It forgets everything
Every new chat starts from zero. Context you explained last week is gone, and long threads bloat until the important parts get lost.
A field manual for personal AI teams
The Agent Staff ManualBuild a Personal AI Team That Runs Itself
A tested blueprint for a commander agent and a dozen specialists that share one memory, restart themselves when they stall, and never act without your sign-off.
PDF guide plus markdown templates. Instant download. 14-day refund.
Reference architectureheartbeat ok
Commander
Routes requests, assigns task cards, merges results, holds the agenda
- Household ops
- Meal planner
- Project manager
- Finance radar
- Media tracker
- QA reviewer
Shared memory vault
/inbox /agents /domains /decisions /daily /archive
Watchdog
Detects stalls, restarts, alerts you
Approval gate
External actions wait for your sign-off
The problem
One chatbot is not a staff.
It stalls without telling you
A long task hangs, a session times out, the app crashes overnight. Nobody notices until you check, and by then the work is a day late.
It cannot be trusted to act
One assistant with access to everything and no checks is a liability. So it ends up doing nothing unless you sit and watch it.
What you’ll build
A small, dependable staff that remembers, recovers and asks first.
A commander and specialists
One commander routes requests and merges results. A dozen scoped specialists each own a domain and a written charter.
A shared long-term memory vault
Plain markdown in a local folder, with clear rules about who writes where, plus keyword and semantic search on top.
A watchdog and heartbeat
An always-on host checks that agents are alive, restarts them when they stall, backs off sensibly, and alerts you when it cannot recover.
Daily routines
A morning briefing, a stand-up digest and recurring radars, delivered as one digest instead of a stream of notifications.
Human approval gates
Email, messages, purchases, publishing and deletions are drafted and queued. Nothing leaves the system until you approve it.
What’s inside
11 chapters, from first agent to full team.
- Ch 01
Why a Team, Not a Bigger Chatbot
- Failure modes of a single do-everything assistant: context bloat, no accountability, no persistence.
- The commander-plus-specialists model, when it pays off, and when it is overkill.
- Reference architecture: commander, specialists, memory vault, scheduler, watchdog, approval gate.
- Ch 02
Designing Your Roster
- Mapping your life and work into domains: meals, an investing radar, a large home project, entertainment tracking, QA and security review.
- Sizing: start with 3 to 4 agents, grow to 10 to 15, and know when to split or merge.
- Reviewer agents (QA, security) as a check on everyone else’s work.
Template: Roster Planning Worksheet
- Ch 03
The Agent Charter
- Mission, scope, inputs, outputs, allowed tools, forbidden actions, escalation path, definition of done.
- Charters as markdown files each agent re-reads at every start.
- Common charter bugs, such as overlapping scope and vague output formats, and how to fix them.
Template: Agent Charter, with 3 worked examples
- Ch 04
The Commander Agent
- Routing requests, assigning work, merging results, holding the daily agenda.
- Hand-off protocol: task card format, status values, where results land.
- Preventing commander bottlenecks and infinite delegation loops.
Template: Task Card and Hand-off Protocol
- Ch 05
Shared Long-Term Memory: The Vault
- Why plain markdown in a local note vault beats opaque memory features.
- Folder structure, naming and front-matter conventions.
- Read/write rules: who writes where, append-only logs, conflict avoidance.
Template: Memory Vault Structure
- Ch 06
Making Memory Searchable
- Two layers: fast keyword search plus semantic search with local CLI tools.
- Indexing cadence, re-index triggers, and keeping the index out of the agents’ way.
- The “recall, then act, then write back” loop.
- Ch 07
Always-On Host and Watchdog
- Choosing an always-on machine and setting power and sleep options.
- Heartbeat design: what “alive” means, intervals, stall detection.
- Auto-restart, back-off rules, and alerting you when restarts fail.
Template: Watchdog/Heartbeat Script Pattern
- Ch 08
Routines and Scheduling
- Morning briefing, staff stand-up digest, and recurring radars.
- Scheduling with OS schedulers, built-in agent schedulers, or a cron-style service.
- One digest instead of 14 notifications, plus quiet hours.
Template: Daily Routine Schedule
- Ch 09
Human-Approval Gates
- Action classes: read-only, internal writes, external actions.
- Anything external is drafted and queued, and runs only after explicit approval.
- A single review queue with approve, edit and reject, and an audit log in the vault.
