Deploy and Host Tasuku on Railway

Tasuku is a self-hosted control plane for agentic engineering. It turns requests from GitHub, Slack, and Linear into durable workflows for research, planning, implementation, and pull-request review while keeping credentials, policy, state, and publication outside agent sandboxes.

About Hosting Tasuku

This template deploys the combined Tasuku server and worker, PostgreSQL 18 with persistent storage, and a private Railway object-storage bucket for generated HTML artifacts. PostgreSQL, object storage, the public HTTPS origin, the master encryption key, and the bootstrap token are wired or generated automatically. Tasuku runs migrations before each deploy and serves its dashboard on the Railway-provided domain.

After deployment, open the dashboard, create the initial organization, register and install the GitHub App, then configure a remote sandbox provider and agent credentials. Daytona is the recommended sandbox provider for Railway because the platform service does not expose a host Docker socket.

Common Use Cases

Turn GitHub issues and pull requests into durable planning, implementation, and review workflows.
Coordinate coding-agent work from Slack and Linear.
Keep credentials, policy, workflow state, and publication outside ephemeral agent workspaces.

Dependencies for Tasuku Hosting

A GitHub account or organization for the generated GitHub App.
A Daytona API key and agent-authenticated runtime image, or another compatible remote sandbox provider.

Implementation Details

The Tasuku service runs tasuku all on port 8080. Railway runs tasuku-migrate before deployment and checks /readyz. The template generates a unique base64 master key and bootstrap token for every deploy and automatically references the bundled PostgreSQL and artifact-storage credentials. Preserve the master key with database backups.

Why Deploy Tasuku on Railway?

Railway provisions the application, PostgreSQL database, durable storage, HTTPS networking, generated secrets, and private object storage from one template. This keeps the control plane easy to operate while leaving agent execution in a separately configured sandbox provider.