Quickstart

Prerequisites

bash

# Install via script (Mac/Linux)
curl -LsSf https://raw.githubusercontent.com/kdeps/kdeps/main/install.sh | sh

Create a project

bash

kdeps new my-agent
cd my-agent

Or manually create the structure:

bash

mkdir my-agent && cd my-agent && mkdir resources

Define your workflow

workflow.yaml:

yaml

apiVersion: kdeps.io/v1
kind: Workflow
metadata:
  name: my-agent
  version: "1.0.0"
  targetActionId: responseResource
settings:
  apiServerMode: true
  apiServer:
    hostIp: "127.0.0.1"
    portNum: 16395
    routes:
      - path: /api/v1/chat
        methods: [POST]

Add an LLM resource

resources/llm.yaml:

yaml

apiVersion: kdeps.io/v1
kind: Resource
metadata:
  actionId: llmResource
run:
  validations:
    methods: [POST]
    routes: [/api/v1/chat]
    check:
      - get('q') != ''
    error:
      code: 400
      message: Query parameter 'q' is required
  chat:
    model: llama3.2:1b
    role: user
    prompt: "{{ get('q') }}"
    timeout: 60s

Add a response resource

resources/response.yaml:

yaml

apiVersion: kdeps.io/v1
kind: Resource
metadata:
  actionId: responseResource
  requires: [llmResource]
run:
  apiResponse:
    success: true
    response:
      data: get('llmResource')

Run it

bash

kdeps run workflow.yaml

Test the API:

bash

curl -X POST http://localhost:16395/api/v1/chat \
  -H "Content-Type: application/json" \
  -d '{"q": "What is AI?"}'

How it works

Request arrives at /api/v1/chat with {"q": "..."}
Validation checks q is not empty
LLM resource sends the prompt to the model
Response resource formats the output
API responds with the result

Quickstart ​

Prerequisites ​

Create a project ​

Define your workflow ​

Add an LLM resource ​