Launch HN: Terminal Use (YC W26) – Vercel for filesystem-based agents

Published: 5 hours ago (March 9, 2026 at 12:53 PM EDT)

3 min read

Source: Hacker News

Overview

Hello Hacker News! We’re Filip, Stavros, and Vivek from Terminal Use (https://www.terminaluse.com/). We built Terminal Use to make it easier to deploy agents that work in a sandboxed environment and need filesystems to do work. This includes coding agents, research agents, document‑processing agents, and internal tools that read and write files.

Demo

A short demo of the platform is available on YouTube:
https://www.youtube.com/watch?v=ttMl96l9xPA

The Problem

Hosting agents traditionally requires stitching together multiple pieces:

Packaging the agent
Running it in a sandbox
Streaming messages back to users
Persisting state across turns
Managing file transfer to and from the agent workspace

Our Approach

We wanted something like Cog from Replicate, but for agents: a simple way to package agent code from a repo and serve it behind a clean API/SDK, without constraining the agent logic or harness.

Packaging & Deployment

Package your agent from a repository with a config.yaml and a Dockerfile.
Deploy it using our CLI.

You define the logic of three endpoints that track the lifecycle of a task (conversation):

on_create
on_event
on_cancel

The config.yaml contains details about resources, build context, etc.

SDK Compatibility

Out of the box we support Claude Agent SDK and Codex SDK agents via adapters that convert SDK message types to ours.
For custom harnesses, you can send messages using our types (compatible with Vercel AI SDK v6).
Front‑end integration is provided through a Vercel AI SDK provider and a messages module, removing the need to manage streaming and persistence manually.

Storage – First‑Class Filesystems

We treat filesystems as first‑class primitives, separate from the task lifecycle. This enables:

Persisting a workspace across turns
Sharing a workspace between different agents
Uploading / downloading files independent of an active sandbox

Our filesystem SDK provides presigned URLs, allowing users to transfer files directly without proxying through your backend.

Benefits

Decoupled agent logic and storage makes it easy to iterate on agents without worrying about sandbox files.
Deployments can auto‑migrate existing tasks to a new version, or keep existing tasks on the old version while new tasks use the updated deployment.
Support for multi‑filesystem mounts with configurable mount paths and read/write modes, keeping storage durable and reusable while allowing task‑specific mount layouts.

Deployment Experience

Inspired by modern developer platforms, Terminal Use offers:

Simple CLI deployments
Preview and production environments
Git‑based environment targeting
Logs and rollback capabilities

All configuration needed to build, deploy, and manage resources lives in the config.yaml, making CI/CD integration straightforward.

CLI for Coding Agents

Our platform is explicitly designed for CLI‑based coding agents:

Send messages to deployed agents from the CLI
Download filesystem contents to inspect agent output

A typical testing workflow:

Create markdown files describing user scenarios.
Use Claude Code to impersonate users and chat with the deployed agent.

Roadmap

What we do not have yet:

Full parity with general‑purpose sandbox providers (e.g., preview URLs, low‑level sandbox.exec(...) APIs). These features are on the roadmap.

Call for Feedback

We’re excited to hear any thoughts, insights, questions, and concerns in the comments below!