Filecoin Onchain Cloud Explained

At a Fil Dev summit in Argentina, Juan Benet stood in a room full of developers and said something that surprised even the people who had been building alongside him for years. "Finally after six years, IPFS and Filecoin can finally work together," Benet said. Sarah Thiam, Head of DevRel at Filecoin and founder of the six-person developer relations entity Fil Builders, was in that room. "We were like, oh my gosh, yeah, it took us six years, what?" That moment, at the pre-launch of Filecoin Onchain Cloud (FOC) during Argentina's DevCon, marks the point where years of separate infrastructure finally started converging into a single developer-facing stack.

IPFS vs Filecoin Explained

The confusion between IPFS and Filecoin has been one of the more persistent issues in the ecosystem. Thiam is direct about the distinction. "IPFS is a peer-to-peer network, it's not blockchain. It does allow you to store, but more like cache, and then retrieve it really quickly in a peer-to-peer network. But it doesn't mean that your data is guaranteed to be stored forever." Filecoin, by contrast, is backed by a blockchain and a network of storage providers who are penalized if they lose data. The two technologies share overlapping primitives, including Merkle trees and CIDs for data integrity, but operate with fundamentally different economic guarantees.

Thiam joined Protocol Labs in late 2022, a few months before the Filecoin Virtual Machine (FVM) was set to launch. The FVM brought EVM-compatible smart contracts to Filecoin, which mattered because before that, making a storage deal meant manually contacting a storage provider. "We can't go around making storage deals by manually contacting through phone or email a storage provider and being like, hey, can you store my data? Like that's just not going to cut it." The FVM solved the programmability problem, but it surfaced a different one: speed.

72 Hours Down to Under a Minute

Filecoin was an early adopter of zero-knowledge proofs, using them as part of its storage verification process long before ZK proofs became a mainstream topic in the broader blockchain space. Storage providers are required to produce a proof showing they still hold the data, and generating that proof is computationally heavy. The result, even after the FVM launched, was that storage deals could take anywhere from 24 to 72 hours to finalize. The minimum amount of data that could be stored was 4 gigabytes, roughly the size of a 1080p movie, which was a meaningful barrier for most application developers.

Filecoin Onchain Cloud, which Thiam says began taking shape in late 2024 and early 2025, was designed around fixing those constraints. The work was led by Nicola, a researcher from Protocol Labs' cryptoeconomic lab, and built out by the FilOS engineering team under Molly and Jennifer. A new proof format called Proof of Data Possession (PDP) replaced the heavier ZK proof for this layer of the stack. The minimum storage size dropped from 4 gigabytes to 127 bytes.

"We are down to like seconds now rather than the hours I mentioned before, which is insane. 72 hours down to like... I think the fastest we got close to just under a minute, and we're still improving that drastically."

FOC went to mainnet last month, which Thiam describes as a significant milestone. The retrieval side of the stack got its own infrastructure in the form of Filebeam, a custom CDN that can return data in under ten seconds. For a file in the 1MB to 10MB range, retrieval currently runs around one second. Thiam notes that the cost of retrieval through Filebeam is comparable to AWS and slightly lower, with latency still being actively reduced.

Store, Pay, and Retrieve

The FOC stack is built as a modular set of smart contracts. Thiam describes three core components available today: the Filecoin Warm Storage Service (FWSS), Filecoin Pay, and Filebeam.

FWSS handles the storage side. It takes data, prepares it, sends it to a storage provider, and receives back the PDP proof. Filecoin Pay is the payment contract, and Thiam describes it as arguably the more foundational piece of the stack. The design ensures that payment only moves when verification has occurred.

"Only at the point when the storage provider returns the proof of storage or the proof of data possession, only then can they receive the payment and only then does the customer's funds actually get paid across. So until then, if nobody sends the proof back, you're not actually paying for anything."

That property makes Filecoin Pay a verifiable payment layer rather than just a billing system. Developers building on FOC are not trusting a service to report whether storage actually happened. The proof triggers the payment on-chain.

Filebeam sits outside the smart contract layer as a CDN, handling retrieval. Together the three components form what Thiam calls a store, pay, and retrieve foundation. The longer-term vision includes adding compute as a fourth component, so that workloads can run physically adjacent to where data is stored, reducing both cost and latency. Protocol Labs is actively looking for partners on the compute side.

For developers, the entry points into the FOC stack include the Synapse SDK and a Model Context Protocol (MCP) server that connects FOC storage to AI tooling. Filecoin Pin, a separate application built as a demonstration of the FOC stack, bridges the older IPFS interface to Filecoin-backed guaranteed storage, letting developers pin data to IPFS with automatic backup to the Filecoin network.

Agent Logs and Exabytes

The AI angle has become a meaningful part of how Filecoin is positioning the network. Thiam points to several use cases that have appeared during recent hackathons, including FIL Genesis and Synthesis: agent metadata, agent logs, episodic memory, and conversation context, all sitting in the kilobyte to gigabyte range and being stored on FOC today.

Further out, Thiam sees vector databases as a natural fit. "Filecoin has, I think, the largest decentralized storage network in terms of storage capacity today, we have like exabytes of data available," she says. Vector databases involve a mix of hot and cold storage that maps well to what Filecoin can offer, and synthetic data generation is another area she mentions as a candidate for the same infrastructure.

On the ERC-8004 Agent Registry side, Thiam describes a workflow where a developer takes the metadata URI of an agent card, stores it to IPFS via Filecoin Pin, and registers the result on-chain. Because the data is backed by Filecoin storage, the agent's record does not disappear if an IPFS node goes offline.

A grants program is in development, with Thiam saying the focus will shift toward milestone-based traction rather than purely build-focused deliverables, reflecting how quickly developers can now ship with tools like the Synapse SDK and MCP server. The AWS outage she referenced from Q4 of last year, which took down a range of services across the web, is the kind of event that keeps the decentralized storage conversation alive. Filecoin's capacity sits at exabyte scale, distributed across a network of providers with no single point of failure, and whether that capacity gets put to use at AI scale now depends largely on whether the developer tooling can keep pace.