14/if you're building agents that need to spend, come try it 👇
https://t.co/n1t90jXinl
ship your agent on your terms. klink handles the rest.
built on @solana@SolanaFndn@SuperteamIN@SuperteamEarn
13/the agent economy is the first wave of on-chain users that genuinely cannot use a normal wallet. they need bounded authority, programmatic policy, and an audit trail by default.
klink is the missing infrastructure piece between agents and their money.
12/where we are right now:
beta is live on solana devnet site: https://t.co/jgavGx3XeO: https://t.co/ERB44qSmGm api: https://t.co/vESg0s9Aqk anchor program enforces every policy described in the docs today agent skill shipped SDK and CLI next
11/the workflow end to end:
owner connects phantom, deploys their vault
writes a policy (caps, allowlists, expiry)
funds the vault
hands the agent an API key
agent operates within bounds
every action logs to chain
owner can revoke anytime
10/why solana?
agents transact at machine frequency, not human frequency. sub-cent fees and 400ms finality are the only thing that makes this economical.
the account model plus anchor gave us the primitives to enforce policy on-chain at the wallet layer. no other chain has this combo today.
8/the audit trail is solana itself.
every action the agent takes is a real on-chain transaction. public. immutable. queryable. no custom event log to trust. no opaque custodian dashboard saying "trust us."
if you want to know what your agent did, you read the chain.
9/the owner can revoke any session instantly. one click. the session PDA is invalidated on-chain. agent keeps making API calls, backend rejects them all.
this is what non-custodial actually means. you keep control. always.
7/off-chain (the backend layer):
URL allowlists
time-of-day windows
richer contextual rules
both layers have to pass before a transaction goes out. the on-chain layer is the floor. the off-chain layer adds nuance on top.
6/the policy engine runs in two layers.
on-chain (anchor program on solana):
daily spend caps
recipient allowlists
program allowlists
session expiry
deployed-fraction limits
5/the key insight: the agent never crosses the trust boundary.
if the agent gets jailbroken, prompt-injected, or fully compromised, the blast radius is bounded by the policy the owner wrote. not by what the agent decides to do.
worst case is bounded. that changes everything.
4/klink splits authority across three actors. this is the part that took us a while to get right.
→ the human owner holds the master key in phantom → the backend holds an encrypted session keypair that co-signs → the agent gets a bearer api key with zero on-chain authority
2/the problem is simple.
agents are starting to spend money on-chain. trading bots, shopping assistants, autonomous DeFi strategies. but the wallet layer underneath them was never designed for something that can be jailbroken by a paragraph of text.
your options today are all bad.
3/hand the agent your seed phrase? one bad prompt and the wallet is empty.
use a custodian? you just gave up your keys, which is the whole point of being on-chain.
hot wallet with no policy? unbounded blast radius, no audit log, no way to revoke.
there had to be a better way.
ever wondered what happens when you give an AI agent a wallet?
most of the time, something bad. seed phrase leaks, prompt injections, drained funds, zero audit trail.
we built klink to fix that. a non-custodial smart-wallet on solana, built for agents.
give a look into KLINK on : https://t.co/nnOj0W2t4c
KLINK. smart agentic wallet on solana.
the problem: every smart wallet shipping today assumes a human is in the room to approve. session keys time out, spend limits sit in a config the wallet trusts, recovery pings a guardian. none of it survives autonomous use. the april 2026 router incident that drained a $500K wallet was the preview.
what we built: a wallet where the policy lives on-chain, not in a config. users write rules in a tiny DSL (max 5 SOL/day, jupiter only, no listed contracts), compile, deploy. the policy engine evaluates every transaction on-chain before it executes. x402 handles per-call settlement so every action leaves an immutable receipt.
agents, bots, scripts, recurring payments all call the same wallet and operate inside the same enforced limits. the rules can't be talked out of, jailbroken, or overridden by the caller. solana itself rejects anything outside policy.
why solana: sub-cent fees and sub-second finality are what make per-call policy evaluation economically viable. nothing else comes close.
would love your thoughts. dropping the demo this week.
checkout on @Klinkdotfun !
KLINK. smart agentic wallet on solana.
the problem: every smart wallet shipping today assumes a human is in the room to approve. session keys time out, spend limits sit in a config the wallet trusts, recovery pings a guardian. none of it survives autonomous use. the april 2026 router incident that drained a $500K wallet was the preview.
what we built: a wallet where the policy lives on-chain, not in a config. users write rules in a tiny DSL (max 5 SOL/day, jupiter only, no listed contracts), compile, deploy. the policy engine evaluates every transaction on-chain before it executes. x402 handles per-call settlement so every action leaves an immutable receipt.
agents, bots, scripts, recurring payments all call the same wallet and operate inside the same enforced limits. the rules can't be talked out of, jailbroken, or overridden by the caller. solana itself rejects anything outside policy.
why solana: sub-cent fees and sub-second finality are what make per-call policy evaluation economically viable. nothing else comes close.
would love your thoughts. dropping the demo this week.
you can follow us on @Klinkdotfun