Skip to main content
ZWRM Agents are AI workers (Claude Code, Codex) that run in isolated Firecracker microVMs. An agent is pure configuration — instructions, default model, effort, VM size, connectors, skills, and budgets. Execution state lives in workspaces: a persistent volume plus a session-continuity key. VMs are ephemeral: one boots when a run or chat starts and is torn down when it finishes; the workspace volume (files, repos, conversation transcript) is what carries everything forward. Agents have full access to the ZWRM platform — the zwrm CLI is pre-installed and authenticated inside every VM. Your agent can build, deploy, scale, and manage applications directly.

​
Quick start

On first use, the CLI prompts for required credentials (Claude/Codex auth, GitHub token, git config). They are stored as agent secrets and injected into every VM at boot; gh is authenticated automatically.

​
The model

There is no built-in repo concept: a repo is whatever the prompt (or a skill) clones onto the workspace volume. gh is pre-authenticated, so “clone acme/api and open a PR” just works — and the clone persists on the volume for the next run with the same session key.

​
Agent types

Agent names are unique per organization, across types: default can name a claude agent or a codex agent in an org, not both. Connecting with the wrong type for an existing name is rejected. Connecting to a name that doesn’t exist creates it.

​
Configuring an agent

Everything an agent is can be changed with a single command — only the flags you pass change:

​
What agents can do

From inside any agent VM, the agent (or you, over SSH) can:
  • Deploy apps — zwrm deploy to build and ship directly from the agent
  • Manage secrets — zwrm secrets set to configure environment variables
  • Create databases — zwrm postgres create to spin up Postgres instances
  • Scale services — zwrm scale to adjust replicas
  • View logs — zwrm logs -f to stream application output
  • SSH into apps — zwrm ssh to debug running VMs

​
Where to go next

Runs

Dispatch autonomous tasks and iterate on them.

Workspaces

Session continuity, files, and retention.

Messaging

How agents wait, ask, and get answered.

Interactive sessions

SSH sessions, dashboard chat, and managing agents.

OpenAI-compatible API

Chat with any agent from any OpenAI client.

Connectors

Attach MCP tools from the gateway — credentials never enter the VM.

Skills

Reusable instruction bundles synced to the workspace.

Memory

Durable knowledge that survives sessions and workspaces.

Capabilities

Scope what an agent may do on the platform.

Budgets

Cap spend and run counts per day.

Automations

Trigger runs from webhooks and schedules.