Company profile · 4 min read
Archil: versioned storage and sandboxes for stateful AI agents
Archil gives AI teams governed filesystem access to production data, with sandboxes, versioned disks and agent-facing workflows.
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
Published · Updated
Archil is trying to remove the copy step between S3 and the agent.
What it does
Archil turns cloud data into a filesystem that AI applications can use without first pulling everything onto a local disk. Its current homepage is aimed at agent platforms: versioned disks, sandboxes, databases, shared context and artifact sharing. The documentation describes connections to S3, serverless execution, SDKs and integrations with agent frameworks.
The useful product boundary is not “faster storage.” It is whether an agent can work against governed data where it already lives. Archil’s homepage shows a workflow in which an agent mounts a filesystem, searches a customer folder, runs code, writes an artifact and preserves the filesystem after the sandbox is released. Those are company-described examples, not an independent performance test.
Why I’d look closer
The strongest fit is a team running stateful agents against large or sensitive datasets. Archil says context can be layered by customer, agent and business, with permissions attached to each layer. Its checkpoint and branching model is also a sensible answer to a familiar agent problem: let the agent make a mess in a branch, then keep the result that worked.
The pricing makes the first decision fairly clear. The Developer plan is listed at $0 per month with 10 GB of performance storage and 30 sandbox minutes. Team is listed at $500 per month with 1 TB, 1,000 sandbox minutes and unlimited file systems; usage beyond the included amounts is priced separately. Those terms are on Archil’s pricing page. Enterprise adds options such as BYOC or on-premises deployment, SSO/SCIM and dedicated support.
Hunter Leath’s public YC profile says he spent nine years building cloud-storage products at AWS and Netflix. That background is relevant because Archil is selling a storage abstraction, not only an agent wrapper. It does not prove the product is right for every data platform.
What I’d ask
I would ask how permissions map to a real organization, how branch and checkpoint retention is billed, what happens when an agent writes back to a source bucket, and which workloads need the Team plan rather than the free developer tier. The docs are strong on the primitives. A buyer still needs to validate migration, operational ownership and the exact support boundary.
My editorial take
Shortlist Archil if your agents are blocked by data movement, cold starts or duplicated working copies. It is less compelling for a small stateless workflow that can read a few API records and finish. The bet here is that storage, execution and agent context become one operating surface. That is a useful bet when the data is the bottleneck.
Quick facts
| Field | Sourced detail |
|---|---|
| Product | Filesystems, sandboxes and agent data infrastructure |
| Buyer | AI platform, analytics and data-intensive application teams |
| Pricing | Developer $0; Team $500/month; Enterprise custom |
| Primary tradeoff | More governed context and versioning, with a new storage layer to operate |
| Public security/business options | BAA/DPA listed for Team; BYOC, on-premises and SSO/SCIM listed for Enterprise |
Sources checked
| Source | Checked |
|---|---|
| YC company profile | 2026-09-19 |
| Archil homepage | 2026-09-19 |
| Archil pricing | 2026-09-19 |
| Archil documentation | 2026-09-19 |
Cohort context
Archil is listed in Fall 2024. In our 2026-09-18 directory snapshot, 57 of 94 listed companies in that cohort have YC’s primary industry label B2B (60.6%). This is a current-directory comparison, not an original intake count or a performance ranking. Nine-cohort dataset.
Public website snapshot
Observed 2026-09-19T16:14:31.119Z in raw homepage HTML. This records visible metadata and advertised links, not agent execution or product quality.
| Signal | Homepage observation |
|---|---|
| Product description metadata | Observed |
| Canonical link | Not observed in this response |
| H1 or H2 heading | Observed |
| Typed structured data | Not observed in this response |
| Docs/developer link | Observed |
| Pricing link | Observed |
| llms.txt link | Not observed in this response |
| Markdown alternate | Not observed in this response |
Public observations · Collection method. Missing links here do not establish that a capability or file is absent elsewhere.
About the author
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
I cofound Lazyweb and publish Mudpie. This is an owner-written publication, not an independent testing organization. Research notes distinguish observations, sourced reporting and editorial judgment.
First1000 ↗ · X ↗