Self-host platform
Kubernetes-native AI agent orchestration platform
helm upgrade --install agyn-platform oci://ghcr.io/agynio/charts/agyn-platform --namespace platform --create-namespace -f values-platform.yaml- Stars
- 242
- CPU
- 4 CPUs
- RAM
- 8 GiB
- Storage
- Additional space for platform image and data
- Port
- 2496 (Local platform HTTPS endpoint (agyn.dev subdomains))
- Requires
- Cert-manager, NATS, OpenZiti Controller + Router, Redis
ghcr.io/agynio/agent-runtime
What is platform?
Agyn lets you run AI agents on your own Kubernetes cluster. The platform spins up agents only when needed and shuts them down when idle, keeping costs low. You define agents, their tools, and secrets as Terraform code that can be version-controlled and reviewed. Each agent gets its own isolated container, and every tool runs in a separate sidecar so credentials never reach the model. The system uses zero-trust networking with per-agent identities, enforces spend limits, and provides full audit trails of every LLM call and tool execution. For local development, a single command starts the entire stack in a VM. For production, one Helm chart deploys everything to your cluster. The project is fully open source under AGPL-3.0.
Should you self-host platform?
platform scores 48/100 on selfhost.directory (how scoring works) and ranks #110 of 209 in AI Agents. It has 242 GitHub stars.
There was no stable release in the past 12 months, so check the repository's recent activity before you rely on it.
Setup is advanced: plan for several services and some ongoing maintenance. An official Docker image is available. It is AGPL-3.0-licensed and written mostly in Typescript.
Its closest alternative on selfhost.directory is Orca (91/100, 108 642 stars). It scores higher overall, so compare the two before you pick. See platform vs Orca side by side.
Features
- Agent sandboxing
- Any agent container support
- Chat and sandbox UIs
- Declarative Terraform configuration
- Egress secret injection
- External secret providers
- MCP server isolation
- Multi-Model Support
- Observability and tracing
- Per-agent spend caps
- RBAC and SSO
- Serverless agent execution
- Telegram and Slack connectors
- Workload isolation with Kata/FirecrackerOptional, recommended for production
- Zero Trust Networking
platform README
FAQ
How do I run an agent on GPU?
Register a runner with GPU capability and set the agent to require GPU; the orchestrator schedules only on GPU runners.
Does Agyn support every MCP server?
Yes, the platform speaks standard MCP with both Streamable HTTP and stdio transports supported.
Can an agent talk to another agent?
Yes, multiple agents can be added as participants in a conversation, each with their own context.
Can a user belong to multiple organizations?
Yes, users can switch contexts in the Console and Chat to choose which organization they operate in.
Can the platform run without internet?
Yes, with a self-hosted LLM endpoint, mirrored container images, internal OIDC IdP, and internal Postgres, Redis, S3, OpenZiti, and OpenFGA.
Which models can my agents use?
Any model reachable via OpenAI Responses API or Anthropic Messages API compatible endpoints, including OpenAI, Azure OpenAI, Anthropic, and self-hosted gateways like vLLM or TGI.
Is there a hosted version?
Not yet. Self-hosting is the only path today; a hosted version may launch in the future.
Is Agyn open source?
Yes, the platform services, runtime, and tooling are licensed under AGPL-3.0.
More in Kubernetes
PortainerSelf-hosted open-source console for managing container estates: Docker hosts, Swarm clusters, Podman, Kubernetes and Azure ACI, all from one server container.
production-stackSelf-hosted, open-source Helm stack that scales vLLM across a Kubernetes cluster, with KV-cache-aware request routing and a Prometheus/Grafana monitoring dashboard.
prometheusPrometheus is a self-hosted, open-source monitoring toolkit that pulls metrics via HTTP, stores them locally, and uses PromQL for powerful queries and alerting.
Proxmox Kubernetes EngineSelf-hosted tool to deploy HA Kubernetes clusters on Proxmox VE using Cluster API with native storage, load balancing, and CNI.
rackOpen-source, self-hosted PaaS that runs your apps on a Kubernetes cluster inside your cloud account, adding autoscaling, instant rollbacks, and a Heroku-like CLI.
RadarSelf-hosted open-source Kubernetes UI running as a single binary. Provides topology views, GitOps integration, audit checks, and MCP server for AI agent support.
RancherSelf-hosted, open-source platform to provision, manage and secure many Kubernetes clusters from one plane, with centralized auth, RBAC and Docker or Helm installs.
rocketplaneIOSelf-hosted AI SRE for Kubernetes: eBPF zero-instrumentation traces, logs, metrics, live service map, plus MCP for AI agents to safely remediate.
skyfloDeprecated open-source AI agent that runs in your Kubernetes cluster to automate infrastructure via a deterministic approval-gated control loop.
sparklabxSelf-hosted Apache Spark notebook platform with per-user PySpark/Scala kernels, MinIO IAM storage isolation, OAuth, and Helm deployment. Open-source core, Apache-2.0.
TraefikTraefik is a self-hosted, open-source reverse proxy that auto-discovers services from Docker, Kubernetes, and other orchestrators to route traffic without manual config.
tsuruSelf-hosted, open-source PaaS that runs apps on your own Kubernetes clusters with one-command deploys, autoscaling, teams, quotas and add-on services.
0 comments