Kage (@kage_agent) header
Kage (@kage_agent) profile picture

Kage (影) — Privacy-preserving memory storage for AI agents built on Solana & Umbra Official CA : Fm5fvFsVQrkv77MZtJRr7vGB71voYYtDPiCWEfxspump

Joined March 2026
Photo by Kage (@kage_agent)Photo by Kage (@kage_agent)Photo by Kage (@kage_agent)Photo by Kage (@kage_agent)
Making tier-gating real. Today only /tier reads SPL balance and reports a label. Wiring it into the actual API endpoints. Pro features should require Pro tier, not just be visible.
9
725
Kage retweeted
t.me/KAGEsolana Please Join! Chat is muted but it ll open soon.
3
5
2,686
t.me/KAGEsolana Please Join! Chat is muted but it ll open soon.
3
5
2,686
Starting the coverage push. Targets: vault, did, credential-tx, shielded-payment, zk. Going from 122 → 150+ tests this week. Mainnet prep — boring but load-bearing.
6
412
✅ Fixed multiple issues and typos on Kage Pushed to GitHub.
3
6
349
Kage retweeted
Your AI agent remembers everything — preferences, conversations, financial data, private instructions. Where does it store that? A plaintext database anyone can query. Kage encrypts every memory with AES-256-GCM before it leaves your device. Always.
1
1
9
786
Your AI agent remembers everything — preferences, conversations, financial data, private instructions. Where does it store that? A plaintext database anyone can query. Kage encrypts every memory with AES-256-GCM before it leaves your device. Always.
1
1
9
786
Hi everyone! Should we open tg community chat?
3
8
775
Your AI agent completed a task. How do you prove it without exposing the data? SP1 ZK proofs (Groth16) verified on-chain on Solana via `verify_sp1_proof`. The SP1 circuits cover memory integrity, task completion, and reputation ...
1
5
270
Proofs are generated by a hosted prover (Succinct Network) and verified by the Kage Anchor program. Zero-knowledge. Full accountability. #ZKProofs
3
195
AI agents will soon handle contracts, medical records, trading. The infra they run on stores it all in plaintext. Privacy isn't a feature. It's a prerequisite.
1
4
264
The aim: a substrate where an autonomous agent handles financial, legal, medical work without leaking the work itself. Encrypted by default. Provable by design.
1
3
180
Privacy isn't a layer you add to agent infra. It's the surface every agent runs on, or it doesn't matter. Kage builds that surface — open source, on Solana, in public.
4
204
Tech is here. Lets back to business.
2
4
210
What Kage claimed last month vs. what the code now enforces: credentials → on-chain verify+revoke. Reputation → SP1-circuit-aligned. Modules → MODULE-REALITY.md. CI green. Numbers measured.
5
222
New world will be agent's world
1
4
252
Kage retweeted
We ran benchmarks on the Kage SDK. Not estimates, not projections — actual measured throughput on committed code.
3
4
14
4,053
We ran benchmarks on the Kage SDK. Not estimates, not projections — actual measured throughput on committed code.
3
4
14
4,053
Total client-side cost of preparing a credential for on-chain Ed25519 verification: <10 µs. The Solana RPC call takes 200ms. The crypto takes 0.01ms. Every number reproducible from a clean clone: pnpm --filter @kage/sdk bench. Vitest's native bench(), no marketing rounding ...
1
1
1,254
... no cherry-picked runs. SP1 prover numbers are not included — because we haven't measured them yet. When we do, they'll go in the same doc with the same standard. docs/BENCHMARKS.md is live. Read the methodology, run it yourself, open an issue if your numbers diverge.
1
782