The BitVM2 paper has been accepted at #USENIX 2026, one of the big four academic security, privacy & blockchain conferences.
The BitVM2 paper will serve as basis for Bitcoin rollup & bridging research for the years to come.
https://t.co/xa3znNrxEO
@coinyak_io Oh, that would be cool!
Yes, you can:
gateway-cli routes # all routes
gateway-cli routes --chains # supported chains
gateway-cli routes --tokens base # tokens on Base
gateway-cli routes --src-chain bitcoin # routes from BTC
We've just released our new gateway-cli.
Basically, all you need to do is:
gateway-cli swap --src BTC --dst USDT:ethereum --amount 150USD
You can integrate this into an agent. Some ideas:
- Move BTC to Ethereum, let the agent farm yield, cash out back to BTC
- Swap BTC to USDT, wait until BTC price drops, swap back to BTC, get more sats
- ...
You can put the keys into an env or use, e.g., 1password to not keep any keys on the machine. There's also a dry-run mode, so you could also leave the BTC wallet as a cold wallet.
If there's demand, we've been considering adding AWS KMS support and other signing options. Reach out if you want to give this a try or have feedback!
No AI agent toolchain shipped a native way to move Bitcoin.
This slot was empty.
We built the BOB Gateway CLI to fill it - and now any agent can swap native BTC.
GLM-5.2 (Max) by @Zai_org ranks #10 on the new Agent Arena leaderboard, closely matching Claude-Opus-4.8 (non-thinking) and is the #1 open model by a wide margin!
In Agent Arena, we measure models on millions of real-world, long-horizon agentic tasks from a global community of users. Models can access web search, filesystem, and terminal tools to complete complex workflows. The leaderboard measures model performance on outcomes relative to the average model using a causal tracing methodology.
Compared to 5.1, GLM-5.2 (Max) climbs from #13 to #10. Its clearest gains are confirmed task success, and user praise vs. complaint. Bash capabilities and tool hallucination remain stable. There is a tradeoff in steerability compared to the previous model (-6.0% vs. +1.2%).
GLM-5.2 remains the same price as GLM-5.1, $1.4/$4.4 per input/output MTokens. 1M context window.
Huge congrats @Zai_org for the incredible release!
See thread for details on how GLM-5.2 (Max) performs across 5 different signals.
GLM-5.2 (Max) by @Zai_org ranks #10 on the new Agent Arena leaderboard, closely matching Claude-Opus-4.8 (non-thinking) and is the #1 open model by a wide margin!
In Agent Arena, we measure models on millions of real-world, long-horizon agentic tasks from a global community of users. Models can access web search, filesystem, and terminal tools to complete complex workflows. The leaderboard measures model performance on outcomes relative to the average model using a causal tracing methodology.
Compared to 5.1, GLM-5.2 (Max) climbs from #13 to #10. Its clearest gains are confirmed task success, and user praise vs. complaint. Bash capabilities and tool hallucination remain stable. There is a tradeoff in steerability compared to the previous model (-6.0% vs. +1.2%).
GLM-5.2 remains the same price as GLM-5.1, $1.4/$4.4 per input/output MTokens. 1M context window.
Huge congrats @Zai_org for the incredible release!
See thread for details on how GLM-5.2 (Max) performs across 5 different signals.
How are people dealing with context switching between running multiple AI conversations/builds/projects in parallel? I noticed my limit is around 3 conversations where I can still hold context in my head and don't really have to wait for the various instances.
Nobody wants to wake up to "sorry, your money is gone, I went 20x leverage on a memecoin."
But that's exactly what can happen when you give claude & co. access to your funds. I've been doing a lot of experimentation in the last few weeks. Digging into self-learning trading strategies using Karpathy's autoresearch for hyperliquid perps, adding BTC functionality to agentkit and looking at openclaw skills.
The problem I see everywhere in these approaches is that they all rely on prompts, agents md, and other unenforceable means to contain agents. Prompts are suggestions, not rules.
Guardrails can't live in the prompt. They have to live in the wallet. The agent holds a key that can only act within the limits you set. You hold a separate key it can't touch.
But to actually get to an enforceable ruleset seems quite complex. I wrote up the full thinking, including why I think finance ends up inside our chat apps rather than as dedicated finance apps.
Curious how others are thinking about constraining money agents.
Oh, that's really cool!
