This guide takes you through Tejus's basic loop: open a repository, give one coding agent a concrete task, supervise the work, and review the resulting diff.

Before you begin

Install Tejus on Windows. You need at least one agent: the built-in Tejus agent (sign in with Google for Tejus Cloud, or add your own API key), or a supported coding-agent CLI that is already installed and signed in.

First-run setup

A fresh installation without projects opens a welcome flow: a feature tour, your Tejus account, agent selection, appearance, your first project, and a summary.

Your Tejus account is optional. Sign in with Google for free Tejus Cloud models, and choose whether to share anonymous usage analytics. Your own API keys and CLI agents work without an account.
Choose your agents lists the agents detected on this machine as Connected, Needs sign-in, Not installed, or Disabled. Uncheck any you don't want Tejus to use. Agents whose sign-in Tejus cannot check are shown as connected once their executable is found; a real turn confirms account access.

Existing installations with projects do not repeat first-run setup.

Add a Git project

Open Tejus and add a local repository (Ctrl+Shift+O).

Start with a repository whose current changes are already committed or intentionally preserved. A clean starting state makes the agent's work much easier to review.

Create one task

Press Ctrl+N or use the new-task control.

Tejus remembers each project's most recently chosen Local or Worktree mode for later chats. Review that choice when creating a task. For this first task, use the local checkout and run only one agent against the repository. Use a Git worktree when you begin running multiple tasks or want stronger isolation.

Choose an agent and model

Select an available agent, model, and effort or reasoning option.

New chats start with the built-in Tejus agent. Its Tejus Cloud models need a signed-in, activated account; models from your own API keys work without one. See Tejus Cloud. Coding-agent CLIs such as Claude Code or Codex use the runtime and account configured on your machine.

Give the agent a verifiable objective

Describe:

The outcome you want
The files or area involved
Important constraints
The checks that should pass

For example:

Add an empty state to the pull-request list.

Reuse the existing shared panel components.
Do not redesign the surrounding page.
Run the focused browser test and report the result.

A bounded objective is easier to execute, review, and undo than "improve the pull-request page."

Supervise the turn

Follow the transcript and tool activity while the agent works.

Useful controls:

Ctrl+J opens the terminal drawer.
Ctrl+D opens the diff view.
Ctrl+Shift+B opens the browser.
The IDE button in the window bar opens the chat in the Editor view.
Approval and user-input requests appear in the task.
Send a follow-up when the agent needs a correction or additional constraint.
Interrupt the turn when it is clearly heading in the wrong direction.

Do not wait passively for a final message if the intermediate work is already incorrect.

Verify the result yourself

When the turn finishes:

Read the agent's summary.
Check any security review findings reported for the turn.
Inspect the complete diff.
Run the relevant tests or checks yourself.
Look for unrelated files, debug output, generated artifacts, or accidental deletions.
Keep only the changes you understand and intend to ship.

The agent's final message is a report, not proof that the work is correct.