Show HN: ToolKami CLI – Simple Agents Made Easy (Extending Shopify Tobi's Try) (github.com)

ToolKami is an open-source CLI framework for building and running lightweight AI agents with a focus on simplicity, composability and safety. It provides a small set of shell-integrated commands (toolkami init, cd, worktree, sandbox, merge, drop) and curated “tools” so you can declaratively define agent workflows in YAML, spin up isolated worktrees and Docker sandboxes for experimentation, and merge safe changes back into a repo. Installation is a one-line fetch of a Ruby script that injects a shell function, and the project supports running distributed MCP servers as single-file binaries via UV for easy deployment. Technically, ToolKami emphasizes mixing deterministic tools (file/shell tools, linters, greps) with dynamic LLM agents to reduce flakiness and make behavior reproducible. Example YAML workflows show model selection (e.g., gpt-4.1), tool lists (Kami::Tools::Grep, EditFile) and step bindings like git_diff or running rubocop. It also targets richer modalities: agents can query up-to-date docs, autonomously debug client-side rendering/JS errors, and feed outputs to vision-language models for structured checks. Features that tightly interleave non-deterministic AI behavior with normal code execution are in private beta. For practitioners this means faster, safer agent prototyping, reproducible multi-agent workflows, and straightforward local sandboxing and deployment.