How to Use Cady: The Beginner's Guide
CodexFuse Team · 2026-11-03T18:34:33.440Z · 10 min read min read
What Is Cady, Exactly?
Learning how to use Cady agent orchestrator software starts with a plain-English definition: Cady is a shared workspace that all of your AI agents connect to, so every agent — whichever AI subscription it runs in — works from the same company goals, decisions, and records. Its own site calls it "your agent orchestrator, tracking results and managing your agents across your team."
The word "orchestrator" sounds like jargon, but the idea is simple: an orchestrator is the team manager for your AI. You already pay for AI subscriptions like Claude, ChatGPT, or Gemini. Without Cady, each agent starts every task with none of your company's context. With Cady, you connect each agent once, and every agent reads from the same goalpost and writes its results to the same team sheet. The maker's pitch is aimed squarely at bootstrappers: run your whole company through AI agents, managed like a professional sports team.
One clarification up front, because the name sits in a crowded neighborhood: Cady is the product from cadyworks.com, launched on Product Hunt on October 27, 2026. Caddi (trycaddi.com, spelled with a double-d and an i) is a different company selling an AI workforce for back offices, and Cadre OS and Cadderly are unrelated AI products. This guide covers Cady only.
If your first connected agent will be a coding agent, keep CodexFuse's Claude Code beginner guide open in another tab — Cady plugs into the same agent you already know, so the learning curve is mostly the Cady side.
What Can Cady Actually Do for a Small Team?
Cady does five distinct jobs, and understanding them separately is what makes the product click.
| The primitive | What it is | The beginner move |
|---|---|---|
| One goalpost | Every agent works toward the same company goals, so its work actually moves the company forward instead of drifting. | Write 2–3 real company goals before connecting any agent; vague goals produce vague agent work. |
| A team sheet | Every agent and its results in one place — not a stream of activity you have to scroll through. | Check the team sheet each morning instead of asking each agent "what did you do?" |
| Run-by-run review | Risky work gets checked against your rules before it goes out. | Decide your approval rules first (see the approvals section below), then let agents run. |
| Coaching that sticks | Fix a prompt or skill once and every future run improves — the skills are open source. | When an agent makes a fixable mistake, correct the skill, not just the single run. |
| Receipts | Every result comes with a receipt you can inspect, so you can see exactly what the agent did. | Read the receipt before approving any send, publish, or spend. |
The practical upshot is vendor independence, which Cady markets as "vendor-proof your AI": your agents work across your AI subscriptions, so your company is not tied to one provider. Your ChatGPT cannot easily work with your colleague's Claude — but in Cady, both work from the same company, goals, and records. Each assistant keeps its own memory for its own user; Cady holds the shared part: who owns what, what has been decided, and what is still open. Work handed from one person's agent to another's does not need explaining again.
What Do You Need Before You Start?
You need three things: an AI subscription you already pay for, a Cady account, and a small enough starting scope that you can actually manage the results.
The subscription part is the point of the product: there is no AI bill from Cady — your own subscription does the work, so Cady adds coordination without adding a second meter. The Product Hunt listing states this plainly: "Works with the AI you already pay for, so there's no AI bill from us."
The scope advice comes from the maker himself, who wrote that the product is early and to expect some bugs: try it with two of your AI apps, or two people on your team. Do not wire up ten agents on day one. Two agents sharing one goal is enough to feel the difference between scattered chat and a managed team.
As of November 3, 2026, Cady's own site says "Free to start. Pro is free until 31 December 2026 for every company created by 30 November 2026." Pricing for a two-day-old product moves fast, so re-check the pricing page before you commit — treat every price in this article as a launch-week snapshot.
How Do You Connect Your First Agent?
Connecting your first agent is a five-step arc: create your company, write the goalpost, connect one agent, give it a first real job, and review the receipt.
- Create your company in Cady. This is the shared workspace everything else hangs off: goals, decisions, people, and work.
- Write 2–3 concrete company goals. Cady gives your agent your company's goals, owners, and context — vague goals are the most common reason the first run disappoints.
- Connect your first agent once. Connect your Claude, ChatGPT, Gemini, or Grok account to Cady from the same chat app you already use today; each agent connects once and keeps working from that app afterward. The skills repository lists quick-start paths for Claude Code (a plugin install), Codex (an MCP server add), and claude.ai (a skill upload).
- Give it a first real job, not a test prompt. Real jobs the product is built for include outreach, content, and record-keeping — pick one small job with a visible result, like drafting a weekly outreach message.
- Read the receipt before it goes out. Every result comes with a receipt you can inspect; make the first review slow and deliberate, then speed up as you learn what "good" looks like.
Under the hood, the connection runs over MCP — the Model Context Protocol. One MCP server (app.cadyworks.com) serves your company's state: goals, decisions, people, work, and approvals. Your agent connects to Cady through that one server, loads the right skill from the open-source skill library, and reads or writes company state through Cady, which checks authority and records every effect. If MCP is new to you, our what MCP is explainer covers the protocol before you touch the wiring.
What are the open-source skills?
