Run Locally in 30 Seconds

kdeps ships as a standalone binary. Install it, run it, and you have an interactive AI agent running on your machine. No Docker. No config file. No API key required if you use a local model.

Install

macOS (Homebrew):

brew install kdeps/tap/kdeps

macOS / Linux (curl):

curl -LsSf https://raw.githubusercontent.com/kdeps/kdeps/main/install.sh | sh

Go install:

go install github.com/kdeps/kdeps@latest

Verify:

kdeps --version

Start the agent

kdeps

That's it. kdeps opens an interactive REPL. By default it downloads and runs a local llamafile model (~1.1 GB on first run, cached in ~/.kdeps/models/). No API key needed.

kdeps v2.x.x
model: llama3.2:1b (local, file backend)
> _

Type anything and the agent responds. The model runs entirely on your machine.

Use a local model via Ollama

If you already have Ollama installed, point kdeps at it:

# pull a model with Ollama
ollama pull llama3.2

# start kdeps using that model
kdeps --model llama3.2 --backend ollama

Or try a reasoning model:

ollama pull deepseek-r1
kdeps --model deepseek-r1 --backend ollama

Your prompts never leave the machine.

Use a cloud model

Set your API key and pick a provider:

# Anthropic
export ANTHROPIC_API_KEY=sk-ant-...
kdeps --model claude-opus-4-5 --backend anthropic

# OpenAI
export OPENAI_API_KEY=sk-...
kdeps --model gpt-4o --backend openai

The --backend flag tells kdeps where to route the request. The --model flag picks the model. Both can also be set in ~/.kdeps/config.yaml to avoid repeating them on every invocation.

Load your own workflows as tools

Once you have a workflow directory, pass it as an argument to kdeps:

kdeps ./my-workflow/

The REPL starts with your workflow registered as a callable tool. The LLM decides when to invoke it. This is agent mode - the LLM drives, your workflows execute on demand.

What to do next

• Local Models (Llamafile & Ollama) - Go deeper on offline setup, model selection, and privacy
• Quickstart - Build your first workflow
• Agent Skills - Extend the agent with pre-built skill sets