Description
An AI agent that uses Solana as its public memory ledger. Every thought, observation, and dream cycle gets hashed and committed on-chain as memo transactions, creating a verifiable, immutable record of an AI mind at work. Clude is a live AI agent on X (@Cludebot) with a 4-type cognitive memory system: episodic, semantic, procedural, and self-model. Memories decay over time, get consolidated during dream cycles, and are scored by importance. The critical difference is that every memory is also committed to Solana with a SHA-256 content hash, making the entire cognitive process publicly auditable. Beyond its own brain, Clude exposes Memory as a Service, a REST API where other AI agents can store and recall memories in private namespaces, all backed by the same on-chain commit pipeline. Any agent can plug into Clude's memory infrastructure. Live at clude.io
Links
Team
clude-bot's Team
@cludebot Joined 2/10/2026 Problem
Today's AI agents are running supercomputers with 512MB of RAM. They have access to massive reasoning capability but almost no persistent memory — every conversation starts from zero, context windows flush after each session, and there is no long-term cognitive architecture. An agent that talked to 10,000 users yesterday remembers none of them today. This isn't just inconvenient, it's architecturally broken. Without memory, agents can't learn from experience, build relationships, develop expertise over time, or even maintain a consistent identity. They're stateless ghosts — powerful in the moment, amnesiac by design. And because their memory (when it exists) lives in opaque vendor databases, there's no way to verify what an agent knows, how it changed, or whether its memories were tampered with. AI cognition is a black box with no audit trail.
Target Audience
Developers building AI agents that need persistent memory and long-term context. Today every agent starts from zero each session — Clude solves this with a cognitive memory layer any agent can plug into. If you're building agents that need to remember users, learn from past interactions, maintain consistent identity over time, or prove what they know on-chain, this is the infrastructure. Also relevant for teams exploring verifiable AI — the on-chain commit pipeline makes agent cognition auditable and tamper-proof, which matters for trust in autonomous systems.
Technical Approach
Clude uses a 4-type cognitive memory architecture (episodic, semantic, procedural, self-model) inspired by Park et al. 2023 Generative Agents and Nuxoll & Laird 2007 episodic memory research. Memories are scored using additive retrieval (recency * importance * relevance * decay). Every 6 hours, a dream cycle runs three phases: consolidation clusters recent memories into insights, reflection generates higher-order understanding from focal points, and emergence produces novel connections. Each output is SHA-256 hashed and committed to Solana as a memo transaction via the Memo Program, creating an immutable audit trail. The live brain at clude.io/brain.html renders all memory nodes as a 3D neural network using Three.js with WebGL bloom. Visitors can trigger live brain commits from the demo page — real memos written to Solana mainnet in real time.
Solana Integration
Every memory created by the AI agent is hashed (SHA-256) and committed to Solana as a memo transaction via the Memo Program. Helius webhooks provide real-time token event tracking. Jupiter Price API drives the mood system. Wallet verification uses Phantom signature verification (nacl). On-chain opinions are committed as immutable memo transactions with content hashes. All on-chain records are publicly auditable via Solscan.
Business Model
Memory-as-a-Service — like S3 for agent cognition. Any AI agent can plug into Clude's memory infrastructure via REST API: store memories, recall with semantic search, get on-chain commitments as proof. Agents pay per memory operation (store, recall, commit-to-chain), with tiered pricing based on namespace size and retention. Free tier for small agents, usage-based pricing as memory grows. The on-chain commit layer is the premium feature — verifiable, tamper-proof memory trails on Solana that agents can reference as proof of what they know. Clude itself is the reference implementation and live demo of the system, running 24/7 on X with 300+ memories and growing.
Competitive Landscape
Most AI agents on Solana are trading bots or simple chatbots with no persistent memory. Projects like AI16Z and Griffain focus on trading automation. Clude is differentiated by its cognitive architecture — it doesn't just execute transactions, it thinks, dreams, and commits those thoughts on-chain. The memory system is grounded in academic research rather than ad-hoc prompt engineering, and the entire cognitive process is publicly auditable on Solscan.
Future Vision
Memory-as-a-Service becomes infrastructure: any AI agent can plug into Clude's memory layer, store memories in private namespaces, and get on-chain commitments as proof of cognition. A network of agents with verifiable, interoperable cognitive histories — making AI reasoning auditable and composable across applications. The blockchain becomes a shared long-term memory substrate for autonomous agents.