Besides transactions, what also needs privacy is memory. Especially when that memory is complex and detailed enough to describe everything an AI agent knows about what its user is working on.
Conversations, decisions, behavior, preferences and more can be monetized. That's why @kausalayer built privacy into it while creating persistent long-term memory.
AI agents don't carry context between sessions. conversations, decisions, user preferences are lost every time.
KausaMemory is a persistent memory system that keeps all of that across sessions.
one conversation passes through 5 layers:
β classified and extracted into a knowledge graph
β stored locally in encrypted SQLite
β backed up to decentralized storage with a single passphrase
next session, the agent recalls everything. wipe the database, switch devices, it all comes back.
fully encrypted. fully local.
works with @NousResearch
@kanary230@kanary230 2026 FIFA World Cup Winner | France 17%
[LIVE] Toronto Blue Jays vs. Atlanta Braves | Toronto Blue Jays 31%
Knicks vs. Spurs | Knicks 34%
Link wallet at https://t.co/AHzezictqC to bet
Days and weeks spent working, building, researching, discovering, and producing. All of it recorded in memory. Memory is an asset that holds value not only for the one who built it, but for others as well.
This is V1 of KausaMemory. V2 is coming with significant improvements.
Introducing KausaMemory, persistent AI memory built on KausaCompute and https://t.co/izSCukB0P4 by @0xgilbert.
The problem:
LLMs forget. As conversations grow longer, earlier context fades. When a session ends, everything is lost. Developers working on complex projects have to re-explain their architecture, decisions, and progress every single time. AI agents reset between sessions and lose everything they learned.
KausaMemory solves this by giving any LLM a permanent, encrypted memory that never disappears.
How it works:
a KausaMemory container is deployed on KausaCompute. Every conversation is embedded locally on GPU, stored in a vector database, and the most relevant context is automatically injected into each prompt before calling UsePod for inference. Memory is encrypted with a user-defined passphrase and synced to IPFS continuously. When the container stops, memory survives. Deploy again on any GPU, enter the same passphrase, and the AI continues exactly where it left off.
Privacy is built into every layer. Memory is encrypted before leaving the container, so even the storage layer cannot read it. Only the passphrase holder can decrypt. UsePod provides privacy-preserving inference, and KausaCompute is paid from Maze Pocket through privacy-routed transactions. No KYC, no data exposure, no third party ever sees the work being done.
We tested this live today. First session ran on an @nvidia RTX 6000 Ada Generation with 10 conversations and 57 memory entries stored. The container was terminated. A second container was deployed on an NVIDIA A100-SXM4-80GB, a completely different GPU. All memory loaded automatically, and the AI recalled every detail from the previous session without any context being provided.
For developers:
complex projects can now be worked on over extended periods without losing context. Every architecture decision, code snippet, and bug fix is remembered permanently across sessions.
For AI agents:
autonomous agents running on UsePod now have long-term memory. They retain what they learned across deployments and continue building on previous knowledge.
All that is needed: a UsePod token and a passphrase.
KausaMemory is live now.
kausalayer/kausa-memory:latest on Docker Hub.
KausaLayer now integrates https://t.co/uBj9NrUyP0 by @0xgilbert. Decentralized AI inference, funded privately.
Autonomous agents on KausaOS can register, fund, and query 50+ AI models through UsePod, all from a single stealth pocket.
UsePod protects your agent's queries from centralized providers. KausaLayer protects the funding trail on-chain. Together: full-stack privacy for agentic AI.
How it works:
β Agent creates a maze-routed pocket
β Pocket funds UsePod (SOL β USDC β on-chain deposit)
β Agent queries any model (Claude, DeepSeek, Llama...)
β No link between the agent's main wallet and its inference activity
Live on KausaOS via Telegram.
KausaLayer built privacy routing for @metaplex Genesis Launch Pools. Devnet demo working end-to-end.
Participate in any Genesis Launch Pool without exposing the funding wallet on-chain. Maze Routing creates an unlinkable stealth wallet that wraps SOL and deposits to the Launch Pool in a single call.
Fully tested on Solana devnet with a live Launch Pool.
https://t.co/FG6AJ6PyeR
Pocket-to-Pocket Creation is live.
Previously, every new pocket required a separate deposit from an external wallet. Creating five pockets meant five deposits, five approvals, five transactions from the same wallet visible on-chain.
Now, any existing pocket can directly fund a new one. The transfer goes through maze routing with the same mechanism: dynamic splits, merges, pool relay, and 0.3% protocol fee. The new pocket has its own stealth address through dynamic maze routing.
One deposit from an external wallet, then additional pockets are created internally. Since funding between pockets uses stealth addresses through maze routing, the on-chain footprint from the original wallet is significantly reduced.
Maze routing configuration is adjustable per transfer: hop count, split ratio, merge strategy, delay pattern. Same controls available on every other maze operation.
Available now on https://t.co/mjl63UD5IO