Archil

Archil, formerly named Regatta Storage, builds a cloud file system aimed at AI-native applications. Its core product virtualizes object storage — primarily Amazon S3 buckets — and presents it as a POSIX-compatible, locally mounted file system with effectively unlimited capacity. The company states the product delivers roughly 30x faster access than reading S3 directly and about 90% lower cost than manually managed EBS disks, using custom data protocols and NVMe SSD-backed shared caching. Because the product presents as a standard file system (supporting renames, appends, and locking), the company says applications require no code changes; data written locally is flushed back to the underlying S3 bucket within minutes. Mounting requires only a volume name, cloud region, and data source location.

The company has extended beyond storage into broader agent infrastructure. Current documentation describes versioned/branchable disks with checkpointing, serverless execution and persistent sandboxes for running untrusted AI-generated code against mounted data, native SQLite database support on the file system, immutable log support, and a strongly consistent S3-compatible REST API for sharing artifacts between agents and humans. Multiple context sources (for example separate S3 buckets for skills, customer data, and run logs) can be composed with per-source permissions into a single workspace mounted at /mnt/archil and shared live across agents without copying data. Integrations exist with agent frameworks (AI SDK, eve, Mastra, LangChain), TypeScript and Python SDKs, sandbox providers (Daytona, e2b, Superserve), Kubernetes, and Terraform.

In a June 2025 post the company framed itself as a data company rather than a storage vendor, outlining planned expansion into connectivity with additional data sources (Hugging Face, GitHub, data warehouses, SFTP), serverless data transformations at the storage layer, and built-in versioning, locality awareness, and access controls.

Founding story

Founder and CEO Hunter Leath spent nine years building cloud storage products, leading Amazon's Elastic File System product and Netflix's cloud storage team. He observed that AI and analytics teams store data in S3 for cost and scalability but must copy it into local file systems to use it, causing hours-long downloads, fixed capacity limits, no sharing, and roughly 10x higher cost; Archil was created to remove that data-movement step. The broader team says it previously built and operated storage at AWS, Netflix, Uber, Datadog, Pure Storage, Snowflake, and Chroma.

Business model

Archil sells cloud storage and adjacent compute infrastructure directly to developers and engineering teams on a usage-based basis, positioning itself as a replacement for self-managed block storage (EBS) and direct S3 access.

Consumption pricing that is described as fully elastic: for disks synchronized to a customer-owned S3 bucket, the customer pays only for "Active Data" (recently read or written), and nothing when data is idle and resident only in S3. Sandbox compute is billed only for active time (site example: $0.18/hr active compute, $0 idle), and each 1 TB of Active Data includes 1,000 free minutes of sandbox compute.

Traction

The Y Combinator profile lists a team size of 11 and status Active; the company's about page lists 17 named team members across engineering, GTM, operations, and talent. The website states it is used by teams shipping AI in production, without naming customers. Total disclosed funding is $18M.

Latest developments

The website's current banner announces Serverless Sandboxes, and messaging has shifted toward running large numbers of stateful agents with versioned disks, serverless sandboxes, and SQLite databases. The about page discloses an $11M Series A led by Standard Capital and $18M raised in total.