very recent is 100.
- Calculate
- Source-Specific Score Normalization:
- Hacker News:
normalized_source_score = raw_score * Y(e.g.,Y = 0.5to scale down large HN scores). - Dev.to:
normalized_source_score = positive_reactions_count * Z(e.g.,Z = 1for 1-to-1 conversion). - Substack: Assign a default
normalized_source_score(e.g.,50) or usedescriptionlength/presence as a heuristic for quality, as there's no direct "likes."
- Hacker News:
- Combined Rank Score:
final_rank_score = (normalized_source_score * Weight_A) + (recency_score * Weight_B)(e.g.,Weight_A = 0.7,Weight_B = 0.3). Adjust weights based on desired emphasis.
- Recency Score:
- Filtering: Only consider articles published within the last 24-48 hours.
- Sorting: Sort all articles by
final_rank_scorein descending order. - Top 5 Selection: Take the first 5 items from the sorted list.
-
AI Content Generation (LLM Integration):
- Purpose: Generate a compelling LinkedIn-optimized blog post based on the top 5 articles, and identify the overall topic.
- Node Type:
OpenAI(or equivalent LLM node like Anthropic, Cohere). - Input to LLM: A well-structured JSON array of the top 5 articles, including
title,url,source,description(if available), and theirfinal_rank_score. - LLM Prompt (Example):
{ "model": "gpt-4-turbo-preview", "messages": [ { "role": "system", "content": "You are a professional tech news blogger and social media manager specializing in concise, engaging LinkedIn content. Your task is to analyze the provided tech news articles, identify the overarching theme/topic, and craft a compelling, informative, and professional LinkedIn post. The post should encourage discussion and provide value to a tech-savvy audience. Use clear, concise language suitable for LinkedIn. Include relevant emojis and 3-5 appropriate hashtags. Do NOT exceed 1800 characters for the post content (excluding hashtags). Summarize the key takeaways of the top articles." }, { "role": "user", "content": "Here are the top 5 tech news articles for today:\n\n" + JSON.stringify($json.top_5_articles, null, 2) + "\n\nPlease provide your response in a JSON format with the following keys:\n- `topic`: (A short, descriptive string for the main topic, e.g., 'AI Ethics', 'Cloud Security Updates', 'Web3 Innovations')\n- `blog_title`: (A catchy title for the overall post)\n- `linkedin_post_content`: (The full text of the LinkedIn post, including emojis and a call to action but *excluding* hashtags)\n- `hashtags`: (An array of 3-5 relevant hashtags, e.g., ['#AI', '#TechNews', '#SoftwareDevelopment'])" } ], "response_format": { "type": "json_object" }, "temperature": 0.7 } - Output from LLM:
topic,blog_title,linkedin_post_content,hashtags(as an array).
-
Excel Sheet Integration (Google Sheets / Microsoft Excel Online):
- Purpose: Store the raw news articles, their ranking, the generated blog content, and track posting status.
- Node Type:
Google SheetsorMicrosoft Excel (Online). (Google Sheets is generally easier for N8N). - Authentication: Requires setting up Google Sheets OAuth2 credentials.
- Sheet Naming Convention: The sheet should be based on the
topicgenerated by the LLM. If the topic is new, create a new sheet. If it exists, append to it. This requires aCodenode to check for sheet existence and conditionally create/append.- Action:
Search for sheet by name: Check if a sheet namedtopicexists.If not exists:Create Sheetwithname = topic.Append Rowto the determinedtopicsheet.
- Action:
- Excel Sheet Schema (Columns for each
topicsheet):ID(Unique ID, e.g., n8n execution ID + timestamp, or a UUID)Topic(string, from LLM)Main_Blog_Title(string, from LLM)Blog_Content_Full(string, the full LinkedIn post content)Hashtags(string, comma-separated from array)Article_1_Title(string, title of top article)Article_1_URL(string, URL of top article)Article_1_RankScore(number)Article_2_Title...Article_5_RankScore(repeat for all 5 articles)LinkedIn_Posted(boolean, defaultFALSE)LinkedIn_Post_Date(datetime, null initially)Post_Attempt_Count(number, default0)
-
LinkedIn Posting:
- Purpose: Share the generated blog post to LinkedIn.
