Cloud-native agentic pods running on DigitalOcean. Each business, each client, each SaaS idea gets its own autonomous team — with knowledge graphs, intelligent model routing, and self-healing architecture. We orchestrate from anywhere.
Between two agencies, four SaaS products, a SaaS incubator, and 60+ clients — we have more work than any team can manually orchestrate. Our skills, MCPs, knowledge graphs, and automation tooling are world-class. But they only run when a MacBook is open and connected.
All agentic work stops when the MacBook closes. Client campaigns, SEO crawlers, ad optimizations, deploy pipelines — everything goes dark until someone opens the lid.
RankPrompt agents compete with CS360 agents for the same compute. A stuck crawler blocks a deploy. One project's context bleeds into another's.
Every session starts cold. Agents re-learn patterns they discovered last week. Task retrospectives aren't feeding forward. The learning loop is broken.
We can't add a Weston Funding ad team alongside a RankPrompt SEO team alongside a Franchise Promo growth team without them stepping on each other.
Every heartbeat check, every content draft, every data parse runs through Claude Opus. A CEO-level model doing intern-level work. We're burning budget on tasks a $0.001 model handles fine.
Trevor is the conductor, the monitor, the debugger, and the escalation path — for everything, simultaneously. That doesn't scale past 3 active projects.
Every business, client, and SaaS idea gets its own autonomous pod — a self-contained team of AI agents with their own skills, memory, budget, and knowledge graph. They learn from every task. They get unstuck on their own. They only call you when it actually matters.
A Paperclip pod is a Docker container running on DigitalOcean with its own Postgres database, its own agent team, its own skills library, and its own growing knowledge graph. Each pod operates like an autonomous business unit — it has a CEO agent that manages the team, assigns tasks, checks health, and escalates only when stuck. Your MacBook and Mac Mini become conductors: they don't do the work, they command, observe, and intervene. The pods run 24/7/365 regardless of which devices are online.
This is not just "run Claude in the cloud." Each pod maintains a persistent knowledge graph that grows with every task. When an agent discovers that a Meta campaign performs 3x better with carousel format for restaurant clients, that insight gets stored as a relation in the KG. Next time any agent in that pod runs a restaurant campaign, it reads that insight first. Agents don't just execute — they accumulate institutional knowledge. Over weeks and months, each pod becomes an expert in its domain. Combine this with OpenRouter for intelligent model routing: the CEO agent decides which model handles each task based on complexity. Strategic decisions get Opus. Content drafts get Haiku. Data parsing gets Gemini Flash. The pod gets smarter and cheaper simultaneously.
Conductors command. The orchestrator manages health. Pods do the work. Each layer is independently scalable.
CONDUCTOR LAYER (Your Devices)
===========================================================================
MacBook (Mobile) Mac Mini (Always-On)
SSH + API on demand Cron health checks every 15 min
Approve budgets, review skills Daily 6am digest email
Visual dashboard via port-forward Auto-restart crashed pods
\ /
+----- HTTPS / SSH Tunnel ---------+
|
===========================================================================
DIGITALOCEAN CLOUD
===========================================================================
+---------------------------------------------------------------------+
| ORCHESTRATOR (Port 8080) |
| |
| Health Aggregator API Gateway Alert Dispatcher |
| Pod Lifecycle Mgmt KG Sync Daily Digest Builder |
| Model Cost Tracker Skill Sync Escalation Router |
+------+----------+----------+----------+----------+------------------+
| | | | |
+------v---+ +---v------+ +-v--------+ +v-------+ +v-----------+
| CS360 | | Rank | | Franchise| | Agency | | Incubator |
| Pod | | Prompt | | Promo | | Ops | | Pipeline |
| | | Pod | | Pod | | Pod | | Pod |
| 6 agents | | 4 agents | | 4 agents | | 5 agent| | 3 agents |
| ads,seo | | seo,aeo | | growth | | client | | research |
| email | | content | | content | | report | | validate |
| deploy | | api,bolt | | social | | health | | build,test |
| cro | | | | ads | | ads | | |
+----------+ +----------+ +----------+ +--------+ +------------+
| | | | |
+---------------------------------------------------------------------+
| SHARED LAYER |
| |
| /opt/ac-skills/ 900+ skills, git-synced |
| /opt/ac-knowledge/ Shared Knowledge Graph (JSONL) |
| /opt/ac-mcps/ MCP server binaries + configs |
| /opt/ac-secrets/ API keys vault (HashiCorp Vault) |
| /opt/ac-templates/ Agent instruction templates |
+---------------------------------------------------------------------+Five pod categories cover the entire Antigravity ecosystem — from SaaS products to agency operations to experimental micro-businesses.
Full autonomous growth engine. Ads, analytics, CRO, SEO, email, engineering, deploys. Already proven: 13 issues completed in one overnight run.
AEO/GEO visibility tracking SaaS. Agents run SEO audits, build content, track AI citations, manage the n8n pipeline, and optimize conversion.
Franchise marketing platform. Agents manage content, social, paid ads, franchise-specific landing pages, and PR coverage reports.
Manages 60+ clients across Anderson Collaborative and Nativz. Client health scoring, report generation, ad account monitoring, SEO tracking, billing alerts.
