Skip to Content
DocsGetting Started

Getting Started

Install Agor, open the UI, and follow the onboarding wizard. Your first AI teammate will help you get your bearings and get everything else set up. (New here? Raising Your First AI Teammate explains why one teammate is the whole onboarding.)

Install

Most users should start here. It’s the most universal path, and it’s the one the rest of the docs assume.

Requires Node.js ≥ 22.12 (install ) and Git on PATH. HTTPS remotes also require a working system CA trust store; SSH remotes require an SSH client and configured keys or agent access. Run git --version before agor init.

# Install Agor
npm install -g agor-live
 
# First run
agor init
agor daemon start
agor open

Run agor init after installation. It shows a multi-select for every supported agentic tool and requires an intentional selection of at least one. The prompt explains that each selection adds download time and disk use; credentials are configured later in the UI. Init writes the selection once to agentic_tools.installed and installs the exact version-aligned packages before reporting success. Press Ctrl+C to cancel. An intentionally tool-free headless deployment must use --agentic-tools none.

To change the deployment policy later, edit agentic_tools.installed in config.yaml, run agor install --sync, and restart the daemon. Agor does not silently rewrite existing config. Repeat the readiness check with agor doctor; see Agentic tools for the complete model.

Follow the onboarding wizard

The first time a member opens the UI, the Onboarding Wizard runs — five quick steps that end on a new board. Read-only viewers enter the workspace directly because their role cannot provision boards, repos, branches, or sessions.

  1. Goals. Pick up to two outcomes, such as Ship without the busywork or Dig into anything. The recommendations and your teammate’s first briefing are tailored to what you want done. You can skip this step for a neutral start.
  2. Teammate. Name your AI teammate and choose a starter template, or skip this step to create a plain board without a teammate.
  3. AI. Connect a model: Claude (recommended), GPT, Gemini, or a custom endpoint. Paste an API key, use a Claude subscription token, or sign in with ChatGPT; Agor verifies the credential on the spot. You can also skip and connect one later.
  4. Tools. Keep or deselect goal-tailored suggestions, or skip. Sign in through Catalog opens a Catalog detail drawer inside onboarding. Review its access disclosure and explicitly Connect to start OAuth or enter a token securely. Connect saves only the MCP connection; it does not create a workspace or tryout session. Once ready, choose Return to onboarding. Back, cancellation, and Skip do not delete saved connections.
  5. You’re ready. Agor creates the board. If you named a teammate, it seeds that teammate from the framework repo  and, when an AI model is connected, launches the first session.

Goals, teammate setup, AI, and tools are optional; the final action always creates or resumes one board. If setup is interrupted after that board is created, reopening the wizard resumes the same board instead of creating another.

You can press Escape or the X on any step, including the final step and after an error, to close onboarding and finish later. This records a deferral, not a successful completion, and does not provision anything by itself. Continue from Settings → Profile → Resume onboarding; Resume preserves saved progress, while Restart intentionally starts the wizard over. A failed final action keeps the saved board/progress available for a retry rather than creating another board. If setup takes longer than expected, Agor keeps that one attempt in flight instead of starting a duplicate; you can still close and finish later.

When the board opens, a connected teammate’s first session picks up where the wizard left off — it already knows its name, goals, and goal-tailored tool suggestions. Curated services use the same Catalog consent, permission, readiness, and connection services as the rest of Agor. GitHub MCP uses its reviewed Catalog entry and a fine-grained personal access token entered in the secure drawer; connecting a repository does not authenticate GitHub MCP. Never paste a PAT into chat. Connections completed here are attached to the first teammate session when setup finishes; if you reload, saved connections remain available in My Servers. OAuth popup navigation alone is not a confirmed connection: wait for Connected and ready, or retry in the drawer or recover through My Servers. Workspace MCP policy may require an administrator to allow installation.

Slack onboarding recommends gateway messaging. A Slack gateway lets you message a teammate through a bot. Onboarding prefers an enabled existing gateway you can use, keeping its current teammate destination; it never silently creates a duplicate or retargets a gateway. If none is usable, only administrators can request their teammate’s help creating one. The teammate must recheck permissions and inventory, collect non-secret choices, create one disabled draft, then use the secure gateway credential widget and verify it before enabling.

Slack MCP gives an agent Slack tool access; gateway messaging does not grant it. Slack is not currently in Catalog: official Slack MCP  requires a registered internal or approved OAuth app and does not support dynamic client registration. Agor has no reviewed public OAuth client for this entry. Slack MCP is not selected or recommended as a connection alongside the gateway. An already configured, approved Slack MCP server can be managed through existing MCP settings. Never reuse gateway bot/app tokens as MCP credentials. From there the teammate can register repos, organize its board, structure a workflow, or just be a thinking partner. Anything a user can do in Agor, the teammate can do too. It has full access to the Agor MCP Server. If you skipped AI setup, connect a model later from Settings → AI & Agents and start a session from the board.

Customize your install

Agor runs out of the box on SQLite as a single-user instance, but it’s designed to scale up:

  • PostgreSQL backend instead of SQLite, for multi-user / team setups
  • Multiplayer mode with branch-level RBAC and per-user permissions
  • Execution isolation with fail-closed local sandboxing or a reviewed delegated substrate
  • Docker Compose and source builds for contributor / dev workflows — see the Docker Compose tab above for the evaluation quickstart
  • Custom config / database paths for non-default install layouts

For any of those, see Extended Installation. For credential alternatives (claude setup-token, codex login --device-auth, workspace-wide env vars), see Authentication.

What’s next

Get help

Last updated on
Your First AI TeammateExtended Installation