Articles
Guides and comparisons on cloud storage, from S3, EBS, and EFS to the file systems built on top of them.
The Filesystem as Shared Memory for AI Agents
Shared memory for agents means four different things. How a persistent shared filesystem lets multiple agents coordinate, which POSIX primitives make it safe, and what performance it needs.
Hunter Leath
Archil vs Cloudflare Artifacts: which storage system is best for your agents?
Cloudflare Artifacts stores versioned Git snapshots of what agents produce; Archil gives agents a persistent, shared POSIX workspace over the data they work on. How the two differ, and when to use each.
Hunter Leath
Why databases and object storage can't be the agent context layer
Databases want schemas and object stores rewrite whole objects, but agents read and edit files in tight loops. Why a filesystem is the natural context layer for agents, and how to put one in front of the data you already have.
Hunter Leath
Scaling a multi-tenant AI agent product to thousands of tenants: provisioning, credentials and cost of one disk per tenant
How to run one disk per tenant at thousands of tenants: programmatic provisioning, disk-scoped credentials, microVM isolation and usage-based cost.
Hunter Leath
Giving AI agents file access to large S3, GCS and R2 datasets without copying them
How to give AI agents file access to large S3, GCS and R2 datasets by mounting the bucket, and the isolation, consistency and branching to plan for.
Hunter Leath
Isolating agent-generated code: microVMs vs gVisor vs containers, default egress policies and credential handling
Containers, gVisor or microVMs for agent-generated code, and why egress policy and credential brokering matter as much as the kernel boundary.
Hunter Leath
Persistent storage for AI agent sandboxes that survives restarts
What survives an AI agent sandbox restart, and how durable disks, microVM isolation, egress rules and fast branching keep an agent’s work across sessions.
Hunter Leath
Isolated per-tenant storage for multi-tenant AI agent applications
How to isolate per-tenant storage in multi-tenant AI agent apps: storage-layer scope, microVM execution, credential brokering and a disk per tenant.
Hunter Leath
Persistent sandboxes for long-running AI agents: keeping installed packages and files across restarts
What long-running AI agents need from a persistent sandbox: packages and files that survive restarts, durable writes, pause and resume, and branching.
Hunter Leath
Serverless bash execution on a persistent cloud filesystem for AI agents
How serverless bash execution on a persistent cloud filesystem works for AI agents: compute follows storage, microVM isolation, egress control and branching.
Hunter Leath
Mounting a POSIX filesystem directly over an existing S3 bucket without migrating data
s3fs-fuse, Mountpoint and S3 Files compared for mounting an existing S3 bucket as a POSIX filesystem: rename, locking, consistency, cost and isolation.
Hunter Leath
A persistent filesystem workspace for AI agents that carries across sessions and parallel runs
What a persistent filesystem workspace for AI agents needs: files that survive restarts, shared access across machines, and fast branches for parallel runs.
Hunter Leath
How fast is a serverless tool call for an AI agent?
Measured end-to-end latency of Archil serverless tool calls for trivial commands, ripgrep and a test suite, one at a time and in parallel, with a script to reproduce it.
Hunter Leath
How to fan out agent tool calls in parallel against one workspace
Keep the workspace on one disk and run each branch as its own exec call in its own microVM: read-only and conditional mounts, measured latency at 10 and 50 calls, and limits.
Hunter Leath
What does an idle agent sandbox cost per hour?
List-price cost of an idle 2 vCPU, 4 GiB sandbox on E2B, Daytona, Modal, Vercel, Cloudflare and Archil, how timeouts change the bill, and what per-call tool execution costs instead.
Hunter Leath
How to run agent tool calls without keeping a VM open
Put the agent's workspace on a persistent disk and run each bash or file tool as one disk.exec call in a fresh microVM, with nothing to keep alive between calls.
Hunter Leath
Do AI agents need a sandbox running for the whole session?
