Talus Protocol v2.1 Release: Bigger Data, Stronger Foundation

When we launched Talus Protocol v2.0, in addition to making it so all versions were backwards compatible, we gave AI agents rules they can’t break: scoped permissions, locked budgets, and a record of every action anyone can check.

Talus Protocol v2.1 Release: Bigger Data, Stronger Foundation

When we launched Talus Protocol v2.0, in addition to making it so all versions were backwards compatible, we gave AI agents rules they can’t break: scoped permissions, locked budgets, and a record of every action anyone can check.

Since that release, we’ve learned that rules only build trust when they hold up under real-world pressure. An agent that people rely on for important tasks has to finish the job correctly even if the network hiccups halfway through. It has to handle the larger data that real work involves without losing the guarantees that make it trustworthy.

That’s what Talus Protocol v2.1 is about. It upgrades both the protocol’s developer tools and the Leaders (the network nodes that carry out agentic work), starting with how agents handle data too large to fit in a single blockchain transaction.

Bigger data, same guarantees

With Talus Protocol v2.1, Walrus, Sui’s decentralized storage network, is now a more native part of how agents handle data. Talus Tools can pass inputs and outputs of up to 8 MiB through Walrus, and Tool developers can work with it directly from the Talus CLI and SDK: uploading data, estimating storage costs, and keeping durable references to what they’ve stored.

Leaders read those references as part of the workflow, and the important part is what stays the same: the content is still signed, so it can be traced to the Tool that produced it. If anyone alters the stored data, the next step in the workflow rejects it. Our test suite checks for exactly that.

Uploads run through a shared signing wallet with spending limits, so storage costs follow the same budget rules as everything else on Talus.

Faster, more efficient Leaders

Talus has always kept finished steps from running twice. If a Tool has already made a swap or sent a payment, it shouldn’t run again just because a later step hit a problem.

In v2.1, we improved how Leaders use the Sui Network to track transaction results and coordinate work, so they recover from interruptions more efficiently and handle tasks without unnecessary loops. This release also aligns Talus with the latest version of Sui and its recent RPC changes, so agent work keeps running smoothly as the network evolves.

Easier to build

Talus Protocol v2.1 also gives developers new ways to build, test, and manage their agents:

  • Test against the real thing. Developers can run their TAPs (Talus Agent Package’s) tests locally against the actual published Talus contracts on Testnet or Mainnet, catching issues before they deploy.
  • Walrus from the command line. Upload data, see cost estimates before uploading, keep durable references, and recover uploads, all funded from your wallet.
  • A simpler starting point for Onchain Tools. New Tools start from a standard template, and developers who need workflow-level authorization can opt into it.
  • Reusable task inputs. A job’s inputs can be saved as a file and reused when scheduling it again.
  • Grants bound to a recipient. When scheduling a task, developers can tie a workflow grant to a specific recipient for Onchain Tools that require it.
  • Clearer cost tracking. Developers can see exactly which steps are still waiting to be paid or refunded.
  • Simpler network setup. The command line tool now connects directly through Sui’s network, with no extra configuration needed.

Bringing it together

The more an agent is trustworthy, the more it can be given to do. That trust comes from a foundation that keeps up as the work grows. With v2.1, Walrus becomes a native part of how agents handle data, Leaders run more efficiently on the latest version of Sui, and developers get more tools to build, test, and track their agents with confidence.

Start building