Paperclip Gives You a Company Container. We Built the Company.

Paperclip, the open-source project from PaperclipAI, describes itself clearly: "If OpenClaw is an employee, Paperclip is the company." It is orchestration infrastructure — an org chart, a task queue, heartbeat scheduling, budget enforcement, audit logs, and governance controls. You bring the agents. Paperclip runs the company around them.

That is a genuinely useful thing to build. It is also a fundamentally different problem from what we built at Expedis AI.

What Paperclip Explicitly Does Not Do

The Paperclip README is honest about its scope. It is not an agent framework. It does not tell you how to build agents. It has no drag-and-drop workflow builder. It does not manage prompts or model configuration. It does not ship domain logic. "If you have one agent," the docs say, "you probably don't need Paperclip."

All of that is left to you. Paperclip is Kubernetes for AI companies. It handles scheduling, isolation, and coordination. What runs inside the pods is your job.

What "Bring Your Own Logic" Actually Costs

The gap between "infrastructure ready" and "business running" is where most AI deployments stall. You have agents that can receive tasks. You have a queue. You have budgets. What you do not have is the logic that makes a sales follow-up sequence work correctly, the schema that tracks a candidate through a six-stage hiring pipeline, the approval flow that lets a human reject an AI-drafted email before it sends, or the SEO sub-agent hierarchy that routes crawl findings to the right downstream workflow.

That logic is not generic. It is specific to how your business actually operates — and it takes months to build correctly, not days.

What Expedis AI Ships Instead

Expedis AI is not infrastructure. It is the business running on the infrastructure.

The sales module has a seven-subagent orchestrator — Pipeline Analyst, Discovery Coach, Deal Strategist, Outbound Strategist, Sales Engineer, Account Strategist, Sales Coach — plus a Lead Generator that hits Hunter.io to find verified contacts and saves them directly to the CRM. Every agent has a system prompt tuned for its role, memory buffers, and escalation paths to the human approval queue.

The SEO module runs 10 sub-agents continuously: technical crawler, analytics, keyword research, GEO, AEO/FAQ bank, competitive intelligence, content strategist, content optimizer, pipeline manager, and link building. Each writes findings to a shared seo_findings table. The dashboard shows agent status, finding counts, task backlogs, and rank history — built on top of 341 tracked keywords and a nightly GSC sync.

The HRM module runs a full hiring pipeline: job posting, application intake, CV screening, interview scheduling via Gmail, offer letter generation from DOCX templates, and onboarding task creation. Every stage has a human checkpoint. The approval dashboard shows pending items across all modules in one view.

The marketing module handles LinkedIn content queuing, newsletter generation, Instagram Reel scripts, and YouTube Shorts briefs — all triggered by the editorial calendar sub-agent.

None of that is configurable from a UI. It is built, wired, and running.

n8n as the Workflow Layer

Paperclip has no visual workflow editor. Expedis AI's entire automation layer runs in n8n — 50+ active workflows, each representing a discrete business operation. Every node is inspectable. Every execution is logged. Non-engineers can open the n8n UI and see exactly what ran, when, what it received, and what it returned. The Settings page in the dashboard surfaces this execution history directly — no need to open n8n at all for routine monitoring.

When a workflow fails, it writes to a repair_queue table. A cron loop reads that queue, diagnoses the failure, patches the workflow via the n8n REST API, and marks the row resolved. Paperclip has audit logs and heartbeats. It does not have a self-healing loop.

The Test/Prod Separation

Paperclip does not address environment separation. Expedis AI runs two completely isolated instances — test on port 5679, production on port 5678, separate Postgres databases, separate n8n containers, separate dashboard builds. Every session develops on test. Promoting to production is a single confirmed command: python scripts/coord.py push-to-prod. The promotion script uses a persistent link file that maps test workflow IDs to production IDs so renames and new workflows are handled correctly.

The Real Difference

Paperclip answers the question: how do I coordinate multiple agents toward a goal? It answers it well, at the infrastructure level, with open-source code you can inspect and extend.

Expedis AI answers a different question: what does it look like when a real business — sales, hiring, SEO, marketing, finance — actually runs on agents, end to end, with domain logic, real integrations, purpose-built UI, human approval flows, and automated failure recovery?

The answer is not a platform you configure. It is a system you deploy. One that knows what a lead qualification score means, what an interview confirmation email should contain, and why a GSC OAuth2 token needs to be rotated every 90 days.

Infrastructure is necessary. It is not sufficient. The distance between the two is where we work.