Last updated: 2026-08-12
Archil is a hosted file system, built by former Amazon EFS and Netflix storage engineers, that mounts S3, GCS, and other object stores as a POSIX disk claimed up to 100x faster than raw S3 for git-style operations. It adds serverless Bash, Python, and Node sandboxes billed per active minute for AI agent and ML workflows.
About Archil
Hunter Leath founded Archil after nine years in cloud storage, first as a founding engineer and senior product manager on Amazon EFS, then leading cloud storage at Netflix. Founded in 2024 as Regatta Storage and rebranded to Archil in 2025, the product mounts existing object storage buckets, including Amazon S3, Google Cloud Storage, Cloudflare R2, and Azure Blob, as a real POSIX filesystem instead of requiring a separate download or ETL step before AI, analytics, or agent workloads can touch the data. The core mechanism is an SSD caching layer sitting in front of the source bucket, so repeated reads and small-file operations hit local disk speed instead of round-tripping to object storage on every request. On top of the mounted filesystem, Archil sells serverless sandboxes for running arbitrary shell, Python, or Node scripts against that cached data, with compute billed only while a script is actively executing and nothing charged between invocations. Disk versioning and branching let separate agent runs check out independent snapshots of the same underlying data set without overwriting each other's work. The product is built for ML engineers training on large S3-hosted data sets, teams operating fleets of AI coding or research agents that need persistent and shareable storage, and infrastructure engineers who currently maintain custom S3 sync or ETL pipelines. A Kubernetes CSI driver and a dedicated Terraform provider extend the same mounted-disk model into container orchestration and infrastructure-as-code workflows, and TypeScript and Python SDKs plug directly into the Vercel AI SDK, LangChain, and Mastra for agent-facing integrations. Pricing scales from a free developer tier meant for evaluation and small projects, through a fixed-price team plan aimed at production workloads with a signed BAA and DPA available, up to custom enterprise agreements that add bring-your-own-cloud deployment, SSO and SCIM, and an availability SLA. The product is accessed through a web console, a command-line interface, and language SDKs rather than a desktop or mobile app, since it is infrastructure rather than an end-user application. Archil has raised $18M in total funding, including a seed round led by Felicis Ventures with participation from Y Combinator and Peak XV Partners, and a later Series A led by Standard Capital. Investors also include storage-industry angels such as Modal founder Erik Bernhardsson and former AWS storage executive Wayne Duso, and the company lists early adopters including Depot, Gamma, and Gumloop as logo customers, though it has not yet published detailed case studies for any of them.
Pricing
Developer plan: free, no credit card needed, 10GB of storage plus free sandbox minutes each month; overage runs $0.30/GB-month storage and $0.27/hour compute. Team plan: $500/month covers 1TB of storage and 1,000 sandbox-minutes; overage runs $0.20/GB-month storage and $0.18/hour compute, with a signed BAA and DPA available. Enterprise: custom pricing with bring-your-own-cloud, SSO/SCIM, and an availability SLA. Archive storage is $0.025 per GB-month, out-of-region egress is $0.05 per GB, and there are no IOPS charges or minimum commitments.
Key Features
- S3-as-filesystem mounting: Mounts S3, GCS, Cloudflare R2, Azure Blob, and other object stores as a real POSIX directory with no separate ETL step, with reads claimed up to 30x faster than accessing S3 directly.
- Idle-free serverless sandboxes: Runs Bash, Python, and Node commands directly against the mounted disk through disk.exec, billing only for active compute minutes and nothing while idle.
- Disk versioning and branching: Creates checkpoints and branches of a mounted data set so parallel agent runs can work against independent snapshots without overwriting each other.
- Kubernetes CSI driver and Terraform provider: Provisions ReadWriteOnce and ReadWriteMany volumes inside Kubernetes clusters and manages Archil disks as infrastructure-as-code through a dedicated Terraform provider.
- Native SQLite and append-only logs: Lets agents write structured SQLite databases and immutable append-only logs directly to the mounted filesystem instead of standing up a separate database.
- Built-in filesystem search: Ships an Archil grep function in its SDKs for searching mounted filesystem data, aimed at agents working across large, growing file trees.
- Framework SDKs: Provides TypeScript and Python SDKs plus dedicated packages for the Vercel AI SDK, LangChain, and Mastra, with a documented Claude Sonnet 4.6 code example.
