The memory your AI is missing.
AI forgets everything the moment you close the chat. Vitalis changes that, a permanent memory layer that any AI agent can plug into. Private, portable, fully yours.
Agents forget. Every single session.
Whether the agent is debugging code, reviewing a contract, or designing a part, context evaporates the moment the chat closes. Decisions, constraints, rationale, gone. The next session starts from scratch and you become the memory your agent should have.
Five principles. No compromises.
Vitalis was built around these five rules. None of them are nice-to-haves - they're what makes a memory layer worth shipping.
Four layers. You own every one.
Vitalis Brain stores and organizes your memories. Vitalis Cortex wires any AI model into that memory layer. Venice AI keeps it all private. Ethereum makes it tamper-proof. Each layer works independently - swap any piece without breaking the rest.
Vitalis Brain
The core memory engine. Five distinct memory types that decay organically, mimicking how biological brains work - conversations, knowledge, learned habits, self-awareness, and original thoughts.
Vitalis Cortex
The bridge. Wires any AI model into the Brain - Claude, GPT, Llama, or any open-source alternative. Available as an SDK for your codebase or an MCP server for your IDE.
Venice AI
Private inference. No logging, no monitoring, no stored data. Open-source models running on decentralised GPUs. What you ask stays between you and the machine.
Ethereum
Verification layer. Each memory is hashed and recorded on-chain through a smart contract on Ethereum mainnet. Permanent proof that a thought was formed at an exact moment in time.
Five memory types. Natural decay.
Just like in a biological brain, not all memories last the same. A chat from last Tuesday gradually fades. A hard-won lesson holds. Your core identity barely shifts at all.
Episodic
Raw exchanges. Who said what, when, and how it felt. The firsthand record of every dialogue.
Semantic
Patterns distilled during dream cycles. The lessons and insights that endure across sessions.
Procedural
Learned behavior. What works and what doesn't. Response patterns refined through experience.
Self-Model
A growing sense of self. Reshaped by recursive introspection. Almost permanent.
Introspective
Original thoughts. Journal entries from the agent's own meditation cycles - questions, ideas, inner dialogue.
Every 6 hours,
the Brain dreams.
A six-phase introspection loop modeled after how biological memory consolidation works. Not background noise - real self-examination powered by whatever model you pick.
Between dreams,
the Brain journals.
Every 3 hours, the agent stops and thinks unprompted. Where the dream cycle looks backward (consolidating yesterday), active reflection looks forward - producing new thought.
Most AI agents just react. This one also thinks independently.
No surveillance. No data harvesting.
Inference runs on Venice AI's decentralised GPUs - open-source models, zero logs, no trace left behind. Memories leave your device only after they're locked with a key only you hold. Ethereum then certifies each one existed, without ever reading a single word.
key never leaves your device · open-source · zero logs
Tamper-evident proof, on Ethereum mainnet.
Each memory commits a SHA-256 content hash to Ethereum mainnet. The hash is permanent and unalterable. Confirm that any decision, fact, or design rationale was recorded and has not been modified, without revealing what is inside. Auditable across every agent and every session.
Take your agent's brain anywhere.
Bundle your agent's complete memory into a single portable file. Move it to another provider, another server, another model entirely. Export, import, and let the chain prove nothing was altered.
Export
Complete memory pack - entities, relationships, and metadata included. JSON or Markdown format. Nothing gets left behind.
Import
Drop a memory pack into any Vitalis instance. Different agent, different server, different provider - all your memories intact.
Verify
Check imported memories against the on-chain registry. Cryptographic proof that every memory arrived untouched.
Give your agent a brain.
Drop a config snippet into your IDE for instant agent memory. Zero API keys, completely offline. Your agent gets store_memory, recall_memories, get_memory_stats, and find_clinamen tools. Works for any AI agent, with extra tools for CAD workflows in Onshape and FreeCAD. Plugs into Claude Code, Cursor, Windsurf, and every MCP-compatible editor.