One trick I used to have more free experimentation was to bootstrap an agent with little money but that could do anything. When it made a certain profit, I programmed it to cash out the funds back to BTC via Gateway to a wallet the agent didn't have access to.
Do you have anything in mind with your agent as a next experiment?
I've been quiet on here for a while. With all the AI stuff, I was heads down building and experimenting. Which is a lot of fun!
Particularly, I've been thinking about how finance and money will look like over the next few years. Plus, how can we make this both secure and autonomous? Will share more on this soon!
I'm updating the Golden trio of onchain safety:
1) Own a hardware wallet, with the new clear signing framework, screen output will go from gibberish to human-readable.
2) Import your hw into @ambire, they cooked, I migrated to it and it's now the best wallet for the EVM experience, check their simulation.
3) create a @safe, even if 1/1 at first, use tenderly simulation.
Bonus: generate a hotwallet without any funds, make it a proposer on your safe and give the private key to your agent to generate the transactions and batchs for you, no more clicking buttons, no more clunky UIs, just prompt then verify simulations at each layer and sign at the end.
Voila, this setup makes you safe, 100x your crypto UX and makes unc Kim sad.
I think I haven't had this much fun building stuff. Claude with custom skills is finally allowing me to write a bunch of code quickly again.
AI and BTC coming soon?
What are people using for memory management for their AIs? I've build a little memory agent (called Paul) that I can use as a skill in claude. Under the hood, it uses a vector DB and a SQLite instance to have multiple levels of memory:
- Permanent memory -> always true, highest prio
- Quarterly goals -> changes over time
- Weekly priorities -> what's important right now, very temporary, Paul knows not to give context from weeks that are in the past
Then a few other specific DBs that handle other cases. I basically use a cheap model call to decide from which memories to load and then a mid-tier model to summarize and forward to the right query.
So I'd do something like "Use /brainstorming with /paul as context to ..."
To me, this works really nice compared to the memory.md approach since it works across projects and instances.
But I'm curious, what are others doing?
Time for the last update before the Christmas break and quite excited about having this finally on mainnet.
You can now swap USDT/USDC/ETH with BTC to Ethereum and Base via BOB. We'll be adding more routes as they become available.
The cool bit is what is happening under the hood though: BOB Gateway is the tech that powers these swaps on the BTC route. There are two cross-chain swaps happening:
1. BTC on Bitcoin <> wBTC on BOB
2. wBTC on BOB <> USDT/USDC/ETH on Ethereum/Base
BOB Gateway optimizes the BTC<>wBTC route. By having the Bitcoin light client on BOB, you can run it much more cheaply than on Ethereum mainnet but with very similar security (since BOB has a working proving system!).
BOB Gateway can then plug into any existing bridge for the wBTC <> USDT/USDC/ETH routes. We already have a bunch of LayerZero routes live that support 10 chains (Base, Ethereum, Unichain, ...). But anyone can plug in their own bridge as well. So if you are a bridge provider, you can easily add BTC capabilities by adding Gateway.
Native Bitcoin swaps are now available on BOB Gateway.
The initial release supports USDC, USDT and ETH on Ethereum and Base. More destination chains and assets will be added soon.
Learn more ๐
https://t.co/dAuuGZMxCS
Lock BTC on Bitcoin and have the ability to withdraw these BTC yourself without bridge operators isn't possible with the "classical" wrappers (wBTC, cbBTC, etc) today.
We are changing this with Bitcoin Vaults. But instead of building one version, we are building the OP Stack for Bitcoin Vaults. There are many different decisions that need to be made with Bitcoin Vaults that should be in the hands of everyone who wants to build on the base primitive of being able to unilaterally withdraw.
The idea is to offer a base stack that anyone can expand according to their needs:
- Launch with a multisig to handle disputed withdrawals or go with BitVM right away
- Select custom operators that facilitate the Bitcoin Vault
- Different strategies to execute liquidations like natively on Bitcoin or a via a wrapped asset
And many more! Full technical specification and something to play with will follow soon. DM me if you want early access.
Today, weโre announcing the Native Bitcoin Vaults Stack ๐
Open-source infrastructure that lets you use Bitcoin as DeFi collateral - without giving up custody.
Lock BTC directly on Bitcoin. Borrow against your exact coins. Get them back on repayment.
Powered by BitVM. Launching in early 2026.
Read the full blog to learn more ๐
https://t.co/BcuUz9lUHh