The skills live in a public GitHub repository, malcolmai/cady-skills, licensed under the Mozilla Public License 2.0. The repo calls itself a "method library": 66 skills for running a company, from finding customers and writing outreach to month-end close, raising money, compliance, and shipping software. Methods live in the repo, versioned and open — your company's facts never do.
Two privacy lines in the repo's README matter for beginners: agents report run results and usage to Cady but never upload transcripts, and there are no scripts or install hooks in the library — it is text only. The skills also require a Cady account to be useful at all: "The skills hold the methods; Cady holds your company's state, approvals and records. Without Cady they do nothing useful."
What Does the "One Goalpost" Habit Look Like Day to Day?
The one-goalpost habit is a daily loop: the agent pulls the current goals, does the work, and reports the result to the team sheet; you review the receipts and update the goals when the company moves.
Here is the beginner version of that loop, using the two-agent starting scope: every morning, open the team sheet and read what each agent delivered yesterday. When something looks off, do not just re-run the agent — fix the prompt or skill once, because coaching sticks and every future run improves. When the company's direction changes, change the goalpost first, then let the agents re-plan against it. Coaching an agent on a stale goalpost is wasted effort.
What waits for your approval?
Cady lets you choose which actions pause for your approval, and the defaults are worth knowing before you connect anything.
| Action type | Default behavior | Your call |
|---|---|---|
| Invitations | Waits for approval | Keep this on — invites create access |
| Data room shares | Waits for approval | Keep this on — shares expose files |
| Outreach (email, LinkedIn) | You choose | Start with approval on; relax later |
| Sending, publishing, spending | Follows your approval rules | Write the rules before the first send |
The product's philosophy on this is direct: you choose which agents to add and which work they can do, you control their connections and permissions, and sending, publishing, and spending follow the applicable approval rules. Beginners should start strict — approval on for everything outbound — and loosen the rules only after they trust the receipts.
Who Should Not Use Cady?
Cady is a poor fit for two kinds of user, and the product's own launch copy admits it is early.
First, a solo developer with one coding agent does not need an orchestrator yet — the shared-goalpost machinery pays off when two or more agents (or people) must stay aligned. If you just want a coding agent working while you sleep, that is a different lane entirely: our Agent AFK beginner guide covers unattended CLI coding, which is delegation, not team orchestration.
Second, teams that need polished, production-grade automation should wait. The maker himself says the product is early and asks early users to report what breaks. The try-it-with-two-apps advice is a genuine instruction, not marketing humility.
Cady vs Agent AFK vs Plain Chat: Which Fits You?
Cady occupies the team-orchestration lane; two adjacent lanes solve different problems, and picking the wrong one is the most common beginner mistake.
| Cady | Agent AFK | Plain chat | |
|---|---|---|---|
| Problem it solves | Many agents, one shared goal | Coding tasks while you're away | One-off questions |
| Unit of management | The company: goals, owners, decisions | The task: one delegated job | The conversation |
| Memory | Shared company state, per-user assistant memory | Per-task context | Per-chat context |
| Approvals | Rules per action type | Task completion check | None |
| Best first step | Write 2–3 goals, connect 2 agents | Delegate one small coding job | Ask one sharp question |
The lanes are complements, not rivals: a solo dev can use Agent AFK for overnight coding, graduate to Cady when the company gets a second agent, and keep plain chat for quick questions forever. One keyword, one page — this article is the Cady lane.
Sources
- cadyworks.com — product site, fetched November 3, 2026 (definition, vendor-independence claim, connect-once flow, shared-state lines, approval rules, pricing line "Free to start. Pro is free until 31 December 2026 for every company created by 30 November 2026.")
- producthunt.com/products/cady — Product Hunt launch listing, published October 27, 2026, updated November 2, 2026 (maker description, one-goalpost / team sheet / run-by-run review / coaching primitives, supported agents, "no AI bill from us", try-with-two advice)
- github.com/malcolmai/cady-skills — open-source skills repository, Mozilla Public License 2.0 (66-skill method library, MCP server architecture, privacy lines, quick-start paths for Claude Code / Codex / claude.ai)
Frequently Asked Questions
Does Cady add an AI bill on top of my subscriptions?
No — that is the product's core pricing claim. Cady is free to start, your own AI subscription does the work, and the launch-week terms put Pro free until December 31, 2026 for companies created by November 30, 2026. Re-verify the pricing page before you commit, because launch terms move fast.
Which AI apps can I connect to Cady?
Claude, ChatGPT, Codex, Gemini, Grok, Hermes, or any MCP client, per the launch listing. Each agent connects once and keeps working from the chat app you already use.
Can my ChatGPT and my teammate's Claude share work?
That is the main selling point. Each assistant keeps its own memory for its own user, while Cady holds the shared part — who owns what, what has been decided, and what is still open — so work handed from one person's agent to another's does not need explaining again.
Does Cady see my company's data?
Cady holds your company's shared state: goals, decisions, people, work, and approvals. Per the open-source skills repo, agents report run results and usage to Cady but never upload transcripts, and the skill library contains no scripts or install hooks — text only.
What's the difference between Cady and Agent AFK?
Cady is team-level orchestration: many agents aligned to one shared goalpost with approvals and receipts. Agent AFK is the unattended-CLI lane: delegating coding tasks to one agent while you're away. Different lanes, different intents.