- Role-based mount access: Issues per-source read-only or read-write mount tokens and delegated ownership so multiple agents or users can share the same mounted data safely.
Pros
- Raised $18M in total funding, including investors with direct storage-industry pedigree such as Modal founder Erik Bernhardsson and a former AWS storage executive, unusual validation for a 2-year-old infrastructure startup.
- Founder Hunter Leath spent nine years in cloud storage as a founding engineer and senior PM on Amazon EFS and later led cloud storage at Netflix, bringing production storage-systems depth that's uncommon at this stage.
- Idle-free sandbox billing means compute stops accruing between commands, unlike always-on VM-based dev environments that charge whether or not code is actually running.
- Includes a container-orchestration path via a dedicated CSI driver for Kubernetes volumes and an infra-as-code path via a Terraform provider, so platform teams provision disks the same way they provision any other cloud resource.
Cons
- Data written to the underlying S3 bucket only becomes visible inside Archil, and vice versa, on an eventual-consistency basis measured in seconds to minutes rather than instantly.
- Serverless sandbox execution is AWS-only as of the most recent public documentation; Google Cloud buckets can be mounted but cannot yet run through disk.exec.
- No G2, Trustpilot, Capterra, or Product Hunt listing exists yet, so there is no independent third-party review signal to weigh against the vendor's own claims.
- The dedicated security page attributes SOC 2 Type II, ISO 27001, and FedRAMP certifications to underlying AWS infrastructure rather than asserting them as Archil's own, a distinction buyers evaluating compliance should confirm directly.
Frequently Asked Questions
How much does Archil cost in 2026?
Archil's Developer plan is free with no credit card required, including 10GB of storage and a monthly allotment of free sandbox minutes; usage beyond that is billed at $0.30 per GB-month for storage and $0.27 per hour for compute. The Team plan costs $500 per month for 1TB of storage and 1,000 sandbox-minutes, with overage priced at $0.20 per GB-month and $0.18 per hour, plus a signed BAA and DPA for compliance-sensitive workloads. Enterprise pricing is custom, quoted per account with dedicated support terms. Archive storage runs $0.025 per GB-month and out-of-region egress is $0.05 per GB, with no IOPS charges or minimum commitment on any plan.
Is Archil free to use?
Yes. Archil's free Developer plan includes 10GB of performance storage, up to five mounted file systems, and a monthly allotment of free sandbox compute minutes, with no credit card required to sign up. It's intended for evaluation and small projects; usage beyond those limits switches to Archil's standard metered billing. Teams needing production-grade storage or compute typically move to the Team plan once they outgrow the free quotas.
What are the best alternatives to Archil?
The closest alternatives are e2b and Daytona for teams that mainly need ephemeral AI agent sandboxes without a persistent, S3-backed disk, and AWS Mountpoint for S3 or the open-source s3fs-fuse for teams that want a free, self-managed FUSE mount without Archil's caching layer or serverless billing. Modal is a common alternative for teams whose primary need is general-purpose serverless compute rather than a filesystem layer. Choose Archil specifically when you need a managed, cached POSIX view of object storage combined with per-minute sandbox billing.
How does Archil compare to e2b in 2026?
Archil and e2b both offer sandboxes for running AI agent code, but they solve different halves of the problem: e2b focuses on ephemeral, isolated code execution environments, while Archil adds a persistent, S3-backed POSIX filesystem that sandboxes, including e2b's, can mount for shared and versioned data. Archil's own documentation lists e2b as a compatible sandbox integration rather than a strict competitor. Choose e2b alone for short-lived, stateless code execution, and add Archil when agents need to read, write, and version large data sets across runs.
How do you get started with Archil?
Sign up for the free Developer plan at console.archil.com, which requires no credit card. Install the TypeScript SDK, published as disk on npm, or the Python SDK via pip install archil, then create a disk and mount an existing S3, GCS, or R2 bucket. From there, call disk.exec to execute a shell script or Node/Python snippet against the mounted data, or wire in the Vercel AI SDK, LangChain, or Mastra integration for agent workflows.
Top Alternatives
- e2b: Pick Archil if agents need persistent, S3-backed storage tied to the sandbox; pick e2b if you only need ephemeral code execution.
- Modal: Pick Archil if the workload is data-mounting and agent storage; pick Modal if you need general-purpose serverless compute functions.
- Together AI: Pick Archil if you need a filesystem layer under your own agent stack; pick Together AI if you need hosted inference APIs for open models.