weirdGuy
Kastor is an open-source, declarative source-of-truth layer for AI agents. It lets developers define agents, tools, prompts, models, and deployment targets in typed HCL, then validate the spec, compile it into runnable framework code (LangGraph as the first target), and reconcile hosted agents with Terraform-style plan/apply/state workflows.
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 command-line tool and HCL-based language for AI agents. Provides parsing/validation of agent specs, a formatter, LangGraph code generation, and Terraform-style plan/apply/destroy with local state management and drift detection. Installable via Homebrew, an install script, or the Go toolchain.
Market Position
Positioned as 'Terraform for AI agents' - a declarative infrastructure-as-code layer and source-of-truth contract for agents, rather than an agent runtime or framework. It sits above frameworks like LangGraph (its first codegen target) and is intentionally not trying to describe the full control loop or become a runtime.
Leadership
Founders
Artem
Solo creator of Kastor, publishing under the GitHub handle weirdGuy and the Hacker News handle weirdguy. Building the project in Go. Frames Kastor as a bet that the agent runtime layer will keep changing while the need for reviewable, diffable agent contracts will not.
Founding Story
Kastor began as a solo, early proof-of-concept project in mid-2026. Rather than trying to standardize the agent runtime or control loop (which the creator argues is not yet stable), it aims to standardize the outer contract around agents: what inputs an agent accepts, what outputs it promises, which model and prompt/template it depends on, which tools it can call, and where it should be built or applied. The goal is to turn agent configuration, currently scattered across framework code, prompts, configs, and deployment notes, into versionable, reviewable, declarative files.
Target Markets
- Defining AI agents as a single reviewable, version-controlled source of truth
- Generating runnable LangGraph framework code from a declarative spec
- Managing and reconciling hosted agents with plan/apply/state workflows
- Treating agent configuration as diffable, code-reviewable contracts