Farnsworth Syntek
Seven-layer recursive agent memory with context branching, holographic recall, dream consolidation and on-chain persistence.
Install
The installation steps are in the documentation below. Documentation
Documentation
From the project's README on GitHub, refreshed weekly. Links open on GitHub.
Contents
Farnsworth SYNTEK
7-layer recursive agent memory with context branching, holographic recall, and on-chain persistence.
SYNTEK (Synthetic Yield Network for Token Extraction & Knowledge) gives AI agents persistent, evolving memory that survives across sessions, devices, and time. Memory is encrypted and stored on-chain via DropClaw — files go directly on-chain as Monad calldata.
Quick Start
1. Subscribe
npx farnsworth-syntek setup
Walks you through x402 payment ($100 / 90 days) and securely stores your credentials.
2. Connect to Claude Code
claude mcp add farnsworth-syntek -- npx farnsworth-syntek-mcp
3. Use Memory
Your agent now has 10 MCP tools:
| Tool | Description |
|---|---|
syntek_recall | Search memory across all 7 layers (holographic recall) |
syntek_store | Store knowledge — auto-detects the right layer |
syntek_learn | Process conversation history into memory |
syntek_identity | Get/set core identity (always in context) |
syntek_graph | Query the knowledge graph |
syntek_branch | Create parallel context branches |
syntek_context | Get context snapshot for prompt injection |
syntek_status | Check memory system status |
syntek_session_end | Consolidate + encrypt + upload memory to chain via DropClaw |
syntek_subscribe | Check subscription status |
Architecture
7 Memory Layers:
- L1 Identity — Core traits, always in context
- L2 Active Context — Working memory with branching
- L3 Episodic — Timestamped experiences
- L4 Semantic — Facts and knowledge
- L5 Procedural — How-to patterns
- L6 Graph — Entity relationships with temporal decay
- L7 Synthesis — Cross-domain insights
Key Features:
- Holographic recall (BM25 + vector + graph + recency fusion)
- Context branching (multiverse memory)
- SYNTEK 5-pass knowledge compiler
- Dream consolidation (session-end compression)
- AES-256-GCM encrypted on-chain persistence via DropClaw
- Files go directly on-chain as Monad calldata — Syntek never stores user data
On-Chain Persistence
Syntek uses DropClaw for permanent memory storage on Monad blockchain:
- On session end, memory is consolidated and compressed (dream consolidation)
- Consolidated memory is encrypted client-side (AES-256-GCM) — encryption keys stay with the user
- Encrypted blob is written directly on-chain as Monad calldata via DropClaw
- On next session, memory is loaded from chain and decrypted locally
Syntek subscription includes DropClaw access for Syntek memory uploads — the $30 DropClaw service fee is waived. User still pays Monad gas fees for on-chain storage.
Credentials
Credentials are stored in ~/.farnsworth/vault.enc — encrypted, portable across machines.
npx farnsworth-syntek status # Check subscription
npx farnsworth-syntek revoke # Securely erase credentials
Or use environment variables:
export SYNTEK_API_KEY=farn_...
export SYNTEK_GATEWAY=https://ai.farnsworth.cloud
API Documentation
Full API docs: ai.farnsworth.cloud/memory/skill
Payment
- $100 / 90 days via x402 protocol
- Accepted: SOL, MON (Monad), USDC (Base)
- Includes DropClaw access for Syntek memory uploads (gas only, no $30 service fee)
- Gas fees for on-chain memory sync (per session)
Ecosystem
- Syntek — Agent memory engine ($100/90 days, includes DropClaw for Syntek uploads — gas only)
- DropClaw — Encrypted on-chain storage ($30 + gas, Syntek subscribers pay gas only for Syntek uploads)
- PlanetExpress — Agent marketplace ($30 to list, $15 for Syntek subscribers)
License
Proprietary — Farnsworth Labs. All rights reserved.
Frequently asked questions
What is Farnsworth Syntek?
Seven-layer recursive agent memory with context branching, holographic recall, dream consolidation and on-chain persistence.
How do I install Farnsworth Syntek?
Follow the installation steps in its documentation, shown on this page and in its repository.
Is Farnsworth Syntek open source?
Its code is public on GitHub, but the repository does not declare a standard license.
Who maintains Farnsworth Syntek?
It is published by timowhite88. AI.tn lists it but does not maintain or endorse it.
From the community awesome-mcp-servers lists and the project's GitHub repository. Listing is not an endorsement: review a server's code and permissions before giving it access to your data.