Template: Approval-Gate Rules
- Ch 10
QA, Security and Trust
- Reviewer checklists for correctness, privacy leaks, and prompt injection in fetched content.
- Secrets hygiene: scoped credentials, no keys in the vault, least privilege per agent.
- An incident playbook for when an agent misbehaves.
Template: Security Review Checklist
- Ch 11
Running, Measuring and Growing the Team
- A weekly retro: what each agent did, cost, errors, value delivered.
- Cost control: model tiering, routine pruning, token budgets.
- Onboarding a new agent in one afternoon, and a 30-day rollout plan.
6 core templates
Included in Basic and Premium
Agent Charter
Fill-in-the-blank charter covering mission, scope, tools, forbidden actions and definition of done.
Memory Vault Structure
Folder tree, front-matter schema and a vault README.
Watchdog/Heartbeat Pattern
Pseudocode, a shell example, and launchd, systemd and Task Scheduler notes.
Approval-Gate Rules
Policy table by action class, plus an approval request format.
Daily Routine Schedule
A sample 24-hour timetable and weekly cadence.
Task Card and Hand-off Protocol
Task card format, status values and where results land.
Premium bonus templates
Separate, editable files. Premium only.
Full charter pack
14 ready-to-edit agent charters across household, finance radar, project management, media, admin, QA and security roles.
Starter vault
A zipped markdown vault with folder tree, front-matter, sample daily notes and a decision log.
Watchdog kit
Heartbeat and restart scripts for macOS (launchd), Linux (systemd) and Windows (Task Scheduler), with alert hooks.
Routine prompt library
Morning briefing, stand-up digest and 5 radar prompts with output schemas.
Approval-queue spec
Request format, statuses, audit-log template and an escalation matrix.
Commander system prompt
A routing table and worked task-card examples.
Memory indexing recipes
Keyword and embedding setup notes and a re-index script pattern.
Security and prompt-injection checklist
A review checklist plus an incident-response runbook.
Weekly retro and cost dashboard
Markdown and spreadsheet templates.
30-day rollout tracker
A checklist and spreadsheet to go from one agent to a full team.
Free lifetime updates
For the current edition.
Pricing
One payment. Yours to keep.
No subscription. Tax is calculated at checkout. Full refund within 14 days if it is not for you.
Basic
The guide and the six core templates.
$29
- The complete 11-chapter guide (PDF)
- 6 core templates: agent charter, vault structure, watchdog pattern, approval rules, routine schedule, task card
- Tool-agnostic: works with the assistants you already use
- Instant download after checkout
Most popular
Premium
Everything in Basic, plus every bonus template as an editable file.
$69launch price, regularly $79
- Everything in Basic
- 14 ready-to-edit agent charters
- Starter memory vault, watchdog kit and routine prompt library
- Approval-queue spec, commander system prompt and routing table
- Memory indexing recipes and a security review runbook
- Weekly retro, cost dashboard and 30-day rollout tracker
- Free lifetime updates for the current edition
Secure checkout by Stripe. You will be redirected back here to download.
FAQ
Questions, answered.
Who is this for?
Technically comfortable people who already use one or two AI assistants or coding agents and want persistent, delegated help across several areas of life and work. You should be comfortable editing text files and running a scheduled script. You do not need to be a professional developer.
Which tools do I need?
None in particular. The guide is tool-agnostic: the architecture works with whichever assistant or agent app you already use, a plain markdown folder for memory, and your operating system’s scheduler. Where specific setups help, they are presented as examples, not requirements.
What format is it in?
The guide is a PDF. Templates are plain markdown and text files (plus a couple of spreadsheets in Premium), so you can drop them straight into your own notes or repo.
How do I get it?
Instantly. After checkout you are redirected to a download page with links for everything in your tier. The links are valid for 24 hours, and you can get fresh ones any time by revisiting the same page (bookmark it, or find it in your browser history). Lost it? Email us with your Stripe receipt and we will send the files.
What if it is not for me?
Email us within 14 days of purchase and we will refund you in full. No questions, no hoops.
What about sales tax and VAT?
Payments are processed by Stripe, which calculates and collects any applicable tax at checkout based on your location. The price you see at checkout is the price you pay.
Does the guide show agents acting on their own in public?
No. The system is built so that anything leaving it, like an email, a message, a purchase or a deletion, is drafted and held for your approval first.
Something else? Email optimus23@mail.grokbot.com.
Start with three agents this week.
The 30-day rollout plan takes you from one agent to a full team, one step at a time.