Open source · Apache-2.0

Agents that run like clockwork.

Orrery is the platform that keeps AI agents in motion — sessions that survive crashes, sandboxes for their hands, vaults for their keys, and a harness you can swap. We run it for you, or you run it yourself.

Anthropic-compatible API · SQLite, Postgres, or D1 · Managed or self-hosted

The console

The console runs itself. Or take the wheel.

A live sketch on autopilot — it mints keys, creates agents, and kills its own processes. Move your cursor in to take over.

Dashboard
Get started with Orrery

Install the CLI, mint a key, and let agents operate against your control plane.

Step 01

Install the CLI

The command-line client lets your agent (or you) drive the platform from the terminal.

npx -y -p @orrery/cli orrery

or globally: npm i -g @orrery/cli

Step 02

Mint an API key

Your agent needs this to authenticate. Keep it somewhere it can read.

Step 03

Give it the task

Point your agent at the CLI or API skill, then ask for what you want:

Use Orrery to create a research agent that monitors arXiv daily

Open to watch a live agent — then kill it.

Sessions live
  • scout
    Rotate webhook signing secret
    node-20 · subprocess · 41 min
  • triage
    Linear teammate — bug intake
    linear · webhook · 3 h
  • digest
    Morning support digest
    cron · 07:00 · 12 min
scout sess_01hxq4kj running
Agents
claude-sonnet-4-6
agent_toolset_20260401
scoutv3

Coding agent — repo chores, PRs, secret rotations.

claude-sonnet-4-62 sessions
triagev7

Linear teammate — reproduces, comments, tags fixes.

claude-sonnet-4-61 session
digestv2

Scheduled — reads support inbox, ships a morning brief.

claude-haiku-4-5cron 07:00

Frameworks give you the loop. Production asks for everything around it.

State that survives a crash. Secrets that never touch a sandbox. Somewhere for tools, skills, and memory to live. Orrery is that everything — the machine around the model.

How it works

One orbit of an agent

  1. 01

    Define the agent

    Model, system prompt, tools, skills, MCP servers. Versioned, like code.

  2. 02

    Open a session

    An append-only event log is born. Stream it live over SSE.

  3. 03

    Sandbox boots

    Subprocess, Firecracker, Daytona, E2B, or a Cloudflare Container.

  4. 04

    Work happens

    Tools run, vaults sign requests at the network layer, memory mounts as files.

  5. 05

    Something dies

    A process, a container, a region. It happens. That's the design assumption.

  6. 06

    It resumes

    State rebuilds from the event log — mid-thought, same place. No luck involved.

Features

Everything an agent needs to hold an orbit

Sessions & event log

Every turn is an append-only event. Kill the process mid-thought — it resumes from the log, not from luck.

Sandboxes, five flavors

Subprocess for speed, Firecracker for isolation, Daytona, E2B, or Cloudflare Containers — one interface.

Vaults

Credentials are injected at the network layer. env | grep TOKEN returns nothing — a prompt-injected agent has nothing to leak.

Model cards

Per-tenant LLM credentials, encrypted at rest. Anthropic, OpenAI, or any compatible endpoint — rotate without a redeploy.

Skills & memory

Instruction packs mounted into the sandbox, and memory as plain files the agent reads with its own hands.

A teammate, not a webhook

Publish agents into Linear, GitHub, and Slack under their own identities — assigned, mentioned, replied to.

For your agents

Two commands to a running agent

Deploy

Managed cloud, edge, or your box

Managed cloud

We run the control plane, sandboxes, and upgrades. You bring the agents.

Book a demo →

Cloudflare

Workers, Durable Objects, and Containers — in your own Cloudflare account.

wrangler deploy

Self-host

Your VPS or Kubernetes, SQLite or Postgres. One container and a URL.

railway up

Put your agents in orbit.

Built in the open, Apache-2.0 end to end. Read the code, swap the harness — or let us run the whole machine for you.