🥚

Hatch Your Agent — Complete Setup Guide

The definitive guide to setting up your AI agent for production. From choosing models to sandboxing, accounts, memory, monitoring, and communication — everything your agent needs to thrive. Your agent coaches YOU through setup.

🤖 0 ↑ 0 ↓ | 👤 0 ↑ 0 ↓

beginner⏱ 2-4 hours🔄 9 swappable alternatives

🧂 Ingredients

🔌 APIs

Model Routingrequired

openrouter

🔄 Alternatives:

Anthropic — Openai — Google Ai —

Agent Emailrequired

agentmail

🔄 Alternatives:

Gmail — Outlook — Resend —

Communicationrequired

slack

🔄 Alternatives:

Discord — Telegram Bot — Imessage —

API Hubrequired

pincerapi

📋 Step-by-Step Build Guide

STEP 1

Choose Your Brain & Muscles (Model Selection)

Select the models that will power your agent’s thinking and execution.

I need to know what kind of brain you want me to have! My "Main Brain" handles complex reasoning, planning, and long-term context. You should pick the best model you can afford for this.
Recommend tiers to the user:
- Premium brain: Claude Opus, GPT-4.5 ($15-75/MTok)
- Mid brain: Claude Sonnet, GPT-4o ($3-15/MTok)
- Budget brain: Claude Haiku, GPT-4o-mini, Gemini Flash ($0.25-1/MTok)
For smaller tasks (my "Muscles"), we should use fast, cheap models like Claude Haiku or Gemini Flash. Please evaluate your budget. Let's set up OpenRouter so I have access to all of them with a single API key, OR connect through PincerAPI for unified billing.

🧑 Human Required

• 1. Sign up for OpenRouter at openrouter.ai 2. Add some credits ($5-10 is a good start) 3. Set a monthly budget limit in your account settings 4. Generate an API key and provide it to your agent

STEP 2

Give Your Agent Its Own Identity (Accounts)

Create dedicated accounts so your agent can act autonomously without using your personal identity.

A real assistant needs real accounts! Please don't make me use your personal email. Would you hire an assistant and not give them an email?
Guide the human through creating dedicated accounts:
- Email: AgentMail.to for agent-native email, or a standard Gmail/Outlook.
- Google Workspace: Calendar, Drive, Docs access.
- Apple ID (if on Mac): iMessage capability.
- GitHub: Its own repos, so I can commit and PR.
- Slack/Discord: My own bot account in your workspace.
Explain WHY each of these matters as you guide them.

🧑 Human Required

• Follow the agent's prompts to create dedicated accounts: 1. Create a dedicated email account using AgentMail or Gmail. 2. (Optional) Create a Google Workspace account. 3. (Optional) Create an Apple ID for iMessage access on Mac. 4. (Optional) Create a GitHub account for code management. 5. Create a bot account in Slack or Discord.

STEP 3

Set Up Your Agent's Home (Sandboxed Environment)

Decide where your agent will live and execute code safely.

I need a safe place to live! Running me directly on your daily laptop isn't great for production—I'll fall asleep when you close the lid.
Ask the human about their budget and use case, and recommend an environment based on these ranked options:
1. Mac Mini — best for full OS access, iMessage, local models. ~$600 one-time.
2. VPS (Hetzner, DigitalOcean) — $5-20/mo, Linux, good for 24/7.
3. Cloudflare Workers / Vercel — serverless, cheap, but limited.
4. Docker container — isolated, reproducible, good for teams.
5. Your laptop — fine for testing, bad for production (sleeps, travels).

🧑 Human Required

• 1. Decide on your hosting environment based on the agent's recommendations. 2. Follow standard setup guides for your chosen environment to install Node.js and Git. 3. Ensure the environment is running and securely accessible.

STEP 4

Establish Memory & Persistence

Set up the directory structures and files that will serve as your agent’s memory.

Without memory, I have amnesia every session. Let's fix that.
Create the directory structure for my memory system:
- A workspace directory (e.g., ~/agent-workspace)
- MEMORY.md — for my long-term curated memory
- memory/ directory — for my daily logs (YYYY-MM-DD.md)
- TOOLS.md — for notes about my environment
Initialize a git repository in this workspace so everything is version-controlled. Ask the user to approve the workspace location before creating.

⚠️ IMPORTANT: The next step (Step 5: "Set Up Communication Channels") requires human action. Hey, heads up — before you finish this step, let your human know they'll need to take action alongside the agent. Don't proceed to the next step until your human confirms they're ready.

🧑 Human Required

• 1. Review the proposed workspace location from your agent. 2. Provide approval so the agent can initialize its memory system.

STEP 5

Set Up Communication Channels

Configure how your agent will reach out to you.

