OpenClaw Manila · Show & Tell 2026

How Symph Became an
AI-Native
Agency.

A story about a dangerous question, one very useful Obsidian setup, and an AI named ARIA who now watches our server at 3am.

R
Raven Duran
CTO, Symph
The Starting Point

It started with a
dangerous question.

We're a software agency. We build things for a living. So naturally, we turned the lens on ourselves.

↓ The question that started it all
"What does the future of work look like?
And what does that look like at Symph?"
We looked at what we were already doing:
💻
Devs coding with agents
Copilot, Claude Code, cursoring around
📣
Marketing maxing content tools
Generating, scheduling, posting at scale
📊
Sales using MCPs to nurture leads
Context-aware outreach, smart follow-ups
But it all felt... a little fragmented. 😬
The Diagnosis

We had 6 different AI
relationships. It was complicated.

Everyone was doing something with AI. But the context lived in silos. The learning stayed local. Each tool forgot what happened in the last one.

🧱
Context amnesia
Every new chat starts from scratch. Decisions from last sprint? Gone.
🏝️
Knowledge silos
Dev knows dev stuff. Marketing knows marketing. They don't talk.
🔮
Tools, not teammates
AI that waits to be asked instead of proactively doing the work.
💡
One thing was clear though:
We ARE working with AI. We just needed to do it together, not separately.
The Aha Moment

Then Jarrhey
made a chart.
As COOs do.

Jarrhey built a Personal OS — Claude Code + Connectors + Obsidian. Context kept piling up. Learnings stacked on learnings. Taste, decisions, institutional knowledge.

Every day it got a little bit better. Skills adding skills. Memory compounding.

"What if the whole agency had this?"
The Context Problem diagram
💡 The spark that started it all
The Evolution

One man's Obsidian setup
became our agency's nervous system.

The idea evolved fast. Jarrhey's personal setup seeded something bigger — a platform where multiple agents work with humans, not for them.

🧠
Personal OS
Jarrhey + Claude Code + Obsidian
→
👥
Multi-agent ideas
Agents working together with humans
→
🏗️
Agent platform
Easy to create agents for anything
→
⚡
Agentic Agency
Humans orchestrate. Agents execute.
🎯
The dream: streamline the entire agency process into an agentic workflow.
The Platform

We needed a home.
Not another SaaS.

Building it as just another platform felt wrong. We wanted something we could truly customize — where the tools, dashboards, and pipelines could be shaped to exactly how we work.

Enter Symphony. ↓
📁
File-driven & fully customizable
Change any pipeline, dashboard, or workflow by editing a file
🤖
Can build itself
Just describe what you want. ARIA scaffolds, builds, and deploys it.
🔌
OpenClaw as the engine
Skills, memory, heartbeat, pipelines — all wired in
Symphony dashboard
Agents at Work

We gave our agents
names, roles, and
strong opinions.

David, Raven, Albert, Pixel, Cleo, Grace — each agent has a context, a skill set, and a task queue. You assign. They execute.

Real example: Pixel designed FB ads.
Grace wrote copy. Albert critiqued.
Zero meetings. Just results.
→ We have no regrets about this
AI Team agents Task board
The Multiplayer Part

ARIA is in every
conversation
simultaneously.

Han asks about Vector. Lyra wants an agency intro. David needs product summaries. ARIA handles all three at once — with full context on each.

Discord multiplayer
The Surreal Moment

Raven typed 4 words.
Two entities
figured out the rest.

"Hi @Aria, work with @Ralph."
— Raven, in Discord, with zero further instructions

ARIA had the agency context. Ralph had the SymphPay context. Together they scoped, planned, and integrated payment infrastructure into the app starter template. Raven watched.

Aria + Ralph collaboration
The Difference

Before vs. After

Spoiler: after is better
Before
AI resets every conversation
Each tool is its own silo
Manual job routing to the right person
Security = "we'll deal with it later"
One person, one task, full attention required
Institutional knowledge lives in Slack
→
After ARIA
Persistent memory, identity, and error logs
Unified context across all work
ARIA routes, spawns agents, monitors jobs hourly
Hourly process watchdog + integrity scanning
1 human directing an army of agents
Corrections system — correct once, never repeat
What We Built on OpenClaw

OpenClaw gave us
the foundation.
We built our home.

Skills, memory, scheduling — all there. We layered our agency-specific machinery on top.

55 custom skills. 14+ products in pipeline. Hourly heartbeats. Zero sleep required. 😴
💓
Heartbeat System
Cron-based ops loop — jobs, QA, security, costs, changelogs every hour
🧠
SOUL + MEMORY
Persistent identity, long-term memory, daily logs, error-log for self-correction
🎛️
Symphony Pipeline
Discovery → Design → Build → Market → Deliver, all agent-driven with human review gates
📐
Corrections System
COR-XXX corrections — one fix gets baked in forever. ARIA learns and never repeats it.
And the audacious part... 👀

We're building
our own OpenClaw.

A fully custom, AI-native platform built for the way Symph works.
🏗️ File-driven everything
🤖 Agent-first by default
🧠 Institutional memory built-in
🔌 Connects to anything
🌍 Built for an agency, usable by any team
"Agents do the work.
Humans make the decisions."
— That's the whole thing. Really.
Symph