Install

Walk through the Cygent onboarding wizard end-to-end — provisioning, integrations, and the first actions to take.

Overview

Cygent is provisioned per-customer — an Enterprise Beta instance spun up by Cyfrin with your organization's name, your integrations, and your repos. This page walks through what happens from the moment someone from Cyfrin hands you the onboarding URL to the moment you run your first audit.

ℹ️

If you don't yet have an instance, apply to the Enterprise Beta waitlist. Cyfrin will provision the instance and send you an invite.

Before you start

Have these on hand before opening the wizard — you'll move faster if you're not context-switching mid-flow:

RequirementNotes
Chat workspaceSlack, Discord, or Telegram. Slack has the deepest integration surface.
GitHub admin rightsNeeded on the org or repos you want to connect.
Repos in scopeOne or two active repos — not every repo you own.
Default review channelWhere audit results and PR reviews will post. #security or #cygent works.

Step 1 — Sign in and name your organization

Sign in with the email Cyfrin invited. You'll be prompted for an organization name and slug — the slug shows up in dashboard URLs and in a few notifications, so pick something you're happy with long-term.

Step 2 — Create your first agent

An "agent" is one Cygent instance with its own configuration, integrations, and projects. Most organizations start with one agent. Give it a name (the name your team will see in Slack DMs), pick a description, and move on.

Step 3 — Connect a chat platform

Pick Slack, Discord, or Telegram. You'll be sent through an OAuth flow:

PlatformWhat happens
SlackInstalls the Cygent Slack app. Choose the workspace, approve the scopes, Cygent will DM you to confirm.
DiscordAdds the Cygent bot to a server. You pick which server.
TelegramStart a chat with the bot link, or add the bot to a group.

You can connect more than one later. Start with whichever platform your team actually uses.

Step 4 — Connect GitHub

Install the Cygent GitHub App on the org or user account that owns your repos. You can grant access to all repos or a specific list — a specific list is almost always the right call, and you can add more later.

ℹ️

The GitHub App powers PR reviews, issue creation, and the /cygent commit comment trigger. See PR review comments for what this unlocks.

Step 5 — Pick your first repos

The repo picker shows every repo the GitHub App can see. Check the ones you want Cygent to audit and review. These become your Projects.

💡

Start narrow. One active repo on day one is better than twenty stale ones — you'll get sharper results and less noise while you calibrate.

Step 6 — Choose a default review channel

Select the Slack / Discord / Telegram channel that receives audit results and PR review summaries by default. You can override this per project later from Settings.

Step 7 — Pick an autonomy preset

Cygent has four presets. Balanced is the sensible default — the bot responds to messages, posts PR comments, and asks for confirmation before running a review. You can change this any time. See Behavior & Autonomy for the full trade-off analysis.

PresetStance
RestrictedDormant unless @mentioned
BalancedReplies, posts PR comments, asks before running reviews
AutonomousFull proactive mode — auto-review, auto-issue, smart interjection
CustomToggle each behavior individually

Step 8 — Finish and wait for provisioning

Cygent provisions the agent. This typically takes a minute or two — you'll see the instance move through provisioningrunning in the dashboard, and the agent will DM you in your chosen chat platform once it's live.

What happens during provisioning

While the status sits at provisioning, Cygent is:

  • Creating your isolated instance — its own database, its own queue, its own configuration. No shared tenancy with other customers.
  • Indexing the repos you connected. For each project, CARA pulls the code, detects the stack (Solidity, Rust, Go, a mixed Web2/Web3 monorepo), and builds initial context.
  • Encrypting your integration credentials at rest. GitHub App tokens, chat workspace tokens, RPC URLs for Battle Mode, anything you add later — all encrypted.
  • Registering webhooks on your GitHub repos so PR opens, pushes, and issue events reach Cygent.

You'll know it's ready when

Dashboard is 'running'

The agent card on the dashboard flips from provisioning to running with a green indicator.

You get a DM

Cygent sends a welcome DM in your chosen chat platform, confirming the connection and listing your projects.

Projects show up

The Projects tab on the dashboard lists every connected repo with its detected language and a Run Audit button.

Once the agent is running and you've received your DM, provisioning is complete. Move on to your first actions.

First actions after provisioning

Don't sit and wait. The agent only gets useful once it has seen a real audit:

  1. Run your first audit. Click Run Audit on a project — or in Slack: @cygent run a full audit on <repo>. Watch the live progress view.
  2. Skim the findings. Some will be real, some will be invalid in your specific protocol. Mark the invalids as Invalid with a short reason — that's training Cygent on your codebase context.
  3. Open a PR, any PR. Even a trivial one. Watch what the inline review looks like on GitHub so you know what to expect when real work lands.
  4. Ask Cygent a question in Slack. @cygent what's the worst finding you found? or @cygent give me the big picture. Get a feel for how it responds.

The full first-week playbook is at What to do first.

When to create a second agent

Most teams run on a single agent. You create a second one when you want isolated context between teams or protocols:

ReasonWhat you get
Separate protocolsA lending protocol and a stablecoin don't share audit context.
Separate teamsTwo teams shipping independently get separate review channels, settings, backlogs.
Separate environmentsAn "internal experiments" agent distinct from the "production" agent on mainnet code.
ℹ️

A second agent is a clean slate. It doesn't inherit memory, notes, or findings from the first one. That's usually what you want — but it means you'll re-run initial audits and re-teach context on the new agent's projects.

Troubleshooting

Next