Step 01
Brief intake
Mojo (the orchestrator) receives the client brief via WhatsApp and structures it into a sprint plan.
Fig · Mojo
Live — running real client work
Mojo
A 7-agent AI production system that runs creative sprints end-to-end.
The problem
What was broken
Creative agencies move slowly. A typical production cycle — research, strategy, copy, production, localization — takes around 5 days. Most of that time isn't creative work. It's coordination, handoffs, status meetings, and rework.
The bottleneck isn't talent. It's orchestration.
The build
What I made
Mojo is a 7-agent orchestration system that coordinates the entire sprint. Each agent owns a discipline — research, strategy, copywriting, production direction, localization, ops — with its own knowledge base, protocols, and learning loop.
Built from scratch in 36 hours. Runs on OpenClaw. Coordinated through WhatsApp. Outputs to Google Drive. Tracked on Google Sheets. Designed with HITL gates so a human approves every major handoff.
Same sprint scope. Minutes instead of days.
How it works
The pipeline, step by step
Step 02
Scout runs deep research
Perplexity for narrative intel, Apify for structured data extraction.
Step 03
Strategist maps the thesis
Positioning, audience, and narrative thesis synthesized from Scout's findings.
Step 04
Writer produces the creative
Ad scripts, hooks, headlines, taglines — with a HITL gate before approval.
Step 05
Producer directs production
Production briefs and visual direction. Asset generation roadmap lands in Month 2.
Step 06
Babel localizes
Outputs adapted for additional markets and languages — 175+ supported.
Step 07
Ops packages delivery
Final deliverables packaged into a client-ready Google Drive folder. Every action logged.
Proof
The numbers
Build time
36 hours, idea to first live sprint
Sprint duration
Under 45 minutes vs a 5-day production cycle
Cost per sprint
Under $1
Agents in production
7 specialized agents + 1 orchestrator
Languages supported
175+ via the Babel agent
Visibility
Every action visible across WhatsApp + Google Sheets
The stack
What it runs on
Orchestration
- OpenClaw multi-agent framework
Models
- GPT-5.4 (1M context)
- Claude Sonnet
Web intelligence
- Perplexity Search API
- Apify (structured scraping)
Coordination
- WhatsApp Business API
Output layer
- Google Drive
- Google Sheets (via gog CLI)
Infrastructure
- Hostinger VPS (32 GB)
- systemd-managed
Markdown-first agent definitions, no vendor lock-in, full observability at every layer, and the whole thing runs on a single VPS for under $40/month.
What's next
Where this goes
Month 2 — Production tool integrations (Higgsfield, Freepik, HeyGen for asset generation), a single manifest file for the entire system, and n8n integration for deterministic ops workflows.
Month 3–4 — Vertical agent systems for adjacent verticals (lead generation, market research, outreach) using the same architecture.
The bigger bet: agent systems that can be packaged and deployed for specific verticals — small businesses, founders, indie creators — where Apify + AI + domain knowledge unlocks workflows that previously required full teams.
Sign-off
Want to build something like this?
Same operator. New tools. manishdwivedi9639@gmail.com