Ryker / Docs

What is Ryker?

Ryker is an AI agent that does engineering and operations work for your team. You run it on your own server and talk to it in Slack, GitHub, or its browser chat.

Give it a job, such as fixing a bug or finding out why a deployment failed. Ryker reads your repositories, runs commands in a sandbox, writes and tests code, and opens a draft pull request. It can fix failed CI and respond to review comments in the same task. For problems in running systems, it uses Emisar to inspect them and carry out the actions you allow.

You can ask directly, let it respond to alerts already arriving in Slack, or give it work to do on a schedule. Ryker works through the job and asks when it needs a decision, access, or approval.

From a request to a result

Suppose a deployment starts failing after a registry password changes. You ask Ryker to find the cause and fix it. With access to the relevant repositories and systems, it can read the deployment logs, trace how the code loads credentials, and identify the stale password.

If the fix needs a code change, Ryker proposes a task. You press Start task, and it creates a separate working copy. It updates the code to reload changed credentials, adds a regression test, and runs the checks. A task card in the Slack thread updates as the work progresses.

When the change is ready, Ryker can open a draft pull request. If CI fails on its commit, it picks up the task again and fixes the failure. You review the diff, ask for changes, and merge it through your usual process. Ryker keeps the original thread up to date.

A fix to a running system goes through Emisar instead. Its policy determines which actions can run and which wait for a person to approve them. Ryker then checks the result against fresh evidence where the available tools support it.

Ryker, co:op, and Emisar

Ryker connects the conversation to the work. It gathers context, sends work to a model, follows progress, and returns the result. Two other tools provide the places where that work happens:

• co:op runs the model and its commands in a sandbox. It gives a code task its own working copy and limits the files and network destinations the model can reach.
• Emisar gives Ryker access to running systems through defined tools. It checks permissions, handles approvals, executes actions, and records their results.

The installer includes a co:op worker. You connect your own model account, Slack workspace, and GitHub repositories. Emisar is optional; add it when Ryker needs access to your infrastructure.

Code and team knowledge

Ryker can read several repositories together. For example, a task can change an API while reading the client that calls it. An environment groups the repositories and optional Emisar account available to a conversation.

It also learns from conversations and your feedback. It can recall a past investigation, a team decision, or a correction you gave it. You can inspect and remove what it remembers. For current service health, it checks current evidence instead of treating an old answer as proof.

Proactive and scheduled work

In a proactive Slack channel, Ryker can join a useful conversation or investigate an alert without waiting for a mention. It can also stay silent or acknowledge a message with a reaction. An observe-only channel lets it build knowledge without posting.

For repeat work, ask for a schedule such as a weekday health check, or a rule such as reviewing new Terraform plans. You confirm what it should do and where it should report. These requests use the same access settings as other work.

Running it for your team

Ryker stores its work history, settings, and memory in PostgreSQL on your host. Its console shows the context sent to the model, tool calls, results, checks, and cost for each request. You can change models, write team instructions, and investigate an answer that went wrong.

A hosted model still receives the context needed for its work. Your team chooses repository access, controls code merges and releases, and sets infrastructure permissions in Emisar. The access and data guide explains these boundaries.

To try it, install Ryker and ask a question about your code. If your team already runs it, start with a Slack conversation or follow a complete code change.