KDeps

Build and deploy AI agents in YAML.

Workflow pipelines and autonomous agents in YAML. Export as Docker, Kubernetes, ISO, or a single binary. Works with Ollama, OpenAI, Anthropic, and any OpenAI-compatible backend.

Get Started

Why kdeps?

Registry

apiVersion: kdeps.io/v1
kind: Workflow

metadata:
  name: summarizer
  version: "1.0.0"
  targetActionId: summarize

settings:
  apiServer:
    portNum: 16395
    routes:
      - path: /summarize
        methods: [POST]

$kdeps run workflow.yaml

Listening on :16395

$curl -s -X POST localhost:16395/summarize \

-H "Content-Type: application/json" \

-d '{"url": "https://example.com"}'

{"success": true, "data": {"response": "Example.com is used for illustrative examples in documentation."}}

Any backend

Ollama, OpenAI, Anthropic, Groq, or any OpenAI-compatible endpoint. Switch backends in config without touching workflow files.

Deploy anywhere

Export as a Docker image, bootable ISO, single binary, or Kubernetes manifests. The same workflow runs locally and in production.

Component registry

Install pre-built scraper, search, browser, and embedding components from the registry. Compose them into your workflow with one line.

How it works

Three steps from idea to running AI API.

Write YAML

Declare resources -- chat, HTTP, Python, SQL, shell. Wire them with requires:. No glue code.

→

Run kdeps

kdeps run workflow.yaml starts the API server. kdeps serve starts the autonomous agent loop.

→

Call the API

POST to your route, get back structured JSON. Export as Docker, Kubernetes, ISO, or a single binary.

Two modes

Pick the one that fits the task. Mix them in an agency.

workflow

Deterministic pipelines

Resources run in DAG order defined by requires:. Every request takes the same path. Predictable, testable, auditable.

POST /summarize

↓

fetchhttpClient

↓

summarizechat

↓

apiResponse

$ kdeps run workflow.yaml

agent

Autonomous LLM loop

The LLM decides which resources to call and in what order. Every resource auto-registers as a tool. Multi-step reasoning, no wiring required.

stdin prompt

↓

LLMplans steps

↓

http

sql

python

↓

LLMsynthesizes

↓

response

$ kdeps serve workflow.yaml