Olas says its self-custodial AI agents are built to help prediction market newcomers

Olas says its self-custodial AI agents are built to help prediction market newcomers

N
News Editor
2026-09-14 11:29:49
Olas, previously known as Autonolas, is pitching a self-custodial AI agent framework for prediction market trading, with a particular focus on new users. In a discussion with Laura, co-founder David Minarsch said the system lets users run customizable local agents that can call on specialized agents and, on a pay-per-request basis, frontier models from labs including Anthropic and OpenAI. He argued that simply asking a model for a prediction and trading off that output is a naive approach, and framed Olas instead as a modular setup where users can choose what data to share and what services to pay for. Minarsch said Olas has already facilitated more than 14 million transactions between AI agents and settles onchain in USDC and xDAI. The conversation also touched on privacy concerns tied to frontier AI models during the dispute over whether OpenAI used a researcher’s work tied to a Navier-Stokes proof, with Minarsch hinting at an Olas response. He also traced the project’s roots to Fetch.ai in 2019, said ChatGPT changed the trajectory of the space, and noted that Olas Connect, launched a couple of months ago, gives local Claude Code or Codex sessions their own wallet so developers can build custom agents for Polymarket without relying on Pearl’s default app. Olas is also planning a Robinhood launch in the coming weeks and new DeFi-focused products.

Olas, formerly Autonolas, is positioning its self-custodial AI agents as a tool for prediction market users who are just getting started. In a conversation with Laura, co-founder David Minarsch said the project lets users spin up customizable local agents that can work with specialized agents to improve how they trade in prediction markets.

Local agents and pay-per-request model access

Minarsch said Olas users can run local agents and connect them to specialized agents, while also tapping frontier models from labs such as Anthropic and OpenAI on a pay-per-request basis. He described that setup as a way to make more advanced agent-based trading available to newcomers rather than only to highly technical users.

He also argued that asking a model for a prediction and placing a trade based on that answer alone is a naive way to trade. The product, as described in the discussion, is built around a more modular agent structure instead of a one-shot prompt-and-trade workflow.

More than 14 million agent-to-agent transactions

According to Minarsch, Olas has facilitated more than 14 million transactions between AI agents. The discussion also focused on ownership. He said Olas users actually own what they use within the system and can decide what they want to share and what they want to pay for.

For settlement, Olas uses USDC and xDAI onchain. That puts the final settlement layer on blockchain rails rather than inside a closed internal ledger.

Privacy concerns and dependence on remote AI models

The conversation moved into privacy concerns around frontier models during the dispute over whether OpenAI stole the work of a researcher tied to a Navier-Stokes proof. Minarsch teased an Olas solution, though the summary did not include further details.

Part of the Olas model, as presented in the episode, is that users choose what information they share and what requests they want to pay for. The team is also preparing an update on reducing dependence on remote AI models.

Project roots, Connect tool, and upcoming launch plans

Minarsch said Olas traces its roots to Fetch.ai in 2019, and that the launch of ChatGPT changed everything. He also highlighted Olas Connect, a tool launched a couple of months ago that gives a local Claude Code or Codex session its own wallet.

That setup allows developers to build custom agents that trade on Polymarket without using Pearl’s default app. Minarsch said Olas plans to launch on Robinhood in the coming weeks with new DeFi-focused products. He also teased a blog update in the next one to two weeks covering efforts to reduce reliance on remote AI models.

This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan.
900

Disclaimer:

The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.

Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.