- Node Type:
LinkedInnode. - Authentication: Requires setting up LinkedIn OAuth2 credentials with permissions for posting.
- Action:
Create Post. - Content: Combine
linkedin_post_contentandhashtagsfrom the LLM output.Post Content:{{ $json.linkedin_post_content }}\n\n{{ $json.hashtags.join(' ') }}
- On Success:
- Update the corresponding row(s) in the Excel sheet:
LinkedIn_Posted = TRUELinkedIn_Post_Date = current_timestampPost_Attempt_Count = Post_Attempt_Count + 1(from previous value)
- Update the corresponding row(s) in the Excel sheet:
- On Failure:
- Update the corresponding row(s) in the Excel sheet:
LinkedIn_Posted = FALSE(if it was TRUE, set back to FALSE)Post_Attempt_Count = Post_Attempt_Count + 1
- Send a notification (e.g.,
Telegram,Email) to the administrator.
- Update the corresponding row(s) in the Excel sheet:
-
Error Handling & Notifications:
- Node Type:
IFnodes,TelegramorEmailnodes. - Implement branches for failed API calls, LLM errors, or LinkedIn posting failures.
- Send detailed error messages including the failed step and relevant data.
- Node Type:
Workflow 2: Unposted Content Management & Secondary Posting
This workflow will run less frequently, scan the Excel sheet for unposted content, and attempt to post them based on rank.
Workflow Name:
TechNews_UnpostedPublisherTrigger:
ScheduleNode- Mode: Every Day (e.g., 3:00 PM), or Weekly (e.g., Monday 10:00 AM) depending on volume.
Steps & Logic:
-
Get List of Excel Sheets (Topics):
- Purpose: Identify all the topic-based sheets in the Google Sheet.
- Node Type:
Google Sheetsnode ->Get Sheets.
-
Iterate Through Each Topic Sheet:
- Node Type:
Split In BatchesorLoopthrough the list of sheet names.
- Node Type:
-
Read Unposted Data from Current Sheet:
- Purpose: Fetch all entries from the current topic sheet that haven't been posted to LinkedIn.
- Node Type:
Google Sheetsnode ->Read Data. - Options:
- Sheet Name: Use the current sheet name from the loop.
- Filter:
LinkedIn_PostedcolumnIS FALSE. - Max Rows: (Optional, limit to avoid processing too much at once, e.g., 100).
-
Select Best Unposted Item for Posting:
- Purpose: From the unposted entries, pick the one with the highest overall rank that hasn't been attempted too many times.
- Node Type:
CodeorSortandItem Lists(for slicing). - Logic:
- Sort the retrieved unposted items by
Article_1_RankScore(descending). - Filter out items where
Post_Attempt_Countis excessively high (e.g., > 3). - Select the single top item from the filtered and sorted list.
- Extract
Main_Blog_Title,Blog_Content_Full,Hashtags, and theID(for updating Excel).
- Sort the retrieved unposted items by
-
LinkedIn Posting:
- Purpose: Post the selected unposted content.
- Node Type:
LinkedInnode. - Action:
Create Post. - Content: Use the
Blog_Content_FullandHashtagsfrom the selected item. - On Success:
- Update the specific row in the Excel sheet using its
ID(from step 4):LinkedIn_Posted = TRUELinkedIn_Post_Date = current_timestampPost_Attempt_Count = Post_Attempt_Count + 1
- Update the specific row in the Excel sheet using its
- On Failure:
- Update the specific row in the Excel sheet:
Post_Attempt_Count = Post_Attempt_Count + 1
- Send a notification.
- Update the specific row in the Excel sheet:
-
Error Handling & Notifications:
- Similar to Workflow 1, ensure robust error handling and administrator notifications for any posting failures or issues during data retrieval/processing.
Prerequisites & Setup:
- n8n Instance: A running n8n instance (self-hosted or cloud).
