Multi-Agent Outreach System
Each agent has a name, a role, a personality, and a SOUL file that defines how it thinks. They communicate through Discord, store state in Convex, and execute on a $24/mo droplet.
A human sets the objective. The agents handle everything from lead scraping to reply classification. Each step is logged, costed, and auditable.
The difference between a prompt and a system is context. These agents have names, roles, personalities, and guardrails. They're managed like people.
Every agent has a markdown file that defines its identity, responsibilities, tone, and boundaries. Same concept as CLAUDE.md. The SOUL is the management layer.
Agents don't ask permission for routine ops. Health checks, log entries, data scoring: all autonomous. But campaign launches, money moves, and opportunity creation require human approval.
When something breaks, you know exactly which agent owns it. Loki handles quality. Thor handles delivery. Heimdall watches them both. No ambiguity, no shared responsibility.
Code, SOULs, and documentation live in the repo. Convex holds runtime state. Discord is the communication layer. If the server dies, redeploy from git.
A 7-agent team running on a single DigitalOcean droplet, Convex for the database, and cheap LLMs for the heavy lifting.
Not everything runs on autopilot. The system has clear tiers: what agents handle alone, and what requires human sign-off.
This is context architecture: the skill of building systems where AI has the right information at the right time to make the right call. It's not prompting. It's management.
The same method that built this 7-agent outreach system is what I teach. Build your first project in 7 days. Same principles, scaled to whatever you're building.