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

2 hours ago 1

ToolKami is an open sourced "simple" framework with conceptually clear, independant parts that allows you to build and work seamlessly with AI agents. It comes with a Command Line Interface and curated Tools.

Command Line Interface (CLI)

toolkami CLI is a modified version of 's CEO Tobias try implementation. It extends the implementation with sandboxing capabilities and designed with functional core, imperative shell in mind.

Commands:

toolkami init [PATH]: Generate shell function
toolkami cd [QUERY]: Interactive selector
toolkami worktree: Create worktree from current repo
toolkami sandbox: Run Docker sandbox from .toolkami/docker-compose.yml
toolkami merge: worktree changes back to parent repo
toolkami drop: Delete worktree and branch

It is designed to support multiple, concurrent agent workflows:

curl -sL https://raw.githubusercontent.com/aperoc/toolkami/refs/heads/main/toolkami.rb > ~/.local/toolkami.rb # Make "try" executable so it can be run directly chmod +x ~/.local/toolkami.rb # Add to your shell (bash/zsh) echo 'eval "$(ruby ~/.local/toolkami.rb init)"' >> ~/.zshrc

ToolKami's framework let deterministic tools and dynamic agents to work together seamlessly. It is designed on the premise of simplicity, composability and extensibility that scales nicely with LLM's increasing capability.

All the MCP servers can be distributed as a single file binary, thanks to UV script.

I have elaborated on default File and Shell tool in this blog post, along with what can be improved.

Browse tool allows agent to access up-to-date documentations, autonomously evaluate client-side rendering and Javascript errors when development web application.

You can even feed it to a Vision Language Model (VLM) to get a structured output if a sound or animation is played back correctly.

# Install UV ## OSX/Linux curl -LsSf https://astral.sh/uv/install.sh | sh ## Windows powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex" # Start the MCP server `./servers/__main__.py`

Google's AlphaEvolve: ToolKami style

A minimal implementation of AlphaEvolve using this framework with detailed writeup and code.

(Credits to https://deepmind.google/discover/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/)

^^^ Interleave non-deterministic AI behaviour with normal code execution. First, define your AI workflow in a declarative way in YAML.

name: create_tests model: gpt-4.1 tools: - Kami::Tools::Grep - Kami::Tools::EditFile steps: - git_diff: $(git diff) - Generate a list of tests to write based on the changes. - run_linter: $(rubocop)

^^^ Then use node-based editor for non-technical members to customize and test prompts.