July 8, 2026
AI drama, now in config files
Show HN: Kastor – Terraform-style specs for AI agents
A new tool wants to organize AI chaos, but commenters are already asking if it’ll age like milk
TLDR: Kastor is a new tool that tries to turn messy AI assistant setups into clean, trackable files teams can manage like infrastructure. Commenters were split between “finally, some order” and “this field changes so fast that standardizing it now is like nailing jello to a wall.”
A developer rolled into Hacker News with Kastor, a new project pitched as “Terraform for AI agents” — basically, a way to describe chatbots and AI helpers in one tidy, reusable set of files instead of scattering settings across code, hidden menus, and mystery configuration. In plain English: it’s trying to give messy AI projects a single source of truth so teams can review changes, compare versions, and move between platforms without losing their minds.
But the real show was the comment section, where the crowd instantly split into “oh, that’s neat” and “this is way too early, good luck pinning down a blob of chaos.” One commenter gave the breezy thumbs-up — “terraform for agents? neat” — while another came in swinging with the line of the thread, saying this whole idea is like “trying to nail jello to a wall.” Ouch. That hot take captured the big fear: AI “agents” are still such a fuzzy, fast-changing concept that any attempt to lock them into a standard might be obsolete before the ink is dry.
And then there was the classic internet drive-by: “The website is down?” Nothing says launch-day energy like someone checking the front door before reading the pitch. Another skeptic asked the most practical question of all: if this is so close to Terraform, why not just make it an actual Terraform module? So yes, the tool promises order for AI chaos — but the community seems just as interested in whether the chaos is even ready to be organized
Key Points
- •Kastor defines AI agents through typed declarative spec files in HCL and positions itself as a vendor-neutral source of truth for agent configuration.
- •The Go-based toolchain supports `kastor build` for generating runnable projects and `kastor plan`/`kastor apply` for reconciling hosted agents with state, diffs, and drift detection.
- •The project can be installed via Homebrew, a checksum-verifying install script, `go install`, or release archives verified against checksums.
- •The quickstart compiles a weather example into a LangGraph project that requires Go, Python, an OpenAI API key, and a Tavily API key for the hosted MCP search tool.
- •A documented v0 limitation is that upstream agent references are validated and ordered at compile time, but generated code does not automatically execute dependent agents.