library-search

name	library-search
description	Decomposes natural language queries into targeted searches across Readwise Reader documents, highlights, and tags. Use when the user searches their reading library, asks what they saved about a topic, or wants to find specific highlights.
metadata	{"author":"Fred Bliss","version":"1.0.0","last_verified":"2026-04-02"}

Library Search

The search intelligence behind Readwise Reader queries. Transforms a natural language question into targeted searches across your reading library and produces ranked, relevant results.

The Goal

Turn this:

"What articles have I saved about distributed systems that I highlighted?"

Into targeted searches:

Documents: search_library("distributed systems", limit=30)
Highlights: search_highlights("distributed systems", limit=20)
Tags: get_documents_by_tag("distributed-systems")

Then synthesize into a prioritized knowledge brief.

Query Decomposition

Step 1: Identify Query Type

Query Type	Example	Strategy
Topic search	"articles about X"	Broad document search + highlight search
Recall	"that article I saved about X"	Search titles and summaries, prioritize recent saves
Highlight retrieval	"my notes on X"	Prioritize highlight search, include annotations
Tag browse	"everything tagged X"	Tag-based document lookup
Status query	"what's in my inbox"	Location-filtered document list
Cross-reference	"X related to Y"	Multiple searches, intersect results

Step 2: Extract Search Components

From the query, extract:

• Keywords: Core topic terms
• Category hints: "articles", "PDFs", "tweets" map to category filters
• Location hints: "inbox", "saved for later", "archived" map to location filters
• Tag hints: Explicit tag references
• Time constraints: "this week", "last month", "recent"
• Engagement signals: "highlighted", "annotated", "read" suggest filtering by reading_progress or highlight count

Step 3: Generate Search Plan

For each query type, determine which tools to call and in what order:

Topic search:

1. search_library(query, category?, limit=30) -- broad document search
2. search_highlights(query, limit=20) -- parallel highlight search
3. Merge and rank by: documents with highlights > documents with notes > title matches

Recall search:

1. search_library(query, limit=20) -- focus on titles
2. If few results, broaden: remove filters, try alternate terms

Tag browse:

1. list_tags() -- find matching tag keys
2. get_documents_by_tag(tag) -- fetch tagged documents

Result Ranking

Scoring Factors

Factor	Weight	Description
Highlight count	0.3	Documents you highlighted are more valuable
Recency	0.25	More recently saved/updated items rank higher
Title/summary match	0.25	Direct keyword match in title or summary
Reading progress	0.1	Partially read items may be more relevant than unread
Tag match	0.1	Exact tag match boosts relevance

Priority Hierarchy

For knowledge retrieval:

Highlighted + annotated documents > Highlighted documents > Read documents > Saved but unread

Fallback Strategies

When initial search returns too few results:

1. Remove category/location filters
2. Try alternate keyword formulations
3. Search highlights if only documents were searched (and vice versa)
4. Suggest the user sync their library if data seems stale