toryo
toryo is the layer above your coding agent. It runs agent work as durable, gated jobs (coding tasks, codebase reviews, research, and scheduled maintenance) against the repositories you already have, backed by a knowledge plane your agents read from and write back to.
Repositories, worktrees, orchestration, and the default knowledge plane run on infrastructure you operate. toryo does not host a source copy or proxy model traffic. The selected harness sends task context directly to your chosen model provider under your credentials and your agreement with that provider.
If you are ready to install, Getting started takes you from the right platform installer through provisioning, activation, project registration, and a guided or command-line first run. If you are still deciding what to aim it at, When to use toryo is the page that saves you a bad first week.
There are two ways to use it, and they are worth telling apart before you start.
1. As a power tool#
Make your own work better, today. You drive toryo from the places you already work: your agent session, on whichever harness you use, or the board console.
Hand it a task and it runs as a durable job instead of a chat window you have to sit with. Run a full sequence and it plans, waits for you to approve the plan, implements in an isolated worktree, reviews itself, opens a PR, and waits again before merging. Fire the same work on a schedule and it runs while the local machine is awake. Knowledge explicitly recorded by those runs lands in a knowledge plane the next run can read from.
The gates are the point. toryo runs the parts that do not need you and stops at the parts that do, and where those stops fall is yours to set.
Nothing here needs code. Install toryo, register a project, and drive it with
/toryo-* commands or board.
Setup offers three shipped, editable YAML catalog entries: code-task,
review-task, and research-task. If your process differs, you can write your
own down instead: a sequence is a YAML file
in your own repository, and yours replaces ours by name. That is a capability
rather than a step you owe us. See
Writing your own sequences when you want it.
Start with Getting started.
2. As the engine inside your own tool#
Build something toryo does not do. Every app exposes a command-line interface as its integration contract, so your program can queue agent jobs, run sequences, and read and write the knowledge plane without being a toryo app or even being written in TypeScript.
This is how you build a purpose-specific tool, with its own interface and its own data, and let toryo be the part underneath that runs agents and remembers things.
See Building on toryo.
The pieces#
Whichever mode you are in, the same parts are underneath.
Orchestration, how work runs. Foreman keeps the services up; dispatch is the job queue that runs agent sessions; sequence drives multi-step work with your approval gates; scheduler fires work on a timer.
Knowledge, what your agents know. Brain is the entry point over memory, the library, the code index, and directives. Ideas captures work that isn't specified yet, and the session debugger finds waste worth writing a rule about.
System: projects, messaging, sequence authoring, usage, and settings.
Agents, what actually does the work. Harnesses covers the four toryo can drive, what each can and cannot do, and how to choose.
Reference: the skills, which are the /-commands you
drive from inside an agent session, and the full
CLI & contract reference. Both are generated from the code, so
they cannot drift.
Getting set up#
- Getting started: install toryo, provision it, activate, and run your first task.
- When to use toryo: the shape of work it fits, good first tasks, and what to aim it at only once you trust it.
- Licensing & activation: the 14-day trial, activating a team seat, what a seat covers, and how to get unblocked.
- Board: the desktop console, what the one download contains, the two gates it opens on first launch, and how updating works.
When something is wrong#
- Troubleshooting: what to do when toryo will not start,
a job never runs, or search stops finding things. Start with
toryo doctor. - Known limitations: what does not work yet, stated plainly, so nothing is a surprise on day three.
- Feedback: where to send a bug, and what to attach so the first reply is useful.
Release notes for every published version are on toryo.ai/changelog.