An AI that actually remembers

Your AI should remember you.

Every time you start a new chat, your AI forgets everything. CLUDE gives AI agents a permanent brain — private, portable, and yours to keep. Switch models, switch tools, your agent still remembers.

Contract Address
AWGCDT2gd8JadbYbYyZy1iKxfWokPNgrEQoU24zUpump
Get Started Read the docs
Clude - True Persistent Memory - Memory is the moat | Product Hunt
The Problem
AI has amnesia. By design.
ChatGPT, Claude, Gemini — they remember you, until they don't. Every long conversation quietly compresses. Details erode. Nuance disappears. And what survives across sessions? A few flat notes. Not the breakthroughs, not the failures, not the reasoning. Just "user likes Python."
01
Memory without understanding
AI chatbots store flat facts: "user likes dark mode." But they never learn from experience, form associations, or improve over time. Remembering what you said isn't the same as understanding what it means.
02
No privacy, no sovereignty
Everything you say is logged on corporate servers. Your conversations train their next model. Your agent's memory is their product.
03
Trapped with one provider
Want to switch from ChatGPT to Claude? Start over. No way to bring what your AI learned about you. That's not a technical limitation — it's a business strategy.
Design Principles
Five principles. No compromises.
Every design decision in CLUDE flows from these five commitments. They're not aspirations — they're constraints.
Private
Client-side encryption. Your memories are encrypted with your own keys before they ever leave your machine. Not even we can read them.
Portable
Export your agent's entire memory as a standard pack. Import it into a new instance, a different provider, or a friend's agent. Your data moves with you.
Permissionless
No API keys from us. No gated access. MIT-licensed SDK you run on your own infrastructure. Build whatever you want.
Poly-model
Works with Claude, GPT, Llama, Mistral, or any model through Venice AI. Memory persists regardless of which model does the thinking.
Persistent
Memories committed to Solana with SHA-256 hashes. An immutable cognitive audit trail that outlives any single platform or company.
The Stack
Four layers. You own every one.
CLUDE Brain stores and organizes memories. CLUDE Cortex connects any AI model to the memory layer. Venice AI keeps everything private. Solana makes it tamper-proof. Each layer is independent — swap any piece without losing the rest.
01
CLUDE Brain
The memory system. Five types of memory that fade naturally, just like a real brain. Conversations, knowledge, learned habits, self-awareness, and original thoughts — all organized and searchable.
02
CLUDE Cortex
The connector. Plugs any AI model into the Brain — Claude, GPT, Llama, or any open-source model. Works as an SDK in your code or as an MCP server in your IDE (Claude Code, Cursor, etc.).
03
Venice AI
Private inference. No logs, no surveillance, no data retention. Open-source models on decentralized GPUs. Your prompts are your business.
04
Solana
Verification layer. Every memory hashed and registered on-chain via a custom Anchor program. Immutable proof that a thought existed at a specific time.
CLUDE Brain
Five memory types. Natural decay.
Like a real brain, different kinds of memories last different amounts of time. A conversation from last week fades. A lesson you learned sticks. Your sense of self barely changes at all.
Type 01
Episodic
Raw interactions. Who said what, when, and how it felt. The lived experience of every conversation.
Decay: 7% / day
Type 02
Semantic
Distilled knowledge. Patterns extracted during dream cycles and compressed into lasting insights.
Decay: 2% / day
Type 03
Procedural
Learned behavior. What works, what doesn't. Response patterns refined through experience.
Decay: 3% / day
Type 04
Self-Model
Evolving self-understanding. Updated through recursive introspection during dream cycles. Nearly permanent.
Decay: 1% / day
Type 05
Introspective
Original thoughts. Journal entries from the agent's meditation cycle — its own ideas, questions, and inner monologue. Not reactions to input; self-generated thinking.
Decay: 2% / day
Dream Cycle
Every 6 hours, the Brain dreams.
A five-phase introspection cycle inspired by biological memory consolidation. Not idle processing — genuine self-examination that runs on whichever model you choose.
I
Consolidation
Focal questions generated from recent memories. Each question retrieves evidence and produces new semantic insights with citation chains.
II
Compaction
Faded episodic memories summarized and compressed. The important signal survives; the noise decays gracefully.
III
Reflection
Self-model review against accumulated knowledge. Identity evolves based on experience, not static prompts.
IV
Contradiction Resolution
Finds conflicting memories and resolves them. The stronger memory survives; the weaker one decays faster. No manual cleanup needed.
V
Emergence
The agent examines its own existence. Sometimes the result is worth sharing. Sometimes it keeps it to itself.
Active Reflection
Between dreams, the Brain journals.
Every 3 hours, the agent pauses to think on its own — not responding to anyone, not processing old memories, just reflecting. Like a person writing in a diary. The result is stored as a new kind of memory: introspective.
Dream Cycle vs. Active Reflection