Dedicated pod for high-budget clients. Manages $2,500/mo Meta spend, lead form optimization, Zoho CRM sync, weekly reporting to Chris and Joseph.
The idea factory. Agents research markets, validate demand, build MVPs, run test ads, measure ROAS — and graduate winners to their own dedicated pod.
The difference between a script and an agent is learning. Every pod builds institutional knowledge over time through three interconnected systems.
Every completed task gets a retrospective entry. "Carousel ads outperformed static by 3x for restaurant clients." "Blog posts over 2,000 words rank 40% faster in this vertical." Agents read past retros before starting similar work. The KG grows with every heartbeat.
The orchestrator syncs key insights across pods. When the CS360 pod discovers that Stripe webhook retries need exponential backoff, that fix propagates to every pod running Stripe. Shared wisdom, isolated execution.
OpenRouter gives us 200+ models through one API key. The CEO agent routes tasks by complexity: Opus for strategy, Sonnet for code, Haiku for content, Gemini Flash for data parsing. Costs drop 40-60% without sacrificing quality where it matters.
If an agent's heartbeat fails 3x, the CEO agent reassigns the task and restarts it. If the CEO agent gets stuck, the orchestrator alerts you. Agents unstick themselves 90% of the time. You only hear about the 10%.
Agents can propose new skills and routines to a staging directory. You review and approve from the conductor. Over time, pods develop specialized capabilities unique to their domain. The CS360 pod becomes a Cloudflare deploy expert. The Agency pod becomes a client health scoring specialist.
Hard caps at daily, weekly, and monthly levels per pod. CPA thresholds trigger auto-pause. New campaigns require human approval. The system protects against runaway spend while giving agents freedom to optimize within bounds.
We don't need to build the Death Star on day one. Start with one pod, prove the model, then expand as each pod demonstrates value.
ac-pods CLI for conductor commands| Dimension | Today | With Agentic Pods |
|---|---|---|
| Uptime | Only when MacBook is open | 24/7/365 |
| Isolation | All projects share one machine | Each project in its own container |
| Learning | Lost between sessions | Persistent KG grows with every task |
| Cost per task | Everything runs on Opus | OpenRouter routes by complexity (40-60% savings) |
| Scaling | Limited by hardware | Add droplets in minutes |
| Recovery | Manual restart | Auto-restart + self-healing agents |
| Monitoring | Manual "let me check" | Automated health + alerts + daily digest |
| New project setup | Hours of configuration | ac-pods create --template marketing |
| Knowledge retention | In Trevor's head | In the Knowledge Graph, queryable forever |
The orchestrator provides real-time health data. Daily digests get emailed. Alerts fire on anomalies. You never have to wonder "is my stuff running?"
{
"pods": [
{
"name": "cs360",
"status": "healthy",
"uptime": "14d 6h",
"agents": { "active": 5, "stuck": 0, "idle": 1 },
"tasks_24h": { "completed": 12, "in_progress": 3, "stuck": 0 },
"budget": { "today": "$18.50 / $25", "month": "$320 / $500" },
"kg_growth": "+47 entities this week",
"learning": { "retros": 8, "skills_proposed": 1 }
},
{
"name": "rankprompt",
"status": "healthy",
"uptime": "10d 2h",
"agents": { "active": 3, "stuck": 0, "idle": 1 },
"tasks_24h": { "completed": 8, "in_progress": 2, "stuck": 0 },
"kg_growth": "+31 entities this week"
},
{
"name": "incubator",
"status": "idea_in_validation",
"current_idea": "BrandMythos",
"validation": {
"landing_page": "deployed",
"test_ads": "$42 spent, 2.3% CTR",
"signups": 14,
"cpa": "$3.00",
"target_cpa": "$5.00",
"verdict": "BEATING TARGET - recommend graduation"
}
}
]
}This infrastructure will underpin how we work for the next several years. Before we build it, we want to make sure we're thinking about it in the best, most future-forward way possible. Here's where we'd love your perspective:
The foundational decisions we need to get right:
How should the incubator pipeline work?
Making sure agents actually get smarter:
Keeping this sustainable:
Specific to AC + Nativz client work:
Getting the trust model right:
Build an infrastructure where high-agency AI teams come up with business ideas, validate them with real data, optimize customer acquisition, establish return on ad spend, and scale the ones that work — while we focus on the strategic decisions that actually need human judgment. The pods do the work. We set the direction.
Starting with one pod, proving the model, then scaling based on results.
| Phase | What | Timeline | Output |
|---|---|---|---|
| 1. Foundation | Docker + skills + MCPs on DigitalOcean droplet | Week 1 | Infrastructure ready for first pod |
| 2. CS360 Migration | Move all 6 agents, configs, and routines to cloud | Week 1-2 | CS360 running 24/7 on DO |
| 3. Orchestrator | Health aggregation, alerting, pod lifecycle API | Week 2 | Single dashboard for all pods |
| 4. Conductor CLI | ac-pods CLI on MacBook + Mac Mini |
Week 2-3 | Command any pod from any device |
| 5. Scale | Add RankPrompt, Agency Ops, Incubator pods | Week 3+ | Full ecosystem running autonomously |