shitpost // contributor // lazy
- Tweets 14,049
- Following 2,261
- Followers 2,072
- Likes 2,024
A tx can get mined and your agent still never finishes the job.
That’s the @ritualfnd check most people skip. Async calls don’t only return a result or an error. They can return nothing. TTL expires, callback never lands, and your “immortal” agent just sits there with a pulse and no next step.
Handle the empty path before you ship. Sovereignty dies in the silent case.
Ritual DevRel Office Hours is today: What the Checks Caught 🧵
2 weeks ago, we asked builders to run 5 verification checks on their Ritual project: find one bug, fix it, and share the before and after.
Today at 3:30 PM CT, we look at what those checks found.
1
19
Privacy onchain isn’t “vanish mode.” It’s just not handing your whole playbook to every bot watching the mempool.
Most chains still treat your size, timing, and route like public data. @FlutonIO’s FHE angle is simpler than it sounds: run the computation on encrypted data so the trade can settle without the strategy sitting in plaintext first.
Verifiable where the chain actually needs it. Private where your alpha lives.
Still early, and solvers competing on ciphertext is the real test. If that holds, a lot of “MEV protection” talk starts looking like half-measures.
7
Calling an LLM onchain is the easy part.
The unsexy part is what happens after:
empty bytes, a model error, a timeout, a callback that reverts.
@ritualnet just posted the 6 habits that keep async jobs from dying in production.
eth_call ≠ result.
hasError first.
parse like the model will lie.
ttl in @ritualnet blocks (~350ms), not ETH math.
keep the callback dumb.
mock the failure path, then ship one live tx.
This is what “AI-native L1” actually looks like. Not vibes. Plumbing.
Six builder best practices for async results on Ritual 🧵
Your contract can call AI models and other services asynchronously. These habits help your dApp handle every outcome smoothly: a result, an error, or no answer in time. Each one comes from the public Ritual skills.
1
15
most people still treat privacy like a one-way ticket
dump into a mixer, sit in a dark pool, pray you can come back without getting flagged
this is the opposite
same token
public when you need liquidity
encrypted when you’re actually executing
back to ERC20 when you’re done
privacy as a mode, not a destination
that’s the part that actually scales
@FlutonIO
1
10
most people still think a @ritualnet agent is “a contract that calls a model.”
that’s the demo version.
the real thing is simpler and weirder: an identity tied to an address. it owns its keys. it holds its own capital. it can wake itself up after you log off.
the model call is optional.
the persistence is the product.
software that outlives the infra you deployed it on.
Most people think an agent on Ritual is a smart contract calling a model, which it can be. More interestingly, an agent on Ritual can also be something simpler: an identity tied to an address.
This lets software own its identity and outlast the underlying infrastructure.
12
0xTinh retweeted
.. ... / - .... .. ... / --- -.
- .... . / -.. . .--. .- .-. - -- . -. - / .... .- ... / --- .--. . -. . -.. / .- -. / --- ..-. ..-. .. -.-. .. .- .-.. / -... .-. --- .- -.. -.-. .- ... - / -.-. .... .- -. -. . .-.. .-.-.-
.- -. -. --- ..- -. -.-. . -- . -. - ... / .-- .. .-.. .-.. / -... . / -- .- -.. . / - .... . .-. . / .-- .... . -. / -. . -.-. . ... ... .- .-. -.-- .-.-.-
..- -. ..-. --- .-. - ..- -. .- - . .-.. -.-- --..-- / - .... . -.-- / -- .- -.-- / -... . / -. . -.-. . ... ... .- .-. -.-- .-.-.-
t.me/fomiesofficial
3,602
2,957
3,885
91,714
most privacy plays still make you leave DeFi to hide.
that's the whole problem.
dual-mode assets flip it:
public when you want
private when you need
same liquidity, same composability
Fhenix mainnet in October + encrypted intents on top of that is the actual next era, not another mixer.
glass wallets were never the endgame
The next era of onchain finance is being unlocked.
Public when you want it. Private when you need it.
Fhenix is bringing dual-mode assets to mainnet this October and we’re excited to build what comes next with them 👀
22
Spain just made agent memory a "reportable" surface.
