Skip to content
Krypton

Krypton Runtime

Krypton Runtime is a Kubernetes-native runtime for AI agents, self-hosted LLMs, and MCP servers.

For agents, it turns A2A, plain HTTP, and framework-backed containers into cluster resources with stable gateway routing, lifecycle management, scaling signals, and observability.

For model serving, a Model custom resource becomes a llama.cpp deployment: Krypton pulls GGUF weights from Hugging Face and exposes the result through OpenAI-compatible API paths (/v1/models, /v1/chat/completions, /v1/completions, /v1/embeddings). Your applications keep using familiar SDKs while operators manage models as ordinary Kubernetes resources.

For MCP, Krypton runs HTTP-transport servers directly, or bridges stdio servers into the same agent gateway and UI introspection path.

Fast path

Section titled “Fast path”
Install KryptonHelm install, health check, and a no-secrets agent in two commands.

What it does

Section titled “What it does”
Capability What you get
Agents Run A2A, MCP, or plain HTTP containers with an Agent CRD. Bring LangGraph, Google ADK, custom services, or your own image.
Self-hosted LLMs Run Hugging Face GGUF models with llama.cpp using a Model CRD and OpenAI-compatible /v1/models and /v1/chat/completions routes.
MCP servers Host native HTTP MCP servers, or wrap stdio MCP binaries with the bundled mcp-stdio-bridge.
Gateway routing One public gateway handles agent invocation, model routing, streaming responses, and protocol-specific paths.
Operator UI A lightweight control-plane UI lists agents, models, status, and MCP tools.
Scaling signals Per-pod sidecars enforce concurrency, expose in-flight counts, and feed the scaler.
Observability Prometheus metrics for gateway traffic, model calls, sidecar load, scaler decisions, and control-plane health.
Helm-first install Install the runtime as a single OCI Helm chart and keep ingress, TLS, auth, and rate limiting in your existing platform layer.

Quickstart

Section titled “Quickstart”

Install Krypton:

Terminal window
helm install krypton oci://ghcr.io/kryptonhq/charts/krypton \
  --namespace krypton-system \
  --create-namespace

Deploy the no-secrets helloworld agent:

Terminal window
kubectl apply -f https://raw.githubusercontent.com/kryptonhq/runtime/main/examples/agent/python/helloworld/agent.yaml
kubectl -n krypton-system port-forward svc/krypton-gateway 8080:8080 &

Invoke it through the gateway:

Terminal window
curl -X POST http://localhost:8080/v1/agents/agents/helloworld/ \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":"1","method":"message/send",
       "params":{"message":{"messageId":"m1","role":"user",
       "parts":[{"kind":"text","text":"ping"}]}}}'

Open the operator UI:

Terminal window
kubectl -n krypton-system port-forward svc/krypton-control-plane 8090:8090
open http://localhost:8090/ui/

Try an LLM

Section titled “Try an LLM”
Terminal window
kubectl create ns models --dry-run=client -o yaml | kubectl apply -f -
kubectl apply -f https://raw.githubusercontent.com/kryptonhq/runtime/main/config/samples/llm/qwen2.5-0.5b.yaml


curl -s http://localhost:8080/v1/models | jq

Any OpenAI SDK can use http://localhost:8080/v1 as its base_url.

Source

Section titled “Source”

Krypton Runtime is Apache-2.0 and developed at github.com/kryptonhq/runtime.