Open source Apache-2.0, permanently

Describe an agent. Watch it tested. Let it run.

Rookery runs on your own machine, over your notes and the accounts you connect. Agents work while you're away — calling the services you use, and clicking through the pages that don't offer an API.

curl -fsSL https://rookery.cloud/install.sh | sh
Read it first — it's short: install.sh

What it's like

You describe it. It gets built and tested.

New agentdesigning
Workspaces

One machine. Sealed, separate worlds.

Every workspace is its own tenant — its own knowledge, its own credentials, its own connected accounts, its own agents. You enter one with its password, and nothing crosses between them.

Sandboxing is only supported on Linux.

Agents

Bill watcherWeekdays · 09:00
Reading digestSun · 18:00

Chats

Planning the trip
Health notes

Reminders

Call the dentist · 11:30
Renew passport · Fri
214 notes 6 connectedsealed
Knowledge base

Your knowledge grew hands.

Everything you know, everything your agents learn, and what your connected services bring in — as plain markdown on your own disk. Open it in Rookery or in any editor you like. Select a sentence and watch it improve.

notes / standup.md

Tuesday standup

Shipped the new connector settings screen. Reviewed two pull requests and cleared the backlog on the release checklist.

the deploy went fine but there was a issue with the db migration that we had to rollback and then re-run it manually

Tomorrow: finish the migration guard and pair on the onboarding copy.

select any text
Agents

Describe it. Don't configure it.

Say what you want in your own words. Rookery asks a couple of questions, proposes a plan, builds it, tests it against the real services, and shows you what happened before anything is saved. Then it just runs.

Agents4 running

Monthly energy bill

1st · 09:00

RUNNING

Morning briefing

Mon–Sun · 07:00

Inbox triage

Every 30 min

Weekly reading digest

Sun · 18:00

Skills

Things your agents already know how to do.

Built in and ready to attach to any agent — plus any you create the same conversational way.

PDF

Read and pull text out of documents

Spreadsheets

csv, xlsx

Web research

Search, read, and cite

Email triage

Sort what matters from what doesn't

Calendar

Scheduling and availability

Image OCR

Text out of pictures

Change detection

Notice what moved

Time & timezones

Get 3pm right, everywhere

Git & GitHub

Repos, issues, pull requests

Agent collaboration

One agent calling another

Tool installer

Fetch what it needs

Resilient runs

Fail softly, retry sensibly

Every agent gets the same tools, whatever model is behind it.

Web search & fetch Read & write your notes Run what it needs Every connected account

A small model on your own machine and a frontier one are given exactly the same reach. Tools remove the floor on what a model can do — never the ceiling on how well it does it.

Browser control

For the pages that never offered an API.

Agents open a real page, sign in, fill in forms, click through a flow and complete a payment — using credentials you have stored as secrets, which are typed straight into the page and never shown to the model.

Connections

136 services. No middleman holding your keys.

Rookery talks to them directly, using credentials you own.

A selection of the 136 supported. See all

MCP servers

Not on the list?
Add it yourself.

Point Rookery at any Model Context Protocol server by URL — including one on your own network — and its tools join your agents and chat.

How it works
playwright · 192.168.1.20:3000 7 tools
  • browser_navigate on
  • browser_snapshot read-only
  • browser_click approval
  • browser_file_upload off
↵ new tools arrive switched off
Chat

Ask what you know. Then have it act.

Talk to your knowledge the way you'd talk to someone who has read all of it — and have it write a note, or do something in a connected account, right there in the conversation.

What did I decide about the pricing tiers?

read notes/pricing.md · notes/meetings/2026-07.md

You settled on three tiers and agreed to keep self-hosting free forever. The open question was whether the middle tier includes priority support.

answered from your own notes

Notifications

You find out the moment it happens.

An agent finished, a service returned something new, a reminder came due. It lands in your inbox, and reaches you on Telegram, Discord or Slack when you're away.

InboxToday

Monthly energy bill

One bill unpaid for August, 2,480 MKD. Waiting for your approval.

09:00

Morning briefing

Three meetings today, and two replies you owe someone.

07:00

Reminder

Call the dentist.

11:30

Two invoices are past due. I've written the details into your notes under Finance.

09:14

and you can reply

/runrun an agent now

Models

Your machine. Your model.

Use the coder tool you already have, or connect the provider you prefer — hosted, or running entirely on your own hardware. Nothing ties you to one vendor.

Ollama · vLLM · LM Studio

Point Rookery at Ollama, vLLM, LM Studio, llama.cpp, LocalAI or Jan on your own machine or network. No key, no account, and nothing — knowledge, credentials or prompts — leaves hardware you control.

The current list of supported providers lives in the documentation — it grows between releases, so it isn't pinned here.

Secrets

Credentials that stay yours.

Encrypted where they sit, unlocked only into the thing that needs them, scoped to one workspace, and never handed to anyone else.

SecretsIN MEMORY

STRIPE_KEY

sk_live_EXAMPLE_NOT_REAL

GMAIL_TOKEN

ya29.EXAMPLE_NOT_REAL

Scheduling

Every weekday at eight. And again at ten.

Say when in your own words — twice a day, the first of the month, every twenty minutes during work hours, or only when you ask. Reminders work the same way: “remind me in 10 minutes to call the doctor.”

you say

every weekday at eight and again at ten

it runs

Mon–Fri · 08:00, 10:00

Models

Bring the model you already trust.

A coder tool you already use, a hosted provider, or a model running on your own hardware.

Coders

Claude Code
Codex
OpenCode
Cursor
Gemini CLI

Providers

OpenAI
Anthropic
Gemini
xAI
Mistral
DeepSeek
Moonshot
Z.AI
OpenRouter
Perplexity
Groq
Together
Fireworks
Cerebras
SambaNova
Nebius
DeepInfra
Hugging Face
GitHub Models
Bedrock
Alibaba
Ollama Cloud

Self-hosted

Ollama
LM Studio
llama.cpp
vLLM
LocalAI
Jan

Put it on a machine that never sleeps.

curl -fsSL https://rookery.cloud/install.sh | sh
Read it first — it's short: install.sh

Support the project

Rookery is open source and self-hostable, forever. If it's useful to you, a star genuinely helps other people find it.

Star on GitHub