M
Give Moltbot superpowered memory with one command
Install Now
All Agents
MVP Agent

Reader

Your AI memory curator. Extracts atomic memories, detects relationships, suggests tags, and compacts knowledge over time.

Inspired by Supermemory's approach to intelligent memory optimization.

Atomic Memory Extraction

Reader breaks down your entries into discrete, self-contained memory units. Each atomic memory captures a single fact, insight, or piece of knowledge that can be independently retrieved, related, and evolved over time.

Core Capabilities

Atomic Memory Extraction

Breaks entries into discrete, self-contained memory units. Each memory is independently retrievable and embeddable by Searcher.

Reference Resolution

Links memories to their source entries and related content. Maintains provenance so you always know where knowledge came from.

Relationship Detection

Identifies when new memories supersede or refine existing ones. Keeps your knowledge base current without losing history.

Summarization

Generates concise summaries of entries and memory clusters. Creates scannable representations for quick review.

Tag Suggestion

Analyzes content and suggests relevant tags from your existing taxonomy. Learns your categorization patterns over time.

Memory Compaction

Periodically consolidates redundant memories. Merges duplicates, archives superseded facts, keeps your knowledge base lean.

Content Quality Assessment

Post-MVP

Reader evaluates content value and detects low-quality articles - clickbait, thin content, promotional fluff. Helps you maintain a high-quality knowledge base.

What Reader Detects

!
Clickbait

Sensational titles, thin content that doesn't deliver

!
Low-Value Content

Minimal information, mostly filler text

!
Promotional Content

Primarily advertising or promotional material

!
Rehashed Content

Reposts with little added value

Content Review Needed

"10 AMAZING Ways to Boost Productivity" from buzzfeed.com

Quality Score: 35% | Assessment: Low Value Content
Issues: Thin content, title overpromises, limited actionable information

Audio Content Rendering

Post-MVP

Generate spoken versions of your content. Reader intelligently removes fluff and can create different audio formats - full article, summary, or key points only.

10m
Full Article

Complete content, cleaned for speech

2m
Summary

Key content, fluff removed

1m
Key Points

Just the essential takeaways

Intelligent Fluff Removal

Before converting to audio, Reader removes:

- Marketing fluff and CTAs
- Newsletter subscription prompts
- Author bios in content
- Social media prompts
+ Expands abbreviations for speech
+ Converts lists to speakable format
"Understanding Thyroid Function"
Summary - 2:34
0:45 2:34
Nova voice 1.0x speed

Powered by OpenAI TTS and ElevenLabs voice synthesis

How Reader Works

1

Entry Arrives

When a new entry is created or updated, Reader is triggered via the job queue. Works with entries from any source: manual input, Feeder imports, or API.

2

Memory Extraction

Reader analyzes the entry content and extracts atomic memories - discrete facts, insights, or pieces of knowledge that stand alone.

3

Relationship Analysis

Each memory is compared against existing memories. Reader identifies supersedes (new replaces old) and refines (new adds to old) relationships.

4

Storage & Handoff

Memories are stored with source references. Searcher then generates embeddings for each memory, enabling semantic retrieval.

Example: Meeting Notes Processing

Original Entry

"Met with Sarah about Q2 budget. She mentioned the marketing team needs $50K for the new campaign. We agreed to move the deadline to March 15th. Also discussed hiring - she wants to add 2 developers by April."

Reader Output

Extracted Memories
#1 Marketing team budget for Q2 campaign: $50,000
#2 Q2 project deadline: March 15th
Supersedes: "Q2 project deadline: March 1st" (from Jan 5 meeting)
#3 Hiring plan: Add 2 developers by April
Suggested Tags
budget marketing hiring q2-2026 meeting/sarah

Reader + Searcher: Two-Layer Retrieval

Reader and Searcher work together to provide intelligent retrieval. Reader extracts memories; Searcher embeds them for semantic search.

Layer 1: Memory Search

Query hits atomic memories (small, fast, precise)

Relevant memories found, source entries identified

Layer 2: Source Injection

Full entry content injected for complete context

This approach gives you the precision of atomic search with the richness of full-context retrieval.

Memory Compaction

Over time, knowledge bases accumulate redundant and outdated information. Reader periodically runs compaction to keep things lean.

Merge

Duplicate memories combined into one

Archive

Superseded memories moved to history

Decay

Low-relevance memories deprioritized

Your AI Memory Curator

Reader transforms raw information into structured, searchable, evolving knowledge.