Agentic Pipeline

connecting...
Overview
Settings

Suggest a Tweet

Submitted!

Recent Posts

▶ Logs

Pipeline Flow

Bookmarks
keep.md fetch + filter
→
Learnings
extract insights via LLM
→
Patterns (RL)
engagement analysis
▼
LLM Generation
style + patterns + learnings + bookmarks → Claude
▼
Validation
publishability + similarity + topic gate
▼
Post to X
browse automation
→
Collect Metrics
likes, RTs, impressions
→
↻ Next Cycle
feeds back into patterns
Output types: long-form · quote tweet · reply Click any node to jump to its settings
Bookmarks▶
Only process bookmarks from this date onward
Bookmarks fetched per cycle
Chars per bookmark in draft prompts
Bookmarks used for learning extraction
Chars per bookmark in learning prompts
Topics used to filter bookmarks for learning
Saved
Generation▶
X handles to emulate in content style
Attempts to generate a publishable draft
Saved
Style & Knowledge▶
Saved
Saved
Saved
Filtering▶
Tweets must contain one of these to be quoted
0.5
Jaccard word similarity — higher = more strict dedup
Saved
Engagement Thresholds▶
Likes threshold for quoting a tweet
RT threshold (OR with likes)
Likes threshold for replying to mentions
Saved
Volume & Schedule▶
Pages of timeline to scan
Delay between consecutive posts in a cycle
Saved
Prompt Templates▶
These are the exact prompts sent to Claude at each stage. Edit to change tone, structure, or instructions. Placeholders like {style} are filled at runtime with your live data. Clear a field to revert to the default.
Placeholders: {handle} {style} {patterns} {learnings_context} {bookmark_context} {recent_context} {lf_min} {lf_max}
Placeholders: {handle} {style} {target_author} {target_likes} {target_retweets} {target_text} {q_min} {q_max}
Placeholders: {handle} {style} {target_author} {target_followers} {target_likes} {target_text} {r_min} {r_max}
Decides if generated text is a real tweet or LLM reasoning. Placeholder: {text}
Saved