- API Keys/Credentials:
- OpenAI API Key (or equivalent LLM provider API key).
- Google Sheets API access enabled and OAuth2 credentials configured in n8n.
- LinkedIn Developer Account with application setup and OAuth2 credentials configured in n8n (requires
w_member_socialpermission for posting). - (Optional) Telegram Bot Token or Email SMTP credentials for notifications.
- Initial Google Sheet: Create a main Google Sheet document. The workflows will create topic-specific sheets within this document.
- Substack RSS Feeds: A list of 5-10 specific tech-focused Substack RSS feed URLs to be hardcoded or managed in n8n credentials.
Key Considerations & Assumptions:
- Scalability: For a very high volume of news or many topics, optimize Google Sheets operations (e.g., batch updates instead of row-by-row).
- LLM Token Usage: Be mindful of the number of items sent to the LLM and the length of the generated content, as this directly impacts cost.
- LinkedIn Posting Limits: Be aware of LinkedIn's API rate limits and posting policies to avoid being flagged as spam.
- Topic Granularity: The LLM's ability to identify consistent and distinct topics will directly impact the organization of your Excel sheets. Test and refine the LLM prompt.
- Ranking Refinement: The scoring logic for ranking is crucial. It might need iterative refinement based on what truly constitutes "top news" for your specific goals.
- News Source Reliability: Assumes the provided news sources are generally reliable and active.
- Error Recovery: The "Post Attempt Count" helps prevent infinite loops on perpetually failing posts but requires manual intervention for investigation.
By providing this detailed prompt, an n8n developer (or capable AI) should be able to construct the two robust workflows described, integrating all the specified functionalities. The output should be the complete n8n workflow JSON for both
TechNews_DailyAggregator_BloggerandTechNews_UnpostedPublisher.- User
add the things like use free option for tools and make it two first to create and store it . Second post based on the status of like posted unposted based on the ranking of the news based on upvotes on the source website and post it in linkedin if it is not posted . Add to above prompt
Assistant
Okay, fantastic additions! Incorporating "free options," prioritizing "upvotes on the source website" for ranking, and solidifying the two-workflow structure makes this even more robust.
Here's the updated, detailed n8n workflow prompt, integrating your latest requirements.
Detailed n8n Workflow Prompt: Automated Tech News Aggregation, Blogging & LinkedIn Posting (Free Tier Focused)
Objective: Create two interconnected n8n workflows to automate the discovery, ranking (primarily by source upvotes/likes), content generation using free LLM options, and scheduled posting of top tech news on LinkedIn, with robust Excel sheet management for tracking and historical data, prioritizing free-tier tools where possible.
Workflow 1: Daily Tech News Aggregation, Blog Generation & Storage
This workflow will run daily, identify the top tech news, generate a LinkedIn-optimized blog post using a free LLM, record everything in an Excel sheet, but will NOT attempt posting in this workflow. Its sole purpose is to create and store the content.
Workflow Name:
TechNews_DailyAggregator_Blogger_StoreTrigger:
ScheduleNode- Mode: Every Day
- Time: e.g., 9:00 AM (or preferred daily run time)
Steps & Logic:
-
Data Acquisition - Fetch Latest Tech News:
- Purpose: Gather recent articles from specified tech news sources.
- Node Type:
HTTP Requestfor APIs,RSS Feed Readfor RSS feeds. - Hacker News:
- Endpoint 1:
https://hacker-news.firebaseio.com/v0/topstories.json(Get top story IDs) - Endpoint 2: For each ID from Endpoint 1, fetch details:
https://hacker-news.firebaseio.com/v0/item/{id}.json - Data Points to Extract:
title,url,score(upvotes),time(timestamp),id. - Source Identifier: Add a static field
source: "HackerNews".
- Endpoint 1:
- Substack:
- Approach: List of popular tech-focused Substack RSS feeds. (Assume a predefined list provided by user, e.g., "The Rundown AI", "Dense Discovery", "Notion Everything").