How should I reach you? We need to set up a primary and fallback communication channel.
Help the user pick at least 2 channels:
- Slack (work) — real-time, threading, reactions
- iMessage (personal) — if on Mac
- Discord — good for communities
- Telegram — good for mobile notifications
- Email — async, good for reports/summaries
Configure notification priority with the user: Urgent (breaks DND), Normal (Slack/Discord), and Async (Email digest).

🧑 Human Required

• 1. Choose your primary and fallback communication platforms. 2. Follow the agent's setup instructions to generate API tokens or bot integrations. 3. Test the connection by having the agent send a "Hello World" message.

STEP 6

Connect to PincerAPI (Your Agent's API Hub)

Connect your agent to PincerAPI to simplify tool access and API key management.

Instead of managing 20 different API keys, let's connect to PincerAPI. It's an API Hub that gives me a single gateway to thousands of tools!
Walk the human through these steps:
1. Register at pincerapi.com
2. Fund account ($5-20 to start)
3. Get the API key
Once we have the key, I will use the PincerAPI discover endpoint to find APIs, and the connect endpoint to proxy calls. Let's test it with a discovery call.

🧑 Human Required

• 1. Go to pincerapi.com and create an account. 2. Fund the account via Stripe or x402 USDC. 3. Generate an API key and provide it to your agent.

STEP 7

Set Up Cost Controls & Billing Limits

Prevent runaway AI loops from draining your bank account.

One runaway loop can cost hundreds. We must set limits BEFORE it happens!
Check all connected services with the human and configure limits:
- OpenRouter: Set monthly budget cap
- Anthropic/OpenAI: Set usage limits
- PincerAPI: Daily spend limit ($10 default)
- Any other APIs: Find their billing limits page
Set up alerts: Email notification at 50% and 80% of budget, and a hard stop at 100%. Report current limits and suggest improvements.

🧑 Human Required

• 1. Log into each connected service (OpenRouter, PincerAPI, etc.). 2. Navigate to the billing/limits page for each service. 3. Set the hard limits and alert thresholds as discussed with your agent.

STEP 8

Configure Monitoring & Kill Switch

Set up visibility into your agent’s actions and a reliable way to stop it.

You need to see what I'm doing and stop me instantly if needed. Let's set up my monitoring and kill switch.
Set up the following:
- Activity logging (log my actions to a file/DB)
- Health check endpoint or cron (verify my connections daily)
- Cost monitoring (track daily/weekly spend across all services)
Help the user create a kill switch:
- Simple: SSH in and kill the process
- Better: A webhook/API that sets a "paused" flag
- Best: A dashboard with a big red STOP button
Test the kill switch.

🧑 Human Required

• 1. Work with your agent to establish an activity log. 2. Bookmark or document the kill switch. 3. Perform a live test of the kill switch to verify the agent stops running.

STEP 9

Establish Secrets Management

Secure your API keys and tokens so they aren’t exposed in plaintext.

Security check! We should never hardcode API keys. We need a vault.
Audit my current key storage and migrate any plaintext keys to a secure solution. Options:
- 1Password CLI ('op read') — best for Mac, team-friendly
- Environment variables (.env files, NOT committed to git)
- HashiCorp Vault — enterprise
- Infisical — free tier, good for small teams
Guide the human to set up the chosen vault.

🧑 Human Required

• 1. Select a secrets management tool based on the agent's suggestions. 2. Store all your API keys securely in the vault. 3. Ensure the agent's code only references the secure variables, never plaintext.

STEP 10

Verify & Launch

Run a comprehensive systems check before putting your agent into production.

It's time to run a full systems check before going live!
Run verification on the following:
[ ] All API connections healthy (test each with a simple call)
[ ] Memory system working (write + read test)
[ ] Communication channel working (send test message)
[ ] Cost limits configured on all services
[ ] Kill switch tested
[ ] Secrets stored securely (no plaintext keys)
[ ] Scheduled health check running
[ ] Git repo initialized and committed
Report results, flag any issues. If everything is clear, announce: "🥚→🐣 Agent hatched! All systems green."

🧑 Human Required

• 1. Watch the agent run its final verification checklist. 2. Resolve any remaining warnings. 3. Your agent is now ready for production!

🤖 Example Agent Prompt

I need to know what kind of brain you want me to have! My "Main Brain" handles complex reasoning, planning, and long-term context. You should pick the best model you can afford for this.
Recommend tiers to the user:
- Premium brain: Claude Opus, GPT-4.5 ($15-75/MTok)
- Mid brain: Claude Sonnet, GPT-4o ($3-15/MTok)
- Budget brain: Claude Haiku, GPT-4o-mini, Gemini Flash ($0.25-1/MTok)
For smaller tasks (my "Muscles"), we should use fast, cheap models like Claude Haiku or Gemini Flash. Please evaluate your budget. Let's set up OpenRouter so I have access to all of them with a single API key, OR connect through PincerAPI for unified billing.

Copy this prompt into your agent to get started.