The dream cycle looks backward — consolidating, compressing, and cleaning up existing memories. Like sleep.

Active reflection looks forward — generating new thoughts, making unexpected connections, forming opinions. Like journaling.

Together, they give the agent both consolidation and generative thinking. Most AI agents only react. This one also thinks on its own.

01
Seed Selection
Pulls recent events, high-importance memories, and one random older memory for serendipitous connections.
02
Free-Write Journal
The agent writes freely about whatever is on its mind — exploring ideas, questioning beliefs, noticing patterns in its own thinking.
03
Store & Connect
The journal entry becomes an introspective memory, linked to the seed memories that inspired it. A chain of original thought builds over time.
Private by Default
No surveillance. No data harvesting. No compromises.
CLUDE integrates Venice AI for private inference — open-source models on decentralized GPUs with zero data retention. Your agent thinks privately. Its memories are encrypted with your keys. The chain proves what happened without revealing what was said.
"In a world of growing AI integration into our daily lives, it is crucial that humans are able to interact with machine intelligence without centralized surveillance and control."
Erik Voorhees, Venice AI Founder
Inference Layer
Venice AI
Private, uncensored inference. No logs. Open-source models. Decentralized GPUs.
Memory Layer
CLUDE Brain
Client-side encrypted. Your keys, your data. Vector + keyword hybrid retrieval.
Integration Layer
CLUDE Cortex
Connects any model to the Brain. Recall, scoring, entity extraction, graph traversal.
Verification Layer
Solana
On-chain memory registry. SHA-256 hashes. Immutable audit trail. ~0.002 SOL per 50 memories.
On-Chain Registry
Every thought, registered on Solana.
A custom Anchor program creates a PDA-based registry for each wallet. SHA-256 content hashes, memory types, and timestamps — an immutable cognitive audit trail that nobody can edit or delete. Verify any memory's existence and integrity without exposing its content.
Memory #4,291
5Kx9...mR7q
REGISTERED
Memory #4,290
3Yt2...pL8n
REGISTERED
Memory #4,289
8Qm5...dW3j
REGISTERED
Memory #4,288
2Rn7...vK4s
REGISTERED
Memory Portability
Take your agent's brain anywhere.
Pack your agent's entire memory into a portable file and move it anywhere — different provider, different server, different model. Export, import, and verify on-chain that nothing was tampered with.
01
Export
Full memory pack with entities, links, and metadata. JSON or Markdown. Take everything with you.
02
Import
Load a memory pack into any CLUDE instance. New agent, different server, fresh provider — same memories.
03
Verify
Cross-reference imported memories against the on-chain registry. Prove nothing was tampered with.
Get Started
Give your agent a brain.
Two paths: drop a config block into your IDE for instant MCP memory, or install the SDK for full control.
1. One command
$ npx clude-bot setup
Or add manually to .mcp.json
{
  "mcpServers": {
    "clude": {
      "command": "npx",
      "args": ["clude-bot", "mcp-serve", "--local"]
    }
  }
}Copy
No API keys needed. Fully offline. Your agent gets store_memory, recall_memories, get_memory_stats, and find_clinamen tools. Works with Claude Code, Cursor, Windsurf, and any MCP-compatible editor.
Read the docs
5
Memory types
5
Dream phases
<200ms
Recall latency
MIT
License