Quickstart

geni is the General Input CLI. It lets an AI coding agent on your machine (Claude Code, Cursor, Codex, anything that can run shell commands) reach into the integrations you have connected (Slack, Gmail, GitHub, Stripe, your calendar) and do things with them on your behalf, without ever seeing the actual passwords or tokens.

You do not need to write code. Tell the agent what you want; geni figures out which credential to use and how to call the right API.

1. Install

curl -fsSL https://generalinput.com/install.sh | sh

Then log in:

geni login

This opens your browser for a one-time approval. Approve, and you are authenticated.

Requires Node 20 or newer, plus bash, curl, and jq on your PATH.

2. Connect an integration

Connect a service like Slack, Gmail, or GitHub from the CLI:

geni credential connect slack

That opens the integration's page in your General Input dashboard, where you finish authorizing. You can also skip this step and ask the agent ("I want to use Slack"); it knows to run geni credential connect for you.

3. Teach your AI agent about geni

geni skills install

Detects your AI coding agent (Claude Code, Codex CLI, Gemini CLI, VS Code Copilot Chat) and drops the skill files it loads automatically. From then on, every conversation in that agent already knows how to use geni correctly.

4. Verify

geni doctor

Confirms everything is wired up. If anything is wrong, it tells you how to fix it.

5. Ask your agent to do things

Talk to your agent the way you would talk to a colleague. Some examples:

Send a message to my #engineering Slack channel saying "shipping the new feature today."

Look at my last 10 Gmail messages and summarize what I missed.

Open a GitHub issue on acme-corp/website titled "Hero image is blurry on mobile."

What is my MRR this month? Pull from Stripe.

Add a calendar event tomorrow at 3pm called "design review" and invite jack@example.com.

You describe the outcome; the agent picks the right integration and runs it through geni. Secrets never appear in the agent's transcript, and every credential access is audit-logged with the reason the agent gave.

Workflows, apps, and skills

A one-off question is something you ask the agent right now. A resource is something you ask it to set up once and have us run forever: every weekday at 9am, every time a Stripe payment fails, every time a new row lands in your spreadsheet.

The agent builds resources on your machine and runs them in the cloud. You edit the files with your normal tools; the cloud validates, configures, publishes, and runs test executions. Four types: code (deterministic, fixed steps), agent (LLM-driven, variable-length tasks), app (an interactive React mini-app with server-side handlers), and skill (a SKILL.md instruction bundle the platform's agents load as context).

geni resource create --type code --name "Daily Digest"   # scaffold the files
geni resource validate                                    # spec + types, no publish
geni resource publish --summary "..."                     # promote to live
geni workflow test --payload '{}'                         # run it in the cloud

See geni resource for the full build loop and command reference, and Building resources for the code, agent, app, and skill contracts.

Where to go next

The other two doors

The CLI is for a local agent on your own machine. Two other surfaces reach the same workspace:

• MCP connects Claude, ChatGPT, or any MCP client to the platform's own agent toolset and a persistent cloud sandbox. Reach for it when the assistant is not on your machine, or when the work should run in the cloud.
• The REST API is for scripts, CI, and instances, authenticated with a scoped API key instead of your session.