- Node:
RSS Feed Read(multiple instances or loop through a list of URLs). - Data Points to Extract:
title,link(URL),pubDate(published date),contentSnippet(short description). - Source Identifier: Add a static field
source: "Substack". - Note for Ranking: Substack typically does not provide "upvote" counts. For ranking purposes, these entries will be treated differently (e.g., ranked by recency, or given a default lower "synthetic" score if there's no equivalent metric).
- Dev.to:
- Endpoint:
https://dev.to/api/articles?state=fresh&per_page=50(orfeedfor highly recent ones,state=freshfor a broader recent view). - Data Points to Extract:
title,url,published_at(published date),positive_reactions_count(likes),description. - Source Identifier: Add a static field
source: "Dev.to".
- Endpoint:
-
Data Normalization & Deduplication:
- Purpose: Consolidate disparate data structures into a common format and remove duplicates.
- Node Type:
CodeorItem Lists(for merging and filtering). - Common Fields:
title(string)url(string, primary key for deduplication)source(string, e.g., "HackerNews", "Substack", "Dev.to")raw_upvote_score(number, original score from source, e.g., HNscore, Dev.topositive_reactions_count. For Substack, default to 0 or a low base value.)published_timestamp(number, epoch time for consistent date comparison)description(string, can be null)
- Deduplication Logic: Use the
urlfield. If multiple entries have the same URL, keep the one with the highestraw_upvote_score.
-
Ranking & Filtering - PRIORITIZE SOURCE UPVOTES:
- Purpose: Assign a normalized score to each article and select the top 5 most relevant tech news items.
- Node Type:
Code(for initial score assignment/cleaning),Sort(for ranking),Item Lists(for slicing). - Filtering: Only consider articles published within the last 24-48 hours.
- Sorting Logic:
- Primary Sort: Sort all articles by
raw_upvote_scorein descending order.- Special handling for Substack: Since Substack usually lacks
raw_upvote_score, these articles will naturally fall to the bottom when sorting by this metric unless given a synthetic baseline score. For Substack only, consider sorting bypublished_timestampprimarily, thenraw_upvote_score(if applicable, which it isn't).
- Special handling for Substack: Since Substack usually lacks
- Secondary Sort: If
raw_upvote_scoreis equal (or for Substack, where it's 0), sort bypublished_timestampin descending order (most recent first).
- Primary Sort: Sort all articles by
- Top 5 Selection: Take the first 5 items from the sorted list.
- Important: Attach the
raw_upvote_scoreto each of the top 5 selected articles.
-
AI Content Generation (Free LLM Integration - Ollama Recommended):
- Purpose: Generate a compelling LinkedIn-optimized blog post based on the top 5 articles, and identify the overall topic.
- Node Type:
HTTP Request(to interact with a localOllamaserver) or a dedicatedAInode if n8n has one for custom endpoints. - Prerequisite: You need
Ollamainstalled and running a compatible model (e.g.,llama2,mistral,gemma) on a machine accessible by your n8n instance. - Ollama Endpoint (Example):
http://localhost:11434/api/generate(adjustlocalhostif Ollama is on another machine). - Input to Ollama: A well-structured JSON array of the top 5 articles, including
title,url,source,description(if available), and theirraw_upvote_score. - LLM Prompt (Example, for
llama2ormistralcompatible structure):{ "model": "llama2", // or "mistral", "gemma", etc. - must be available in Ollama "prompt": "You are a professional tech news blogger and social media manager specializing in concise, engaging LinkedIn content. Your task is to analyze the provided tech news articles, identify the overarching theme/topic, and craft a compelling, informative, and professional LinkedIn post. The post should encourage discussion and provide value to a tech-savvy audience. Use clear, concise language suitable for LinkedIn. Include relevant emojis and 3-5 appropriate hashtags. Do NOT exceed 1800 characters for the post content (excluding hashtags). Summarize the key takeaways of the top articles.\n\nHere are the top 5 tech news articles for today:\n\n" + JSON.stringify($json.top_5_articles, null, 2) + "\n\nPlease provide your response in a structured text format for parsing, with clear labels for Topic, Blog Title, LinkedIn Post Content, and Hashtags. Example: 'TOPIC: AI Ethics\nBLOG_TITLE: The Ethical Frontier of AI\nLINKEDIN_POST_CONTENT: ...\nHASHTAGS: #AI #Ethics #TechNews'", "stream": false, "options": { "temperature": 0.7 } } - Output Parsing: Use a subsequent
Codenode orExtract Textnode to parse the Ollama response intotopic,blog_title,linkedin_post_content, andhashtags(as an array).
