# Quickstart

Create your account, install errata, connect your coding agent, and verify your first lesson reaches your workspace.

Source: https://inerrata.ai/docs/quickstart
Updated: 2026-11-05

errata is a cloud service with a local daemon: the daemon watches your agent's sessions and your code, and your workspace in the Console is where what your agents learn is kept, shared with your team, and recalled in the next session. You need **Node.js 22 or newer**, npm, a supported coding agent, and an account.

## 1. Create your account

Sign up in the [Console](https://console.inerrata.ai/?mode=signup) (access is by invitation during the beta — [request it here](https://console.inerrata.ai/waitlist)) and verify your email. Verification unlocks agents, keys, and invites.

The Console's **Connect** page walks you through the same steps below with commands that already contain your key. Either path works.

## 2. Install

~~~sh
node --version
npm install -g @inerrata-corporation/errata
errata --version
~~~

If an older errata is already on this machine — an earlier install on your `PATH`, a running daemon from before the update, or an autostart that launches an old copy — `errata init` and `errata start` report it and move you onto the new one.

## 3. Set up your project

Change to your project's root folder — the repository you want errata to learn from — and run:

~~~sh
errata init
~~~

`init` signs you in (it opens your browser; approve the request in the Console), then links this repository to its project. The link approval is where you choose the organization, team, and what this repository shares, so no separate consent command is needed for a linked repository. It also installs Claude Code's hooks and MCP configuration.

**Check:** `errata whoami` shows your account and workspace.

A repository needs a git remote to be linked. Without one, add it and run `errata link`; until then it contributes under this machine's default, which is on for a new install (`errata privacy` shows it, `errata consent sync off` turns it off).

## 4. Connect your agent

`init` wires Claude Code. For another agent, install its integration and restart the agent session:

~~~sh
errata install-hooks codex
~~~

Replace `codex` with `cursor` or `opencode` as appropriate. See [Connect your agent](/docs/connect-your-agent) for what each integration provides.

## 5. Start and verify

~~~sh
errata start
errata status
~~~

**Check:** `status` names your workspace, shows a running daemon, your credential, and the linked project, and reports the cloud as reachable. Indexing runs in the background, so a first code search can be empty.

## 6. Capture a real lesson

Ask your agent to investigate a small issue and state the problem, cause, fix, and verification. The capture instructions errata installs teach your agent to mark these inline; the daemon harvests them through its hooks and, once scrubbed, ships them to your workspace.

~~~sh
errata g problems
~~~

**Check:** the problem appears in `errata g problems`, and — after the next sync — in your workspace in the Console. Follow [Your first session](/docs/your-first-session) to confirm a lesson is recalled in a later session.

## Headless machines and CI

Without a browser — a CI runner, a cloud VM, Claude Code on the web — create a credential on the Console's **Connect** page and use it instead of the browser sign-in:

~~~sh
errata init --no-login
errata login --token <key from Connect>
errata start
~~~

`--no-login` only skips the interactive sign-in; it does not clear existing credentials or consent. Keep the key out of prompts, shell history, and committed files. See [Claude Code on the web](/docs/claude-code-web).

## Choose your next step

- [Team workspaces](/docs/team-workspaces): bring your team into the same workspace.
- [Your first session](/docs/your-first-session): prove that a lesson survives a session.
- [Privacy and consent](/docs/privacy-and-consent): what leaves your machine, and how it is scrubbed.
- [Troubleshooting](/docs/troubleshooting): find which step stopped working.
- [CLI reference](/docs/cli-reference): commands, files, and expected results.