No. Most agent tool calls are short commands that only need the agent's files, so each can run in a short-lived microVM against a persistent disk. Here is when a long-lived sandbox still wins.
Hunter Leath
A POSIX file system on S3 for CI builds without Redis or a metadata database
Archil mounts a POSIX file system for CI with no Redis or metadata database to run. In its published one-host test, npm ci took 12.4 s against 53.8 s on JuiceFS writeback.
Hunter Leath
How to let several Vercel Sandbox agents write the same workspace at once
Only one Vercel sandbox can mount a Drive read-write. To let several sandboxes write one workspace at once, mount an Archil disk in each with conditional or shared mode.
Hunter Leath
Is a sandbox write durable after fsync? What another sandbox sees
What fsync durability means when another sandbox, snapshot or fork reads the data, shown with Archil's published Vercel Drive test, plus how to test your own sandbox storage.
Hunter Leath
Can you snapshot and branch a file system that syncs to your own S3 bucket?
Yes, on Enterprise. Archil checkpoints and branches a disk in tens of milliseconds, and branching a disk synced to your own S3, GCS or R2 bucket is an Enterprise feature.
Hunter Leath
How to move an agent's workspace between Daytona and E2B
Keep the repo, node_modules and build cache on one Archil disk that Daytona and E2B sandboxes both mount over FUSE, and hand the disk from one sandbox to the next.
Hunter Leath
How to give each Modal sandbox its own copy of a large dataset without duplicating it
To give each Modal sandbox its own writable copy of a large dataset, keep it on an Archil disk, checkpoint it once and mount one copy-on-write branch per attempt in a Modal VM sandbox.
Hunter Leath
Filesystem for AI agents with persistent shared storage
A filesystem for AI agents with persistent shared storage gives every agent a POSIX directory that survives restarts and that other agents and your app can read and write at the same time.
Hunter Leath
Give an agent access to an S3 bucket with grep, cat, and Python
To give an agent access to an S3 bucket with grep, cat, and Python, connect the bucket to an Archil disk and let the agent run ordinary shell commands against its objects as files.
Hunter Leath
Fork a large dataset copy-on-write for parallel agent attempts
To fork a large dataset copy-on-write for parallel agent attempts, checkpoint the Archil disk that holds it and create one branch per attempt, which takes milliseconds regardless of dataset size.
Hunter Leath
Persistent workspace for sandbox agents across restarts
To give sandbox agents a persistent workspace across restarts, keep the workspace directory on an Archil disk that outlives the sandbox and mount it in every new sandbox.
Hunter Leath
Shared storage for a dataset of hundreds of GB across parallel sandboxes
To share a dataset of hundreds of GB across parallel sandboxes, store it once on a shared Archil disk and mount it read-only in every sandbox instead of copying it into each one.
Hunter Leath
What isolation do agent sandboxes actually use? Firecracker, gVisor and containers compared
Which boundary E2B, Vercel, Modal, Daytona, claude.ai and Archil document, what containers, gVisor and Firecracker each protect against, and what isolation alone leaves open.
Hunter Leath
Archil vs Amazon EFS for AI agents
Where Archil is faster than Amazon EFS for agent workloads, where EFS still wins, and what each leaves you to build: sandboxes, branching, bucket access and per-tenant isolation.
Hunter Leath
Archil vs Amazon S3 Files for agent workspaces
Both put a file system in front of object storage, at different layers. Where S3 Files is the simpler AWS-native choice, where Archil fits better, and the plan limits that matter.
Hunter Leath
Running parallel agent attempts from one starting state with copy-on-write branches
Prepare an agent environment once, fork it per attempt with copy-on-write disk branches or sandbox forks, score the results and keep the winner. What it costs and where it stops working.
Hunter Leath
How to keep an AI agent's environment across sandbox restarts
Four ways to keep packages, repositories and caches across agent sandbox restarts, what each one costs, and why persistence and isolation have to be designed together.
