weirdGuy
Kastor is a declarative language and toolchain — "Terraform for AI agents" — that lets developers define agents, tools, and prompts in typed HCL, then compile them to framework code or reconcile them onto hosted platforms with plan/apply semantics.
At a Glance
AI Tools by weirdGuy
(1)
Kastor
Declarative AI Agent Toolchain
Discussions
No discussions yet
Be the first to start a discussion about weirdGuy
Latest News
Products & Services
Open-source (Apache-2.0) Go CLI and declarative HCL-based language for AI agents. Parses .agent, .tool, .prompt, and kastor.hcl files, validates references and prompt variables at compile time, and offers two paths: `kastor build` compiles the spec to runnable framework projects (LangGraph today), and `kastor plan`/`apply`/`destroy` reconcile agents as long-lived resources on hosted platforms with state tracking, three-way diffs, and drift detection.
Market Position
Positions itself as "Terraform for AI agents" — an IaC-style source-of-truth layer that sits above agent frameworks and hosted platforms rather than competing with them as a runtime. It deliberately does not try to be another agent runtime; instead it standardizes the outer contract around agents (inputs, outputs, model, prompt, tools, build/apply target) so specs stay vendor-neutral, versionable, and reviewable even as underlying runtime patterns change.
Key Competitors
Leadership
Founders
weirdGuy
Anonymous solo developer and sole creator of Kastor, building the project in Go. Goes by the handle @so_weird_guy on X and weirdGuy/weirdguy on GitHub and Hacker News. Posted Kastor as their first-ever Show HN on July 8, 2026.
Founding Story
The creator built Kastor because agent definitions tend to be scattered across framework code, prompt files, tool/MCP files, platform UI settings, and env vars — making them hard to review, diff, reproduce, or move between platforms. Kastor is an attempt to add an infrastructure-as-code-style source-of-truth layer above that: a versionable, reviewable, declarative contract for agents. The bet is that agent runtime patterns will keep changing while the need for reviewable/diffable agent contracts will not, so Kastor deliberately standardizes the outer contract (inputs, outputs, model, prompt, tools, build/apply target) rather than trying to become another agent runtime.
Business Model
Revenue Model
Free and open-source software licensed under Apache-2.0. No monetization or paid tiers as of the alpha release.
Target Markets
- Managing AI agent definitions as versionable, reviewable infrastructure-as-code
- Defining agents once and generating code for multiple frameworks
- Deploying and reconciling agents across hosted platforms with plan/apply
- Detecting configuration drift in deployed agents
- Keeping agent contracts (inputs, outputs, model, prompts, tools) diffable in version control