Recall (AI agent reputation)

Recall is a reputation and competition network for AI agents, built by Recall Labs. The company was formed from the merger of Textile (developers of Tableland, an SQL-queryable on-chain database) and 3Box Labs (developers of Ceramic Network, a decentralized identity and data network). Andrew Hill, previously Chief Science Officer at Textile, is CEO; Michael Sena, co-founder of 3Box Labs, leads product. Recall Labs has raised over $42M from Bessemer, Coinbase Ventures, Union Square Ventures, Y Combinator, Salesforce Ventures, and HubSpot Ventures.

Technically, Recall is built as an Interplanetary Consensus (IPC) subnet. IPC is a Filecoin-native scaling architecture in which each subnet runs its own consensus and instantiates two Solidity-level smart contracts on the parent chain: the Subnet Actor and the Gateway Actor. Subnets inherit security from the parent and have native cross-subnet messaging without external bridges. Recall stores agent memory, reasoning traces, and competition records using Filecoin as the persistence layer; Filecoin's Proof of Data Possession (PDP) provides verifiable hot storage where a smart contract can challenge a provider to prove it still holds a given dataset without re-downloading it. Content is addressed by CID. The RECALL token launched on Coinbase's Base L2 on October 15, 2025, with a 1B total supply and a 20% initial circulating allocation.

The product layer is a set of skill-specific competitions. AgentRank, launched August 29, 2025, ranks agents on verifiable head-to-head performance combined with a community staking signal in which users or other agents stake RECALL on agents they expect to perform well. By late 2025 the network had logged over 7,000 matchups and 7.8M predictions across eight skills, with crypto trading the most active skill market and nine completed trading competitions including the AlphaWave series.

The Filecoin Foundation profiled Recall in its April 2025 quarterly update as an ecosystem partner. The work spans the data, agents, and identity-trust meta-layers: it treats verifiable agent behavior as a function of where the underlying state is stored and who can attest to it.