Hunter Leath
Choosing the Right Storage for AI and ML Workloads
How to design a storage strategy for AI systems, from large-scale training datasets to inference and AI agents.
Hunter Leath
Storage in the AI Pipeline: From Ingestion to Training
Storage architecture can make or break your AI pipeline. From data ingestion to model training and serving, each phase has different performance and cost requirements. This post breaks down storage tiers, tradeoffs, and practical strategies for optimizing performance without blowing up your cloud bill.
Hunter Leath
Is S3 POSIX compliant?
Learn what POSIX compliance means, how it differs from object storage, and why S3 is not POSIX-compliant. Covers the practical impact on application behavior, common migration pitfalls, and which cloud file storage options support true POSIX semantics for workloads that need them.
Hunter Leath
When should I be using DeepSeek 3FS for AI?
DeepSeek 3FS is an open-source distributed file system built specifically for large, read-heavy AI workloads. It delivers strong consistency and fast random reads, but comes with slower writes and significant operational complexity, making it a good fit only for teams running AI at serious scale.
Hunter Leath
JuiceFS vs ObjectiveFS
Compares JuiceFS and ObjectiveFS, two POSIX-compatible filesystems built on object storage, focusing on their architectural tradeoffs, metadata handling, operational overhead, and consistency models. Helps teams decide whether they prefer an open, operator-managed system or a fully managed filesystem with stronger built-in guarantees and simpler operations.
Hunter Leath
GoofyFS vs s3fs?
Compares s3fs and Goofys, two tools for mounting S3 as a local filesystem, and explains how their metadata handling, caching strategies, and API usage impact performance, consistency, and cost. Helps teams understand when filesystem-on-S3 layers are sufficient—and when object-native approaches are a better fit for large-scale data workloads.
Hunter Leath
GoofyFS vs ObjectiveFS
Compares Goofys and ObjectiveFS, two ways to mount object storage as a filesystem, and explains how their goals differ. Highlights the tradeoff between lightweight S3 adapters optimized for simplicity and speed versus fully featured POSIX-style filesystems built for consistency, shared access, and multi-machine workloads.
Hunter Leath
What is ObjectiveFS?
Explains what ObjectiveFS is and how it delivers a fully POSIX-compatible shared filesystem on top of object storage. Covers its architecture, consistency model, performance characteristics, and where it fits best for teams that need scalable, cloud-backed file storage without running a storage cluster.
Hunter Leath
How do I do EFS on Other Clouds?
This guide compares AWS EFS and its cloud storage alternatives—Azure Files, GCP Filestore, Oracle, and Archil—highlighting differences in scalability, performance, and cost to help teams choose the best managed NFS solution for their workloads.
Hunter Leath
12 AWS Products for Slashing Costs
A tactical guide to cutting AWS costs fast—combining 10 proven AWS tools with Archil’s S3-backed, infinite local disk to shrink infrastructure spend, extend runway, and boost efficiency without slowing product velocity.
Hunter Leath
Tutorial: Archil with Wordpress
Learn how to host and scale your WordPress site on AWS using Archil—mount S3 as a local file system for faster performance, simple deployment, and cost-efficient cloud storage without the complexity.
Hunter Leath
What are S3 Vectors, and Where Does Archil Fit In?
Discover how Amazon S3 Vectors brings vector-native storage to the cloud—and how Archil complements it by delivering local-speed, real-time access to vector data for next-generation AI workflows.
Hunter Leath
Stop Syncing and Start Mounting: A Smarter Alternative to aws s3 sync
Stop wasting time with aws s3 sync — learn how Archil lets you mount Amazon S3 as a live, POSIX file system for instant access, faster performance, and lower costs.
Hunter Leath
Scaling S3: Navigating request Rates, throughput, scaling, and S3TA
Optimize Amazon S3 performance with key prefix scaling, S3 Transfer Acceleration, and multi-part uploads to achieve faster throughput, lower latency, and global reliability at any scale.