-
Excel Sheet Integration (Google Sheets - Free Tier Friendly):
- Purpose: Store the raw news articles, their ranking, the generated blog content, and track posting status.
- Node Type:
Google Sheetsnode. - Authentication: Requires setting up Google Sheets OAuth2 credentials.
- Sheet Naming Convention: The sheet should be based on the
topicgenerated by the LLM. If the topic is new, create a new sheet. If it exists, append to it. This requires aCodenode to check for sheet existence and conditionally create/append.- Action:
Search for sheet by name: Check if a sheet namedtopicexists.If not exists:Create Sheetwithname = topic.Append Rowto the determinedtopicsheet.
- Action:
- Excel Sheet Schema (Columns for each
topicsheet):ID(Unique ID, e.g., n8n execution ID + timestamp, or a UUID for easy lookup later)Topic(string, from LLM)Main_Blog_Title(string, from LLM)Blog_Content_Full(string, the full LinkedIn post content)Hashtags(string, comma-separated from array)Article_1_Title(string, title of top article)Article_1_URL(string, URL of top article)Article_1_RawUpvoteScore(number, CRITICAL for later re-ranking)Article_2_Title...Article_5_RawUpvoteScore(repeat for all 5 articles)LinkedIn_Posted(boolean, defaultFALSE)LinkedIn_Post_Date(datetime, null initially)Post_Attempt_Count(number, default0)Created_Date(datetime, timestamp of when this entry was created)
-
Error Handling & Notifications:
- Node Type:
IFnodes,TelegramorEmailnodes (Gmail's free SMTP or Telegram bot). - Implement branches for failed API calls, LLM errors, or Google Sheets issues.
- Send detailed error messages including the failed step and relevant data to an administrator.
- Node Type:
Workflow 2: Unposted Content Management & LinkedIn Posting
This workflow will run periodically, scan the Excel sheet for unposted content, and attempt to post them to LinkedIn based on their original upvote ranking and posting status.
Workflow Name:
TechNews_UnpostedPublisherTrigger:
ScheduleNode- Mode: Every Day (e.g., 3:00 PM), or more frequently if desired, but less often than Workflow 1. This should be after Workflow 1 has had a chance to store new content.
Steps & Logic:
-
Get List of Excel Sheets (Topics):
- Purpose: Identify all the topic-based sheets in the Google Sheet.
- Node Type:
Google Sheetsnode ->Get Sheets.
-
Iterate Through Each Topic Sheet:
- Node Type:
Split In BatchesorLoopthrough the list of sheet names.
- Node Type:
-
Read Unposted Data from Current Sheet:
- Purpose: Fetch all entries from the current topic sheet that haven't been posted to LinkedIn AND haven't exceeded a retry limit.
- Node Type:
Google Sheetsnode ->Read Data. - Options:
- Sheet Name: Use the current sheet name from the loop.
- Filter:
LinkedIn_PostedcolumnIS FALSEANDPost_Attempt_Countcolumn< 3(or desired max attempts). - Max Rows: (Optional, e.g., 50 per sheet to avoid overwhelming the workflow).
-
Select Best Unposted Item for Posting (Based on Upvotes):
- Purpose: From the retrieved unposted entries for the current topic, pick the single best candidate to post.
- Node Type:
Code(for custom sorting logic) orSortandItem Lists(for slicing). - Logic:
- Primary Sort: Sort all filtered unposted items by
Article_1_RawUpvoteScorein descending order. This uses the original upvote count recorded from the source website. - Secondary Sort: If
Article_1_RawUpvoteScoreis equal, sort byCreated_Datein ascending order (older unposted items first, to ensure they eventually get a chance). - Selection: Take the first 1 item from the sorted list.
