Axelar has introduced AgentFlux, an open-source framework aimed at executing AI agents locally while safeguarding private keys, trading strategies, and client data from the cloud — particularly appealing to institutions interested in onchain finance yet cautious about privacy concerns.
Developed by Interop Labs, creators of the Axelar network, AgentFlux enables financial firms to implement “agentic” automation without disclosing sensitive information to external systems, the company reported on Thursday.
The framework addresses a crucial pain point in AI-driven crypto operations: tool-calling.
Currently, most agents depend on cloud models to determine which blockchain tools to use and how to structure transactions, potentially exposing the sensitive data institutions strive to protect. AgentFlux separates these responsibilities into two distinct, specialized models — one for selecting the right tool and another for formulating the arguments needed to execute. According to Axelar’s team, this approach enhances tool-calling precision by 46% in benchmark tests, making local models nearly as effective as larger cloud systems.
Sergey Gorbunov, Axelar’s co-founder, mentioned in a CoinDesk interview that he envisions AgentFlux benefiting various functions: “The first is highly sophisticated trading strategies, which are incredibly proprietary,” he noted. “You would never want public cloud models to match your strategy. Uploading them would inherently compromise their integrity. That’s a key concern.”
“Secondly, I foresee blockchains being extensively used for analysis. For instance, tax reporting or investigating transaction activities. If you have a specific fingerprint you suspect could indicate malicious actions across several accounts, you wouldn’t want to upload that to a cloud model or an AI system. You’d want to analyze all correlated transactions and suspicious patterns privately,” Gorbunov added.
AgentFlux also integrates into Axelar’s wider multichain strategy. The team positions Axelar as a “gateway to onchain finance,” offering infrastructure to help institutions transfer assets and data across blockchains from one integration point. AgentFlux could allow a single AI agent to assess risk, evaluate exposure, and conduct transactions across different ecosystems — a capability the company views as vital for institutional adoption.
Read more: Canary Capital Files to Launch ETF Tracking Cross-Chain Protocol Axelar