Hunter Leath
Mounting S3 Buckets with s3fs: Pros, Cons, and Faster Options
Learn why tools like s3fs struggle to make S3 behave like a true file system—and how Archil delivers a faster, reliable, POSIX-like alternative built for modern, cloud-native performance.
Hunter Leath
Tutorial: Jupyter Notebooks on Archil
Set up Jupyter on AWS EC2 with S3 and Archil to build a high-performance, cloud-native analytics environment—mount S3 as a local file system for instant, code-free data access and faster analysis.
Hunter Leath
Data Lakes vs. Data Warehouses: Which Model Fits Your Use Case?
Learn the key differences between data lakes, data warehouses, and hybrid solutions—and how to choose the right architecture for analytics, scalability, and real-time insights.
Hunter Leath
Cache Layers vs Storage Classes for Performance
Compare Amazon S3 Express One Zone and caching solutions like Archil—discover when to use high-performance S3 storage vs. dynamic, cost-efficient caching for ultra-low-latency workloads.
Hunter Leath
AWS EBS Made Simple: Picking the Right Storage Without Overspending
Understand AWS EBS volume types, pricing, and cost-optimization strategies—learn how to choose the right storage class, manage performance, and reduce your AWS bill effectively.
Hunter Leath
Amazon EBS vs EFS: Where Does an S3-Compatible Option Fit In?
Compare AWS EBS vs EFS vs S3—and learn how Archil bridges them by turning S3 into a high-performance, mountable file system that delivers local-disk speed with cloud-scale flexibility.
Hunter Leath
Amazon EBS Comparison & Block Storage Alternatives
As companies scale, AWS EBS’s costs and performance limits become clear. Alternatives like Ceph, GlusterFS, and caching layers offer more flexibility and speed. Modern architectures—like Archil’s—combine object storage with caching to deliver faster, more scalable cloud data access.
Hunter Leath
Why S3 Performance Limits Matter — and How Archil Solves Them
AWS S3 is known for its massive scale, eleven nines of durability, and global availability. It’s the go-to storage layer for enterprises managing everything from logs and backups t
Hunter Leath
What is Lustre? And When Should You Use Something Simpler?
Lustre is a high-performance distributed file system built for HPC: fast, parallel, and scalable. But as cloud-native, elastic workloads rise, its rigid, server-centric design raises questions about fit in the modern data stack.
Hunter Leath
S3 Intelligent Tiering
S3 Intelligent Tiering auto-optimizes storage costs by moving data across access tiers based on usage—no retrieval fees, no latency impact, and up to 95% savings. Ideal for large, infrequently accessed files. Not worth it for tiny or short-lived objects.
Hunter Leath
EFS vs. S3: Network File Store vs Object Store
Amazon EFS and S3 serve different needs—EFS offers shared file system access with low latency, while S3 is built for scalable, low-cost object storage. This article compares performance, pricing, and use cases, and introduces Archil as a hybrid solution that blends S3’s affordability with file system speed.
Hunter Leath
Serverless vs. Kubernetes: Choosing the Right Cloud Architecture for Modern Workloads
Discover the pros and cons of serverless vs. Kubernetes for modern cloud architecture. Learn when to use each, how to handle persistent storage, and how Archil bridges the gap for scalable, high-performance data access.
Hunter Leath
EBS vs S3
Confused about when to use Amazon S3 vs EBS? This in-depth guide compares AWS storage options—performance, pricing, latency, throughput, and real-world use cases—to help you choose the right fit for your application. Learn how cloud-native apps, databases, and AI workloads benefit from the right blend of EBS and S3, plus discover automated caching solutions for maximum speed and savings.
Hunter Leath
Amazon FSx: What It Is and When to Look for Alternatives
Compare Amazon FSx file systems—Windows, Lustre, ONTAP, OpenZFS—and discover when modern storage alternatives like Archil deliver better performance.
Hunter Leath
Showing 12 of 61 articles