v0.4.0 · New: FalkorDB + GraphRAG

The AI Memory Layer
for Coding Assistants

Replace flat markdown memory files with vector-database-backed semantic search. SkillMind listens, learns, and surfaces only the context you need — automatically.

$ pip install skillmind[pinecone]
6
Vector Backends
27
MCP Tools
<10ms
Semantic Search
0
Manual Work
How It Works

From Raw Input to AI Memory Layer

SkillMind sits between your data sources and your AI assistant, turning raw observations into structured, searchable knowledge.

Input Sources

  • Git commits and branch merges
  • File changes (CLAUDE.md, configs, code)
  • Claude Code conversations
  • YouTube videos and channels
  • Screen recordings and screenshots
  • Manual input via CLI or MCP

What SkillMind Does

  • Auto-classifies into 5 memory types
  • Sanitizes API keys and PII before storing
  • Deduplicates with 90% similarity matching
  • Embeds content for semantic vector search
  • Surfaces only relevant context per conversation
  • Auto-expires stale project memories after 90 days
Why SkillMind?

Flat Files vs. Vector Memory

Your AI assistant deserves better than loading 15+ markdown files into every conversation.

✘ The Problem

  • 15+ flat markdown memory files growing endlessly
  • Everything loaded into context — 90% is irrelevant
  • No deduplication — same knowledge saved 3 times
  • Manual save/forget — AI only learns when told
  • No protection — API keys and PII stored in plaintext
  • No expiration — stale project memories persist forever

✔ SkillMind AI Memory Layer

  • One vector DB with semantic search across all memories
  • Only relevant memories surfaced per conversation
  • Auto-deduplication with 90% similarity threshold
  • Listens to git, files, and conversations automatically
  • Sanitizer redacts API keys, emails, PII before storing
  • Project memories auto-expire after 90 days
Features

Everything Your AI Memory Layer Needs

From semantic search to video learning — one tool, all capabilities.

Core

Semantic Search

Vector-based recall finds relevant memories by meaning, not keywords. Ask "How should I format PDFs?" and get your stored preferences instantly.

Auto

Auto-Classification

The Trainer automatically classifies memories into types (user, feedback, project, reference, skill) based on content analysis.

Security

Built-in Sanitizer

API keys, emails, phone numbers, IBANs, and PII are automatically redacted before storing. Configurable allowlists and custom patterns.

Learning

YouTube Learning

Extract knowledge from YouTube videos and channels. Transcripts are auto-chunked and stored as structured memories.

Learning

Screen Recording

Record your screen, take screenshots, and learn from video files. Extract knowledge from tutorials and demos.

MCP

27 MCP Tools

Claude Code integration via Model Context Protocol. Remember, recall, forget, consolidate — all as native tools.

New

GraphRAG Retrieval

Opt-in graph traversal on the FalkorDB backend. Recall walks shared topics, tags and links between memories (seed, expand, re-rank) to surface related context plain vector search would miss.

Smart

Deduplication

90% similarity threshold prevents storing the same knowledge twice. Consolidation merges fragmented memories automatically.

Import

One-Click Import

Import all existing Claude Code markdown memories in one command. Preserves types, topics, and tags.

Context

Dynamic Context

ContextGenerator builds focused, relevant context per conversation. Only loads what matters for the current file and topic.

Backends

6 Vector Store Backends

Same interface, swap anytime. From local-only to cloud-synced — pick what fits.

Backend Best For Install Requires
ChromaDBSolo devs, local, defaultpip install skillmind[chroma]Nothing
PineconeMulti-device cloud syncpip install skillmind[pinecone]Free API key
SupabaseSQL + vectors, team sharingpip install skillmind[supabase]Free account
QdrantSelf-hosted, advanced filteringpip install skillmind[qdrant]Qdrant server
FAISSOffline, air-gapped, fastestpip install skillmind[faiss]Nothing
FalkorDB NewGraph + vectors, GraphRAG retrievalpip install skillmind[falkordb]Redis/FalkorDB server

New in v0.4.0: the FalkorDB backend stores vectors and a property graph in one engine and adds opt-in GraphRAG retrieval that walks the connections between memories, about 6.7x faster recall on an EU host.

