Quickstart

Get OpenAgentic running — Docker Compose on one host, or Helm on a Kubernetes cluster.

OpenAgentic is a self-hosted agentic platform for IT operations. You run it on your own infrastructure, against your own models and clouds. It’s local-first: the recommended happy path runs entirely on Ollama — free, offline, no API keys. There are two ways to install it — Docker Compose on a single host, or Helm on a Kubernetes cluster. Both pull pre-built images; nothing is compiled on your machine.

The happy path: local-first on Ollama

The recommended way to get started is to run OpenAgentic against a local Ollama — no cloud account, no API keys, nothing leaves your machine.

# 1. Install Ollama — see https://ollama.com
# 2. Pull the default embedding model + a tool-capable chat model
ollama pull nomic-embed-text
ollama pull qwen2.5:7b

# 3. Install OpenAgentic — the zero-config path probes your local Ollama,
#    picks up the chat model, generates credentials, and brings the stack up.
curl -sSL https://install.openagentics.io | bash -s -- --quick

That’s the whole story: install Ollama, pull two open-weight models, run one command. nomic-embed-text powers the semantic tool router and qwen2.5:7b is the chat model — both run locally, for free. Cloud providers are an optional alternative (see below and Providers & models).

Chat model must be tool-capable. OpenAgentic always attaches tools, so a model with no tool support is rejected at boot (e.g. plain gemma3:1b → 400 — does not support tools). On Ollama, use a tool-capable family such as qwen2.5, llama3.1, mistral, or gpt-oss — or any cloud model. Capability is auto-probed on startup.

Docker Compose — one host

Want the guided install (or a cloud provider instead of Ollama)? One command checks your tooling, fetches a pre-built compose bundle, runs an interactive setup wizard, brings the stack up, and opens the UI.

curl -sSL https://install.openagentics.io | bash

Requirements


Docker + Compose v2	The `docker compose` v2 plugin (not the legacy `docker-compose`), daemon running.
Node.js 20+	Runs the setup wizard. The `--quick` and `--env` paths don’t need it.
Resources	~16 GB RAM and 30 GB free disk to run chat locally on Ollama (≈8 GB RAM if you offload chat to a cloud model — Ollama then only loads the embedding model).
A model provider	Recommended: a local Ollama with a tool-capable chat model + `nomic-embed-text` — free, offline, no keys. Optional alternative: a cloud provider (Vertex AI, OpenAI, Azure OpenAI, AWS Bedrock, Anthropic) and its credentials.

What the installer does: verifies Docker / Compose v2 / Node 20+ → downloads the pre-built compose bundle into ~/.openagentic → runs the wizard (deploy target, admin user, provider, MCP servers, credentials) → docker compose up -d, pulling images from ghcr.io/agentic-work/* → opens http://localhost:8080. First boot runs schema setup, vector-store init, and tool indexing, so allow a few minutes.

Zero-config path: curl -sSL https://install.openagentics.io | bash -s -- --quick — probes the local Ollama at localhost:11434, wires it as the provider, auto-detects a tool-capable chat model already on the host (pulling qwen2.5:7b if it finds none) and the nomic-embed-text embedding model, generates credentials, and brings the stack up. The --quick path requires a local Ollama; if none is detected it stops and points you at --wizard (for a cloud-LLM setup) or --ollama <URL>. It never installs Ollama for you.

Milvus is optional. The default stack is pgvector-only — a bare docker compose up -d boots healthy with no Milvus, and MCP tool and RAG embeddings live in Postgres pgvector. Add the Milvus vector store only for large embedding or RAG workloads: pass --milvus to the installer, or run docker compose --profile milvus up -d with MILVUS_ENABLED=true and SKIP_TOOL_SEMANTIC_CACHE=false in .env.

Full walkthrough: Docker Compose.

Install options

The Docker installer takes a few flags. The default with no flags launches the interactive wizard.

Flag	What it does
`--wizard`	Launch the interactive Ink setup wizard (the default). Requires Node.js 20+.
`--quick`	Non-interactive quick install. Requires a local Ollama; auto-detects a tool-capable chat model (pulls `qwen2.5:7b` if none) and wires Ollama as the provider. Stops if no Ollama is detected.
`--helm`	Install on Kubernetes with Helm instead of Docker Compose.
`--update`	Update an existing install in place — Compose rebuild, or `helm upgrade` with `--helm`.
`--doctor`	Run diagnostics only (Docker, Compose, Node, helm/kubectl, disk, ports) and report — fixes nothing.
`--down`	Tear the stack down and remove its containers, network, and volumes.
`--env <file>`	Skip all prompts and bring the stack up from a prepared `.env`.
`--no-open`	Don’t open the browser when the install finishes.
`--ollama <URL>`	Point at an Ollama you already run (default: `localhost:11434`).
`--vertex`	Use Google Vertex AI via ADC (`gcloud auth application-default login`).
`--vertex-key <key>`	Use Vertex AI via an API key.
`--gcp-project <id>`	GCP project for Vertex.
`--gcp-location <loc>`	Vertex location (default `us-central1`).
`--milvus`	Opt into the Milvus vector store. The default is pgvector-only.

For example, --quick --vertex brings the stack up non-interactively on Vertex AI via ADC, and --down tears it back down (containers, network, and volumes).

Kubernetes — any cluster

Install the chart straight from its hosted URL — nothing to clone.

helm upgrade --install openagentic \
  https://install.openagentics.io/openagentic-1.0.0.tgz \
  --namespace openagentic --create-namespace \
  --wait --timeout 10m

Requirements


helm 3 + kubectl	A current kube-context pointed at a reachable cluster (GKE, EKS, AKS, k3s, or local).
Cluster resources	~6 vCPU and 16 GB allocatable across nodes; ~45 GB of persistent volume (postgres · milvus · etcd · minio · ollama).
A model provider	Same as above — a local/in-cluster Ollama (the chart bundles one) or, optionally, a cloud provider. Configure it through Helm values.
Ingress TLS (optional)	A pre-existing `kubernetes.io/tls` secret in the namespace if you expose it over HTTPS; the chart does not integrate cert-manager.

The chart is self-contained — Postgres, Redis, Milvus, Ollama all deploy in-chart, no subchart dependencies. Pass your settings with -f values.yaml (provider, credentials, which MCP servers to enable).

Full walkthrough: Kubernetes.

Verify it’s up

# Docker Compose
docker compose ps
curl -fsS http://localhost:8080/api/health

# Kubernetes
kubectl -n openagentic get pods
kubectl -n openagentic rollout status deploy/api

First steps

Sign in as the admin account from the wizard (local auth seeds [email protected]).
Chat — talk to your stack in plain language on the provider you configured. The model picks from the connected tools and runs them, behind an approval gate on any mutating action.
Flows — build a no-code agentic workflow in the visual builder.
MCP servers — manage the bundled servers from the admin console; most are enabled by default.

OpenAgentic is not FedRAMP authorized.

Managed option

Prefer not to run it yourself? AgenticWork is the managed deployment — SSO, audit logs, a managed model fleet, and the enterprise Fleet/Mission orchestration layer — installed into your environment and operated with you.