Skip to main content

Quickstart

This guide gets you from zero to a working agent in a few minutes. By the end you will have the desktop app installed, a repository open, and a completed agent task with a reviewable diff.

Prerequisites

  • A 4RGED account and an invitation to an organization. If you don't have one, ask an organization admin to invite you from the Members page.
  • A project on your machine (a Git repository is recommended so the agent can scope changes).
  • Network access to your organization's identity provider and 4RGED services. Corporate proxies may require extra configuration — see Installation and Troubleshooting.

Where apps live in production

After sign-in, 4RGED routes you to org-scoped apps on 4rged.ai: the Agents workspace at /agents, Projects at /projects, and the dashboard at /dashboard. Authentication runs through auth.4rged.ai; backend APIs are api.4rged.ai (policy, desktop sessions) and agents-api.4rged.ai (cloud runs). The CLI binary is 4rged — install from 4rged.ai/cli/install and run 4rged login to bind the same org boundary as the desktop app.

RequirementMinimum
RAM8 GB (16 GB recommended for large repos)
Disk2 GB free for the app and caches
OSmacOS 13+, Windows 10/11, or a supported Linux distro
GitInstalled and on your PATH (recommended)

Install and sign in

  1. Install the desktop app

    Download 4RGED for your platform and install it. See Installation for per-platform details, system requirements, and the CLI.

  2. Sign in

    Launch the app and sign in. You'll authenticate through your organization's identity provider and land in your default workspace. If sign-in loops or fails, check that your invitation is still valid and that any corporate proxy allows 4RGED endpoints.

  3. Open a folder

    Open the folder you want to work in. The agent uses this as its working directory and respects your .gitignore. For best results, open the repository root — not a nested subfolder — so the agent sees the full project structure.

  4. Confirm model access

    Open the model picker and confirm at least one model is available. Your organization controls the catalog; if the list is empty, ask an admin to enable models in Model control.

Run your first task

Start with something small and verifiable — a new endpoint, a test fix, or a documentation update. Avoid multi-hour refactors until you are comfortable with approvals and diffs.

  1. Describe what you want

    Open the agent panel and describe a change in plain language, for example: "Add a health-check endpoint at /healthz that returns 200 OK." Include constraints if they matter: framework, file location, or test expectations.

  2. Review before it edits

    For non-trivial tasks, ask the agent to outline what it will change before it edits files. Read the outline, then approve or refine. You can say: "Outline your plan before editing any files."

  3. Approve tool runs

    The agent asks before running commands or editing files outside the safe set. Approve the steps you're comfortable with. Deny or refine anything that looks wrong — the agent will adjust.

  4. Review the diff

    When the agent finishes, review the diff like any pull request, then keep or discard it. Run your own tests if the agent did not, especially before committing.

# Example: the agent might run commands like these on your behalf
git checkout -b add-healthcheck
npm test
curl -s -o /dev/null -w "%{http_code}" http://localhost:3000/healthz

Model selection

Your organization defines which models appear in the picker. Defaults may differ by surface (desktop, CLI, cloud). Use this table as a rough guide — your admin may have restricted options.

Task typeSuggested approach
Small, local editsA fast default model — good for typos, single-file changes, and quick questions
Multi-file featuresA capable reasoning model — better planning and fewer missed edge cases
Large refactorsOutline first, then a strong model; consider handing off to a cloud agent
CI / headless runsUse the model your pipeline admin configured in Model control
Pick the right model

Your organization controls which models are available. If a model you expect is missing, an admin can enable it in Model control. Switching models mid-task starts a fresh context — finish or discard the current run before changing.

Keyboard shortcuts

Use ⌘K (or Ctrl K) to search docs while you work. In the agent panel, refer to files with @filename when your build supports it — the agent will prioritize those paths. See Prompting for more patterns.

Common first tasks

Once the health-check exercise works, try a few tasks that teach how 4RGED behaves:

  1. Fix a failing test — paste the failure output and ask the agent to fix the root cause, not just silence the assertion.
  2. Add types or docs — low-risk changes that show how the agent navigates your codebase.
  3. Explain unfamiliar code — ask for a walkthrough of a module before you change it.
  4. Run the test suite — approve a terminal tool run and watch how the agent interprets failures.
  5. Open a project task — if your org uses Projects, link your work to a task so progress is visible on the dashboard.

Troubleshooting tips

SymptomThings to try
Sign-in fails or loopsConfirm your invite is active; try signing out of other SSO sessions; check proxy settings
No models in the pickerAsk an admin to enable models in Model control
Agent can't see a fileEnsure the file is not in .gitignore; open the repo root folder
Commands fail immediatelyVerify the toolchain is installed (node, python, etc.) and your shell environment is loaded
Diff looks wrongDiscard, refine the prompt, or ask for an outline before edits
Slow responsesLarge repos need more context-gathering; try scoping with a specific file or folder
Don't skip review

The agent is powerful but not infallible. Always review diffs and run tests before merging. Never paste secrets, API keys, or production credentials into prompts — see Agent security.

For a fuller list of issues and fixes, see Troubleshooting.

What's next