March 9, 2026
Agents, files, and fightin’ words
Launch HN: Terminal Use (YC W26) – Vercel for filesystem-based agents
Kubernetes diehards vs “Vercel for agents” hype: HN wants receipts
TLDR: Terminal Use launched a platform to deploy AI agents with persistent file storage and a simple CLI, but HN split fast: Kubernetes loyalists say they don’t need another tool, others press on storage (NFS, retries) and comparisons to sprites.dev and LangGraph. Everyone wants clearer scope and proofs it’s better than DIY
Terminal Use just dropped on Hacker News, pitching itself as “Vercel for agents” — a simple way to package your bot from a repo and run it with files it can read and write. The founders say it’s all about persistent workspaces, shared storage, and easy uploads via presigned links. It supports Claude and Vercel’s AI SDK out of the box, and promises one CLI to deploy, log, and roll back. But some features — like preview links and a low-level “just run this command” mode — are still on the roadmap. Demo here: YouTube, site here: Terminal Use.
The comments? Spicy. The loudest camp is the “I already run Kubernetes” crew, demanding why anyone needs yet another platform. Storage wonks swooped in with “NFS or roll-your-own?” vibes, prodding about checkpointing and retries while declaring the tooling “up in the air.” Another thread pressed, “Isn’t this just Fly.io’s sprites.dev? ELI5.” Power users kicked the tires with a wild real-world ask: a one-shot coding agent juggling Docker-in-Docker, Supabase, Next.js, and CI. And then the identity-crisis bomb: “Is this a replacement for LangGraph?” Cue the HN bingo card: Kubernetes? Check. ELI5? Check. “Is this like X?” Double check. Bottom line: builders love the storage-first angle, but the crowd wants sharper lines between “host your agent” and “design your agent,” plus proof it beats rolling their own
Key Points
- •Terminal Use launches a platform to deploy sandboxed, filesystem-dependent agents with unified packaging, lifecycle management, streaming, state, and file transfer.
- •Agents are packaged with a config.yaml and Dockerfile, deployed via CLI, and implement on_create, on_event, and on_cancel lifecycle endpoints.
- •Out-of-the-box adapters support Claude Agent SDK and Codex SDK; the platform is compatible with Vercel AI SDK v6 and includes a frontend provider.
- •Filesystems are first-class and decoupled from tasks, enabling persistent workspaces, sharing across agents, direct presigned URL transfers, and version-safe iteration.
- •Roadmap items include preview URLs and lower-level sandbox.exec APIs; deployment features include preview/production envs, git-based targeting, logs, and rollback.