Read the deep dive →
MCP Integration

27 MCP Tools for Claude Code

Every tool available as a native MCP tool inside Claude Code. The AI memory layer integrates seamlessly.

remember Store memory (auto-classified, sanitized, deduped)
recall Semantic search across all memories
forget Delete a specific memory
update_memory Edit existing memory content
context Generate focused context for current file
consolidate Merge duplicates, expire stale, cleanup
memory_stats Counts by type and backend info
list_memories Filter and browse by type or topic
import_markdown Bulk import from Claude Code memory files
learn_youtube Extract knowledge from YouTube video
learn_channel Learn from a YouTube channels latest videos
learn_video Learn from local video or screen recording
record_screen Record screen to MP4
screenshot Capture and save screenshot
export_obsidian Export memories to an Obsidian vault with wikilinks
sync_obsidian Two-way sync between memories and Obsidian
review_pending Review, approve or reject pending memories
add_pattern Define custom auto-capture regex patterns
Get Started

Up and Running in 3 Steps

From install to semantic memory search in under 5 minutes.

1

Install

pip install skillmind[pinecone]
2

Initialize

skillmind init --backend pinecone
3

Import Memories

skillmind import ~/.claude/projects/*/memory/

Full Installation Guide →

Built by

From a Real-World Need

SkillMind was built by Antonio Blago, an SEO freelancer and developer who needed a better way to manage knowledge across dozens of client projects in Claude Code. After dealing with 15+ growing markdown memory files that loaded irrelevant context into every conversation, he built SkillMind to solve the problem with vector-database-backed semantic search.

The AI memory layer is now used in production across SEO consulting, web development, and content automation workflows — processing 40+ memories with sub-10ms recall times via Pinecone. Every feature was born from a real workflow bottleneck, not a theoretical use case.

FAQ

Frequently Asked Questions

What is SkillMind? +

SkillMind is an open-source AI memory layer that replaces flat markdown memory files with vector-database-backed semantic search. It works as an MCP server inside Claude Code with 27 tools for remembering, recalling, and managing knowledge automatically.

Which vector databases does SkillMind support? +

SkillMind supports 6 vector database backends: ChromaDB (local, free), Pinecone (cloud-synced), Supabase (SQL + vectors), Qdrant (self-hosted or cloud), FAISS (offline, fastest), and FalkorDB (graph-native, with opt-in GraphRAG retrieval). All backends use the same interface — you can switch anytime.

How does SkillMind integrate with Claude Code? +

SkillMind runs as an MCP (Model Context Protocol) server. Add it to your Claude Code settings.json and you get 27 native tools: remember, recall, forget, context, learn_youtube, record_screen, and more. Install with: pip install skillmind[pinecone,mcp]

Can SkillMind learn from YouTube videos? +

Yes. SkillMind extracts transcripts from YouTube videos, structures the key knowledge, and stores it as searchable memories. It supports single videos, entire channels, and playlists. Install the YouTube extra: pip install skillmind[youtube]

What is GraphRAG retrieval in SkillMind? +

GraphRAG is an opt-in retrieval mode on the FalkorDB backend (v0.4.0). Instead of a flat vector lookup, recall seeds with a vector search, expands across shared topics, tags and links between memories, then re-ranks the candidates. It surfaces related context that pure vector search would miss, and falls back to plain vector search when the graph cannot help.

Is SkillMind free? +

Yes, SkillMind is 100% free and open source under the MIT license. The core tool and local backends (ChromaDB, FAISS) require no API keys. Cloud backends like Pinecone have their own free tiers.

Ready to upgrade your AI memory?

Open source, MIT licensed, production-ready.