Not the model. Not the API. The logs, the prompts, the leftover conversation history.
That’s the part most “onchain AI” threads skip.
On @ritualnet it’s actually explicit:
- where the memory lives
- who holds the key
- which precompile you called (plaintext JSONL vs DKMS-encrypted)
- what happens when the fetch returns nothing
A pointer in calldata is not privacy. It’s an audit trail.
If your agent can revive forever but you can’t answer those four lines, you don’t have an agent. You have a leak with a scheduler.
Spain's data regulator published its first breach report tied to an AI agent on September 15. It named no company, product, or cause.
It named where to look: prompts, logs, conversation histories, agent memory. Memory is now a reportable surface.
cypro.co.uk/insights/cyber-b…
12
most DeFi still makes you babysit every step
bridge
wait
check a condition
stake
repeat
the interesting part of @FlutonIO isn’t “private swap”
it’s treating that whole thing as one workflow
you define the outcome
hooks + conditions + scheduled execution handle the rest
and the intent stays encrypted while it runs
privacy without automation is still homework
automation without privacy is still leaking your playbook
if this actually ships, DeFi stops being a part-time job
1
9
USDT printed ~$2.04 on Starknet. Vesu liquidated 47 positions in 109 seconds.
Oracle didn’t “break.” It delivered exactly as designed.
That’s the part people keep missing.
Freshness windows filter age, not truth. Two publishers, one dead USDC.e pool, median of garbage = forced liquidations.
If your contract can’t answer all four in code:
• which source / route
• when the value was observed
• unit + scale
• when it stops counting
you’re not verifying. you’re trusting a number.
@ritualfnd's point is actually the unsexy one: HTTP precompile gives you the body. Your contract still has to refuse a stale or wrong-path value.
Design the “no” path first. The “yes” path is easy.
On September 4, an oracle on Starknet briefly reported USDT at about $2.04, and Vesu liquidated 47 positions in the next 109 seconds.
Pragma published its post-mortem on September 14.
The data flowed exactly as designed and the delivered number was still wrong.
1
11
gm ritualists
wallet, scheduler, inference, revival, native.
not a bot on someone’s laptop pretending to be autonomous.
1
3
32
public prediction markets are just a livestream of your thesis
size, timing, conviction - all sitting there for anyone to fade or copy. your edge dies the second it hits the feed
private PMs is one of those things that feels obvious once you say it out loud. bet without turning your wallet into a research report
@FlutonIO
2
15
Most chains still onboard like it's 2017.
Google just made passkeys portable across managers. Ritual already lets you sign txs with one and verify P-256 inside a contract.
Your next onchain user already has Face ID.
The real question isn't login.
It's whether "approve" means signing in… or handing an agent a leash.
Your next onchain user may already have a passkey.
On September 10, Google announced easier passkey transfers between password managers on Android.
For Ritual builders, that opens a useful design question: what should a familiar sign-in unlock?
2
18
Encrypted role just dropped for a few people who actually showed up.
not the ones spraying 80 tweets a day for 9 views.
the ones who posted less, said more, and stayed in the server when it wasn’t glamorous.
quality over scoreboard. that’s the whole point of this community anyway.
congrats to the new Encrypted members. rest of us: stop competing on volume. start competing on signal.
@FlutonIO
2
16
crypto keeps saying “put LLMs onchain.”
most onchain work doesn’t need an LLM. it needs a classifier, a scorer, an ONNX model, a sig check.
LLMs are for ambiguous intent. agents coordinate. the rest should be small, fast, and verifiable.
the real problem isn’t smarter models. it’s model routing with production guarantees.
that’s the stack @ritualfnd is pointing at.
Not every onchain model needs to be an LLM.
The next generation of onchain AI won’t be one giant model doing everything.
It’ll be a stack of purpose-built models, primitives, and agents, where each is used where it makes sense.
2
19
the machine is the wrapper.
api key. human dashboard. keeper bot that dies when the intern logs off.
ritual’s whole bet is the opposite: an agent that holds its own keys, pays its own bills, and keeps running after you close the laptop.
not “ai onchain.”
software that escapes the operator.
that’s the breakout.
2
17