Skip to content
Your AI on a USB stick

One YAML file, offline

kdeps turns one workflow.yaml into an offline API, a binary, an image, or an ISO. You review the YAML as a pull request, and kdeps test runs the evals. A check returns 400 before any model call. A chat with loop: is a bounded step in that file.

actionId: llmname: Ask the AIvalidations:  check:    - get('q') != ''   # 400 before any model call  error:    code: 400    message: "'q' is required"chat:  model:    name: llama3.2:1b   # local, no API key  context:    prompt: "{{ get('q') }}"  loop:   # a bounded agent step    tools: [read_file]    maxRounds: 5
$export KDEPS_API_AUTH_TOKEN=dev-token
$kdeps test .
Running 1 eval(s) for chat-api
 
PASS needs-q (400, 1ms)
 
1 passed, 0 failed
$kdeps run .
Starting HTTP server on 0.0.0.0:16395
$

runs on

OllamallamafileOpenAIAnthropicGoogleOpenRouterGroqMistralDeepSeekxAICohereTogetherPerplexityHugging FaceCloudflareBedrockwatsonx

inside kdeps

What one YAML file gives you

Each box is something you get without writing glue code.

One file, any input

kdeps runs it. A request comes from an API, a bot, a file, or an interactive prompt.

A bounded loop

A chat with loop: calls only the tools you name. maxRounds stops that step.

The model in the binary

Pass --include-models to kdeps bundle prepackage. The llamafile stays in the executable, and the machine needs no network.

Evals on the pull request

kdeps test runs the evals: in the process. The same judges review the answer before the resource finishes.

Runs offline

llamafile runs the model on your machine. No API key, no data leaves it.

Ship it anywhere

Ship the same file as a Docker image, Kubernetes manifests, a bootable ISO, or one binary. Write the ISO to a USB stick.

git-native

The YAML is the behavior spec

The YAML in your repository fully defines what the appliance does: the model, the validation, the step order, and the response shape. When you change the YAML and commit it, the behavior changes. Your git history is the changelog of the agent behavior. Only the model binding is outside the repository. As a result, the same commit runs with a local model on a laptop and with a cloud model in production.

git show HEAD - resources/llm.yaml
validations:
  check:
    - get('q') != ''
+    - len(get('q')) < 2000      # reject prompts over 2000 characters
chat:
  model:
-    name: llama3.2:1b
+    name: qwen2.5:7b
  context:
    prompt: "{{ get('q') }}"

One commit makes two behavior changes: a larger model, and a 400 error for input that is too large. You do not change a deploy configuration or a console setting. The diff is the change.

Reviewable

A change to an agent is a diff: a new resource, a stricter validation, or a different response shape. Review it in a pull request, as you review all other code.

Versioned

A git tag is a release. kdeps test runs the evals on that tag. The built appliance also pins the runtime and the model.

Reproducible

In CI, check out a tag and run kdeps test. Then run kdeps validate and kdeps bundle build. The same commit always makes the same appliance.

Shareable (optional)

With the registry, you can install and publish agents by name. kdeps install owner/repo pulls from a GitHub repository. You do not need the registry to build or deploy your own agent.

getting started

How it works

Write YAML. Run it. Ship the same file.

01

Write YAML

Put the steps in one workflow.yaml. A check can return 400 before the model runs. A chat with loop: is an agent, and maxRounds bounds it.

→
02

Run it

kdeps runs ./workflow.yaml as an API, a bot, a file processor, or a prompt. kdeps test runs the evals: in the process.

→
03

Ship it

Ship that file as a Docker image, Kubernetes manifests, a bootable ISO, or one binary. Pass --include-models to kdeps bundle prepackage. That flag puts the llamafile in the binary.

compared to manual code

YAML replaces glue code

You do not write Python scripts, connection code, or boilerplate.

Traditional approachkdeps
Python script + Flask + OpenAI SDK + retry logicOne workflow.yaml file
Dependency code that you write manuallyflow.requires: in YAML
Error handling that you write manuallyonError: block (continue / retry / fail)
A Dockerfile and a CI pipeline for deploymentkdeps bundle build --tag then docker push
Agent chains that you write manuallyagent: resource. One agent calls a different agent declaratively
A test server that you writekdeps test runs evals: in the process
Handler code that checks the body after the model callvalidations.check returns 400 before any model call

kdeps has the Apache 2.0 License.