Readwise Chat (RAG over Highlights)

What this skill carries — grep references/ for any subject the names below miss: !d=${CLAUDE_SKILL_DIR}; command -v skill-toc >/dev/null 2>&1 && exec skill-toc "$d"; s=$HOME/.claude/skills/plugin-utils/bin/skill-toc; [ -x "$s" ] && exec "$s" "$d"; echo "(skill-toc unavailable: references and scripts are NOT listed here — install the plugin-utils plugin, or start a new session so its bin/ reaches PATH)"

Uses GPT-5.1 to answer questions with RAG over the user's full highlight library via WebSocket streaming.

Prerequisites

Chat requires session cookies (not API token). Check if configured:

grep SESSION_COOKIES ~/.readwise/env

If missing, run:

readwise-custom auth session

This launches Chrome, navigates to readwise.io/chat, and extracts httpOnly session cookies via CDP.

Commands

One-Shot Query

Streams the response to stdout as it arrives:

readwise-custom chat "What are my highlights about fiduciary duty?"
readwise-custom chat "Summarize my notes on proxy advisors"
readwise-custom chat "Compare what I've read about SEC regulation vs FINRA"

With JSON output (includes related highlights):

readwise-custom chat "What do my highlights say about broker-dealer regulation?" --json

Interactive REPL

Multi-turn conversation with context preserved:

readwise-custom chat-interactive
readwise-custom chat-interactive --model 5.1-thinking  # reasoning model

Type exit to quit.

Conversation Management

readwise-custom chat-list                           # List all conversations
readwise-custom chat-get <id>                       # Get conversation with messages
readwise-custom chat-delete <id>                    # Delete conversation
readwise-custom chat-rename <id> "Better Title"     # Rename conversation

Model Options

Model ID	Name	Use Case
5.1	Fast (default)	Quick answers, simple lookups
5.1-thinking	Thinking	Complex synthesis, reasoning

Chat vs Search

Feature	readwise-custom chat	readwise readwise-search-highlights
Returns	Synthesized answer + highlights	Raw highlight matches
Model	GPT-5.1 RAG	Vector similarity
Auth	Session cookies	API token
Speed	~3-10s (streaming)	~2s
Highlights returned	Up to 69	Top-N vector matches
Multi-turn	Yes (interactive)	No

Use chat when: You want a synthesized answer, cross-document analysis, or conversational exploration.

Use search when: You want raw highlight matches, exact quotes, or structured data for further processing.