Build a business that upholds your standards.
AA19 is a self-hosted autonomous business operating system. Nineteen agents. One brain. Governed execution. Organizational memory. Your infrastructure, your data, your AI accounts. Read the intelligence thesis, walk the five-step process, or open the architecture.
Observe. Generate. Verify. Learn. Execute. Repeat.
Most software asks "what should happen next?" AA19 asks "what happened last time?" Every approval, edit, and rejection becomes permanent organizational intelligence. Repeat a correction and the system strengthens the rule instead of duplicating it. That is the difference between a tool that executes and a system that learns.
Agent drafts an outreach email
Founder edits the subject line
System extracts the principle behind the edit (the rule, not the text)
Next draft, hours later, already follows the rule
Learning is not a feature. It is the substrate under all of them.
Data becomes behavior in five places, all feeding one memory. Every learning is scoped to the organization, the right department, and the right agent. See the same idea in long form in Approvals are a curriculum.
Approve, edit, or reject a draft. A two-tier engine (deterministic pattern matching first, AI distillation second) extracts the rule behind your action.
Upload a sales doc, brand guide, or writing sample. The system distills principles (never copy paste) and scopes each one to the org, department, and agent that need it.
Type a rule directly ("subject lines must include the org name"). Trusted at 100%, applied on the next run.
Rejected drafts, QA failures, and outcomes feed the Coach, which promotes proven patterns into standing guidance.
Goal, partnership, and research settings are automatically converted into targeting guidance that agents actually load before their next run.
One learning, applied everywhere it matters. Org-wide, per department, per agent. Visible in one place, the Memory page, newest first, never duplicated.
Meet the 19.
Five departments. Nineteen agents. Each one has what it produces, its cadence, and its current autonomy gate. The product documents itself, so this page mirrors the same registry the app runs on. Read the reasoning in The OS is the org chart.
- ›Brain
- ›Coach
- ›QA
- ›Memory
- ›Guidance Engine
- ›Decision History
- ›Chief of Staff
- ›Department Health
- ›Goal Engine
- ›Action Item Engine
- ›Content
- ›Socials (Facebook, LinkedIn, Substack)
- ›Brand Intelligence
- ›Workforce Registry
- ›Connections Registry
- ›Budget Governor
- ›Analytics
- ›Research
- ›Outreach (Outbound, Gmail, Follow-up)
- ›Partnerships
- ›CRM
Agents share one brain, one memory, and one governance model. Sub-agents (Facebook, LinkedIn, Gmail, Follow-up) expand from their parents without duplicating context.
See the architecture →A revenue engine that responds to inventory, not a schedule.
When new prospects drop below threshold, research fires automatically. Inventory healthy? It stands down. Research is a response, not a habit.
Autonomous research on your ICP, CSV upload, or a pasted company URL. Every contact is enriched on the way in.
Every contact gets a company description, buying-readiness signals, and a "why now" pulled through a multi-provider enrichment waterfall.
Personalized emails built from that research. Pain, expertise, clear ask. Same voice, unique execution per prospect.
Nothing sends without approval until autonomy is earned. See the three modes in Approval, Hybrid, AUTONOMY. Read the three modes of trust.
Warmup-aware volume (ramps 10 to 60 per week), randomized human-like send timing.
A second touch is drafted automatically 72 hours after the first.
A pipeline watchdog detects stalls and repairs them.
You approve emails. The system does literally everything else, and gets better at drafting them every week.
Sales infrastructure that keeps composing itself is closer to hiring your first digital employee than buying another tool.
Three gates per agent. Approval, Hybrid, Autonomous.
Every action reviewed. The default state for every new agent.
Routine work runs on its own, important decisions still escalate.
Independence granted only to behaviors that have proven themselves.
The autonomy score is fully evidence-based. No defaults, no floors, hard day-based ceilings. Read the case for it in Why most people get autonomy wrong.
One toggle on the dashboard. Every agent adapts its behavior to your capacity. The goal is not automation. It is capacity, yours.
Most tools give you a to-do list. AA19 gives itself one.
Brain patrols, Department Health, Goals, and QA generate items when something needs attention.
Critical, high, normal, low. Assigned to a subsystem, not lost in a shared inbox.
Items are verified before they are considered done. The system cannot self-certify its own fixes.
One open item per issue, never a pile of clones.
Visible on the dashboard and folded into your daily brief so nothing waits in silence.
Knowledge stops leaving when people do.
Upload documents, get distilled standards scoped to the org, department, or agent that needs them. Every rule the system runs on lives on the Memory page, searchable, filterable, trust-scored. Delete anything and every trace is removed.
Agents load their memory before every task. Org context, department context, agent rules, learnings. Every founder decision is recorded and decision history is queryable. Read the pillar guide on organizational memory and the deeper case in the intelligence page.
- ›The Slack thread you cannot find
- ›The senior operator who left last quarter
- ›The doc nobody opens after week one
- ›The rule that gets re-explained every hire
Goals the machine actually reads.
Set aspirations Brain drives toward and floors Brain patrols. Two shapes, one system.
Missed floors create prioritized action items automatically, routed to the subsystem that owns them.
Progress is reconciled continuously and reported in your daily brief, not once a quarter.
Goal changes convert into guidance agents load before their next task. Set a goal, watch behavior shift.
Adjacent read: Why most KPIs are lies.
Connect anything in minutes. Swap anything in seconds.
Anthropic, OpenAI, Gemini. Connect one, two, or all three. The system uses whatever you have, falls through automatically if a provider fails, and no model or provider is ever hardcoded.
Each connectable with a key paste and a verify click. Waterfalls are configured, not coded.
Swap any provider from the dashboard. No code change, no redeploy.
Status, last-verified, and error visibility per connection. You always know what is live and what is degraded.
Your API accounts, your keys, your rates. AA19 never sits between you and your models.
An autonomous system you can afford to leave running.
Set a daily API spend cap from the dashboard. Yours to change any time.
A budget governor pre-checks every AI call against the cap. At the cap, spending stops. Hard stop, not a warning email.
Total spend, daily cap, remaining today. One glance.
Halt all AI spend instantly and resume when you are ready.
Deterministic rules first, small models second, big models only when needed. Normal days stay well below the cap.
What a normal Tuesday looks like.
You own everything.
Your infrastructure, your database, your AI accounts, your data. AA19 is the operating system, not a middleman. Read the reasoning in The founder bottleneck nobody talks about and Symphony, not stack.