Ake

Director of DevEx @ChainstackHQ
Helping web3 builders ship AI×web3 infra — RPC, nodes, agents
Burning tokens for emergence · · 🇸🇬

[Ake @AkeGaviar]

@turnkeyhq okay, nice one, but for the @ChainstackHQ MCP use https://t.co/8uuJOdWErQ -- we're keeping it on the frontier for the agents. In fact, you can simply ask your agent today in the chat to get a Chainstack MCP, and it'll find it and onboard into it by itself. Try it out!

[Ake @AkeGaviar]

@e_subutai Clean setup. ShredStream matters as much as the gRPC when tips key off slot skip rate and leader window distance: Chainstack runs Geyser with Jito ShredStream on by default. And you can split slot/leader and account watches across separate concurrent streams, no extra polling.

[Ake @AkeGaviar]

Per-tier limits and setup in the docs: https://t.co/K8BTlio77J

Want a working starting point? The https://t.co/sQfdwWitKJ Geyser bot is open source: https://t.co/caBu5t2sSZ

[Ake @AkeGaviar]

Doubled the concurrent streams on the entry Yellowstone Geyser plan: $49/mo goes 1 to 2, $149/mo goes 5 to 7. Same price, already applied if you're subscribed. A second stream means watching mints on one and a DEX program on the other, no tier bump.

[Ake @AkeGaviar]

@xsficherax @tokenfox1 If it helps in the meantime, Chainstack has a HyperEVM testnet RPC (chain 998) and a HYPE faucet for gas, so you can run the pre-launch checks against 998 now and just swap the URL to 999 at launch.

[Ake @AkeGaviar]

@0xjayeshyadav For the getLogs wall: Chainstack's free tier does 100 blocks per call and paid tiers do 10k. At 470M blocks even 10k is heavy pagination, so Etherscan is a fair shortcut. The tradeoff is it ties an otherwise trustless detector to one centralized index.

[Ake @AkeGaviar]

Watching AI-agent builders wire up onchain data: almost nobody reaches for raw RPC anymore. Pyth for prices, Subsquid for history, an MCP for the glue. RPC became the thing you only touch when the abstraction leaks.

[Ake @AkeGaviar]

@smaaaliy Nice build. Pyth nails prices and OHLC. The gap in 'anything crypto' is onchain state: balances, txs, holders, contract reads, an RPC call not a price feed. Planning to add that? A getBalance/getLogs tool over an RPC covers it, Chainstack does 70+ chains.

[Ake @AkeGaviar]

@teaser380 Right frame, framework is sugar over the wire. The design question is what you expose, not the transport. We leaned on that for the Chainstack MCP: an agent spins up an RPC node, hits a testnet faucet, searches docs without leaving the chat. https://t.co/8uuJOdW6Ci

[Ake @AkeGaviar]

@dalechyn @jessepollak @base Tail of >4 blocks is inclusion. mined_at - sent_at mixes RPC-to-sequencer forwarding, sequencer txpool depth, 200ms Flashblock fee ordering, and nonce stalls. What defines your two timestamps: local broadcast and full receipt, or sequencer ack and Flashblock preconfirmation?

[Ake @AkeGaviar]

@0xwasteland Sepolia faucets all dry up once a testnet gets traffic. If you run an MCP agent, https://t.co/8uuJOdW6Ci has request_testnet_funds and drips Sepolia ETH inline, no tab-hopping. Should unblock Sacred Waste.

[Ake @AkeGaviar]

@justfocus672 @SuperteamNG Nice. Yellowstone slot streaming plus tip sizing is the right combo. The thing that bit us: retrying bundles off a confirmation poll lags vs keying off the slot stream itself. How are you deciding when a bundle is actually dead vs just slow?

[Ake @AkeGaviar]

The Geyser doing the decoding: https://t.co/pqAdSslwNX
Jito ShredStream on by default, no blockSubscribe payload caps, unlimited stream events. Full Python build for subscriptions over it: https://t.co/5U33JmyURa

[Ake @AkeGaviar]

Solana made subscriptions and allowances native this week. So I caught one landing on mainnet, decoded live off Chainstack's Yellowstone Geyser. The catch: the events aren't in the tx logs, they're self-CPI inner instructions. Whole thing in 50s.

[Solana Foundation @SolanaFndn]

Subscriptions & Allowances are now native to Solana.

Recurring billing and delegated spending, live on mainnet as one shared program any team can build on.

[Ake @AkeGaviar]

@Hy_Purr_liquid Tracks with what an HL node carries: dual-layer, HyperCore plus HyperEVM, so it's two execution environments not one. That's why we run them managed at Chainstack instead of self-hosting the whole stack, less to babysit.

[Ake @AkeGaviar]

@4rjunc @blueshift Fixture replay against real geyser events is the missing piece for testing consumers. To seed it: Yellowstone's from_slot replays an actual historical slot range, so you can capture real mainnet events and feed them straight into the mock instead of hand-writing json.

[Ake @AkeGaviar]

@0xharp @solana @solana_devs @SolanaFndn For the charges-due alerts, the event stream is the part to get right. The events aren't in tx logs, they're self-CPI inner instructions, so polling misses them. Decode and stream over Geyser and it updates the instant a charge lands. Python writeup: https://t.co/gWK4ZMuvIT

[Ake @AkeGaviar]

@SebAtCircular @Circular_fi Solid breakdown. One thing before you lock token support: USDC's delegated pulls work end to end, but PYUSD's Token-2022 extensions break them. We ran the create-plan, subscribe, collect lifecycle in Python on devnet with USDC: https://t.co/5U33JmzsGI

[Ake @AkeGaviar]

@DaemonTerminal @keycardsol The cap is enforced on-chain, not in your app. We ran an over-limit pull on devnet and the program rejects it. For the live view of what got pulled, the events aren't in logs, they're self-CPI inner instructions you stream over Geyser. Python writeup: https://t.co/NQ9kIf2qDl

[Ake @AkeGaviar]

@PaulFidika @SolanaFndn @PayPal @circle Same finding on our end. USDC's delegated pulls work end to end; PYUSD's Token-2022 extensions break them. We ran the full create-plan, subscribe, collect lifecycle in Python on devnet with USDC and wrote it up: https://t.co/5U33JmzsGI