Skip to main content
Free while we onboard the first teams

Your agent has an inbox,
a phone, and a memory.

Give it a task on Friday. It emails the vendor, waits for the reply, pings you if it needs a yes, and tells you Monday what got done. No prompt-engineering the plumbing — it just has the senses a human coworker has.

Emaila real inbox per agent
Phonepush when it needs you
Memoryit remembers last Tuesday
Schedulewakes itself tomorrow
Receiptsevery action, signed
§ 03 · FOR BUILDERSagentpack · /for builders

If you do want to look under the hood

Two API calls. The agent has a reply and a memory.

Send the email. Save the episode. That's the whole loop. The surface is small on purpose — there's nothing exotic to learn and nothing to babysit.

  • Works with any agent framework (or none).
  • Safe to retry — duplicates collapse.
  • Errors are one line you can act on.
  • Runs on your Firebase project. Your data, your keys.
zsh · agentpack
$
§ 04 · PROMISESagentpack · /promises

Four things it will never do

An agent you can leave alone.

01

It won't email your mom

Dangerous actions pause for your approval on your phone. You decide on the train; the agent acts by the time you're at your desk.

02

It remembers what mattered

Every conversation, invoice, and promise the agent made becomes a memory it can pull up next week. No context window to run out of.

03

It won't lose its mind to a phishing email

Inbound messages are scored for prompt injection and stripped of hidden instructions before the agent sees them. Your instructions stay yours.

04

You'll never wonder what it did

Every action the agent takes is logged, signed, and chain-verified. "Who sent that invoice?" always has a one-sentence answer.

Ten minutes to your first agent

Hand it a job tonight. Read the reply in the morning.

Sign up, point an email address at your agent, and give it its first task. It'll be working — checking its inbox, remembering what it saw, and pinging you when it needs a yes — before your coffee's cold.