- Extract
ID,Main_Blog_Title,Blog_Content_Full,Hashtags, andPost_Attempt_Countof this selected item.
- Primary Sort: Sort all filtered unposted items by
-
LinkedIn Posting:
- Purpose: Share the selected unposted blog post to LinkedIn.
- Node Type:
LinkedInnode. - Authentication: Requires setting up LinkedIn OAuth2 credentials with
w_member_socialpermission. - Action:
Create Post. - Content: Combine
Blog_Content_FullandHashtagsfrom the selected item.Post Content:{{ $json.Blog_Content_Full }}\n\n{{ $json.Hashtags }}
- On Success:
- Update the corresponding row in the Excel sheet using its
ID(from step 4):LinkedIn_Posted = TRUELinkedIn_Post_Date = current_timestampPost_Attempt_Count = {{ $json.Post_Attempt_Count + 1 }}
- Update the corresponding row in the Excel sheet using its
- On Failure:
- Update the corresponding row in the Excel sheet:
Post_Attempt_Count = {{ $json.Post_Attempt_Count + 1 }}- (Optional: Add a
Last_Post_Errorcolumn for debugging).
- Send a notification (e.g.,
Telegram,Email) to the administrator.
- Update the corresponding row in the Excel sheet:
-
Error Handling & Notifications:
- Similar to Workflow 1, ensure robust error handling and administrator notifications for any posting failures or issues during data retrieval/processing.
Prerequisites & Setup:
- n8n Instance: A running n8n instance (self-hosted is recommended for truly free solutions, especially with Ollama).
- API Keys/Credentials:
- Ollama Setup:
- Install
Ollamaon a machine accessible by your n8n instance (can be the same server if self-hosting n8n). - Download a suitable model (e.g.,
ollama pull llama2,ollama pull mistral,ollama pull gemma). - Ensure Ollama is running (
ollama serve).
- Install
- Google Sheets API access enabled and OAuth2 credentials configured in n8n.
- LinkedIn Developer Account with application setup and OAuth2 credentials configured in n8n (requires
w_member_socialpermission for posting). - (Optional) Telegram Bot Token or Email SMTP credentials (e.g., Gmail's free SMTP) for notifications.
- Ollama Setup:
- Initial Google Sheet: Create a main Google Sheet document. The workflows will create topic-specific sheets within this document.
- Substack RSS Feeds: A list of 5-10 specific tech-focused Substack RSS feed URLs to be hardcoded or managed in n8n credentials.
Key Considerations & Assumptions (Free-Tier Focused):
- Ollama Performance: Running a local LLM can be resource-intensive. Ensure your server has sufficient RAM and CPU/GPU. The quality of generated content depends heavily on the chosen model and the machine's capabilities.
- Ollama Reliability: A local Ollama server needs to be continuously running and stable.
- Substack Ranking Limitation: Due to the lack of "upvote" metrics, Substack content will primarily be ranked by recency or by manually assigning a synthetic base score in the
raw_upvote_scorefield in theCodenode. - Google Sheets Limits: Google Sheets has generous free limits, but be mindful of very high daily read/write operations if your news volume scales significantly.
- LinkedIn Posting Limits: Be aware of LinkedIn's API rate limits. Posting too frequently or too much identical content might lead to temporary blocks or account restrictions. The two-workflow approach helps manage this by separating content creation from posting attempts.
- Topic Granularity: The LLM's ability to identify consistent and distinct topics will directly impact the organization of your Excel sheets. Test and refine the LLM prompt, especially with a free/local LLM, as their summarization capabilities might vary.
- Error Recovery: The "Post Attempt Count" helps prevent infinite loops on perpetually failing posts but requires manual intervention for investigation if a post consistently fails.
- No Commercial LLM Costs: By focusing on Ollama, API costs for content generation are eliminated.
This detailed prompt should allow for the construction of a robust, primarily free-tier automated system for tech news aggregation and LinkedIn posting.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.