oss ▸ clone and run 0

Run the whole engine, free

Writ is open-core: bring up the recorder, API, MCP, and workflow engine on your own server with one Docker command. Unlimited local runs, your own AI keys, no cloud account needed.

View on GitHub Self-host docs

Writ runs on your own accounts, with your own credentials and data, on sites you are authorized to use.

bring-up.sh · bash

git clone https://github.com/usewrit/writ.git
cd writ
cp .env.example .env        # set your secrets + (optional) AI keys
docker compose up -d        # recorder UI, API, MCP, engine, Postgres, Redis

Same codebase as the cloud - recorder, engine, and /v1 + MCP surfaces on your own domain.

recorder UIcoordinatoragentPostgresRedisMinIOMCP

core ▸ open by design 01

open-core

Open-core, by design

The engine and single-tenant control plane are open source under a fair-code license - free to self-host, modify, and use commercially for your own automations. The multi-tenant network is the commercial layer.

Self-hosting is the same code that powers the cloud, with the commercial modules absent - not a crippled build. You operate, update, and secure it yourself; that is the honest trade for running it on your own infrastructure.

The open column is the engine you run yourself. The cloud column adds the managed network.
Capability	Open (self-host, free)	Cloud (managed)
Workflow engine	Yes	Yes
Recorder / builder UI	Yes	Yes
Local API + MCP	Yes	Yes
Bring-your-own AI keys	Yes	Yes
Local execution (free of compute)	Yes	Partial
Your own self-hosted agent fleet	Yes	Yes
Marketplace	No	Yes
Billing & $ wallet	No	Yes
Managed /v1 gateway at scale	No	Yes
Managed network egress	No	Yes
Fleet autoscaling	No	Yes
Support + SLA	No	Yes

What this required from your firewall: ~~no inbound rule~~ · ~~no public hostname~~ · ~~no port opened~~

Nothing dials in. The agent holds one outbound connection, and the call rides back down it. There is no inbound rule to write and no port to open.

Your own machine — free, no account, runs while it’s awake; only the result leaves.

The agent dials out. Nothing dials in — a system behind your VPN becomes callable without opening a single inbound port. Local callers can stay entirely local; add Uplink only when a remote caller needs to reach a workflow behind your firewall.

run ▸ unlimited, free, local 02

local execution

Unlimited local runs, no compute charge

Self-hosted runs carry no compute charge - you own the hardware. There is no cloud time meter and no per-run fee. Run as much as your machine can handle.

No compute meter No cloud time meter, no per-run charge. You own the hardware, so local execution is free.

Your network & data Everything stays on your infrastructure - your Postgres, your Redis, your MinIO, your network. Nothing is sent anywhere.

Full step library The same 30+ step types, recorder, monitors, and notification channels ship in the open-core engine.

ai ▸ your own keys 03

bring your own AI

Full AI, on your own keys

Point the engine at Anthropic, OpenAI, Ollama, or any OpenAI-compatible endpoint. The same AI that powers extraction, self-healing selectors, and agent runs works self-hosted - on keys you own.

AnthropicOpenAIOllamaOpenAI-compatible

Bring your own provider Anthropic, OpenAI, Ollama, or any OpenAI-compatible endpoint. Run a local model and nothing leaves the box.

BYO keys are unbilled Your keys, your provider bill. Writ never meters or marks up AI on the self-hosted build.

Keys stay local AI keys live in your config on your own machine. They are never sent to Writ or anywhere else.

The hosted "build it with AI" concierge is a Writ Desktop and managed-cloud feature. Self-hosted, you run full AI on your own keys - you wire up workflows in the recorder yourself.

fleet ▸ no cloud link 04

self-hosted fleet

Run your own always-on agent fleet

Self-host the fleet coordinator on your server and connect your own light agents. An always-on fleet you operate - with no cloud link required.

The coordinator runs on your infrastructure and dispatches work to the light agents you connect to it. Schedule workflows, run monitors, and keep automations on around the clock - entirely inside your own network, with no dependency on Writ's cloud.

Always-on schedules Run scheduled workflows and monitors around the clock - the coordinator keeps them going while you are away.

Scale on your hardware Add light agents as you need more throughput. You own the machines, so there is no per-agent fee.

Inside your network Coordinator and agents talk over your own network. No cloud link, no data leaving your infrastructure.

pick ▸ three ways to run 05

ways to run

Three ways to run Writ

Self-hosting is the server-side third way, next to the single-machine desktop app and the managed cloud. All three run the same engine - pick the one that fits.

Writ Desktop

One machine

The full engine as a local app on your Mac, Windows, or Linux machine. Free and account-optional.

Runs on a single computer
Local-first, data stays on device
BYO AI keys, unbilled
Includes the "build it with AI" concierge

See Writ Desktop

Self-host

Your server

The open-core engine on your own infrastructure, brought up with one Docker command. This page.

One Docker Compose command
Unlimited local runs, no compute charge
Your own always-on agent fleet
You operate, update, and secure it

View on GitHub

Managed cloud

We run it

The same engine plus the managed network - marketplace, managed egress, fleet scale, and support.

Always-on, nothing to operate
Fleet-scale managed network
Marketplace and managed egress
Managed support and an SLA

See pricing

Self-hosting trades a cloud bill for the engineering time to operate it - count both.

edge ▸ structurally cloud-only 06

cloud-only

What's structurally cloud-only

Some capabilities are network effects or shared infrastructure and can't be self-hosted. Self-hosting gives you the engine, your own fleet, and your own automations; the managed cloud adds the network.

The managed marketplace
Managed residential egress
Managed multi-tenant billing and the $ wallet
Fleet-scale managed network and autoscaling

ask ▸ answered 07

faq

Self-hosting questions

Is self-hosting free?

Yes. Self-hosted runs carry no compute charge - you own the hardware. There is no cloud time meter and no per-run fee. The engine and single-tenant control plane are open-core and free to run for your own automations. You do provide and operate the server yourself.

What license is the engine under?

The engine is open under a fair-code license: free to use, modify, and run internally and commercially for your own automations. You may not resell Writ itself as a competing hosted service.

Do I get full AI when self-hosting?

Yes - on your own keys. Point the engine at Anthropic, OpenAI, Ollama, or any OpenAI-compatible endpoint, and the same AI that powers extraction, self-healing selectors, and agent runs works self-hosted. BYO keys are never metered or billed. The hosted "build it with AI" concierge is a Writ Desktop and managed-cloud feature; self-hosted, you wire up workflows in the recorder yourself.

Can I run my own agent fleet?

Yes. Self-host the fleet coordinator on your server and connect your own light agents for an always-on fleet - with no cloud link required. Schedule workflows and run monitors around the clock, entirely inside your own network.

What is structurally cloud-only?

Some capabilities are network effects or shared infrastructure and cannot be self-hosted: the managed marketplace, managed residential egress, managed multi-tenant billing and the $ wallet, and the fleet-scale managed network. Self-hosting gives you the engine, your own fleet, and your own automations; the managed cloud adds the network.

Can I move to the cloud later?

Yes. Self-hosted and cloud run the same engine, step types, recorder, and /v1/{slug}/{path} and MCP surfaces. Point the same workflows at the managed cloud, connect a wallet, and pick up the marketplace, managed egress, and fleet scale - no rewrite.

Do my AI keys leave my machine?

No. Self-hosted, your AI keys stay in your config on your own infrastructure. Bring your own Anthropic, OpenAI, Ollama, or any OpenAI-compatible endpoint - the keys are never sent anywhere.

View on GitHub Self-host docs