Case Study

A Multi-Agent System Running in Production.

A live deployment of the BNB architecture managing four coordinated agents — financial, marketing, content, and orchestration — across 22 scheduled jobs. Originally built as the founder's personal system, it's now the reference implementation we deploy variations of for every client engagement.

Below: the architecture, piece by piece, and what it actually looks like running.

01 · Foundation

An Agent Lives On A Computer That's Always On.

It runs on a VPS — a small virtual server in a data center. Always powered. Always online. Nothing fancy; the same kind of machine people use to host a blog.

For deployments with sensitive data or strict residency requirements, the same architecture runs on local servers, on-premise hardware, or air-gapped environments. The agent doesn't care where the computer lives — only that it's always on.

This is the environment where everything else happens. Think of it as the body the agent inhabits.

/home/oc/.openclaw/ · the file structure everything lives in
02 · The Agent

An Agent Is A Program That Wakes Up, Thinks, And Writes Output.

Not a chatbot. Not always listening (but always available). A scheduled program that starts when it's supposed to, reads its context, makes decisions using an AI model, and writes results to files.

The whole runtime is one command:

claude -p "Run morning briefing" --max-turns 15
03 · Identity

Every Agent Starts With A File That Shapes How It Thinks.

A markdown file called CLAUDE.md. The agent reads it at the start of every session — before acting on anything. It contains identity, values, rules, ways of thinking.

This particular agent is 731 lines of investment reasoning encoded as identity. The same pattern applies to any domain — operations, sales, compliance, content. The architecture is the asset; the identity file is what makes it yours.

this file IS the agent. replace it and you have a different agent.
04 · Memory

A Shared Folder Acts As Memory Across Wake-Ups.

Agents don't remember yesterday by default. So every run writes its work to shared markdown files — active state, open items, decisions, the last briefing.

Tomorrow's wake-up reads those same files and knows what the world looked like the last time it ran. Memory, simulated.

shared/PORTFOLIO_STATUS.md · shared/OPEN-THESES.md · shared/DECISIONS.md
05 · Senses

Data Feeds Are How The Agent Sees The World.

Five scripts run at 4:30 AM every weekday. They pull from specific sources — internal CRM data, regulatory feeds, public APIs, industry reports, partner integrations. Each writes a markdown file the agent reads during its morning wake-up.

The agent doesn't browse the web in real time. It reads a pre-digested snapshot. Fast. Predictable. Cheap.

5 feeds · 5 files · pulled fresh every weekday before markets open
06 · Voice

Telegram Is How The Agent Talks To Operators.

When the agent has something worth saying — a briefing, an alert, a question — it sends a Telegram message. The operator reads it on their phone like any other text.

No dashboard to check. No app to open. The agent comes to operators when it matters; otherwise, it's silent.

if you'd never see a dashboard anyway, why build one
07 · Heartbeat

Cron Is The Heartbeat. It Runs Everything On A Schedule.

Cron is a standard feature of every Linux system — it runs programs on a schedule. This system has 22 scheduled jobs across a week: a morning briefing at 6 AM, a market open alert at 6:35 AM, a close review at 1 PM, weekly deep dives on Thursday afternoons.

Predictable. Auditable. Cheap. If something misfires, we know exactly when and why.

22 scheduled jobs · the agent never forgets to run
VPS · /home/oc/ agent scheduled program wakes → reads → writes CLAUDE.md identity how it thinks shared/ memory PORTFOLIO_STATUS.md 5 feeds 4:30 AM daily CRM data regulatory feeds public APIs industry reports partner integrations Telegram to operator cron · heartbeat 22 scheduled jobs
How It Executes

Agents Start With Files.

At the start of every session, the agent reads these files in order. They define who the agent is, what it knows, and what it's allowed to do. Change the files, change the agent.

A Note On What Follows

The files below are from the live deployment we run for our own operations — configured for financial intelligence. We're showing them as-is because real production files are more credible than abstract examples. Read them at the architecture level: identity, memory, decisions, audit trails, schedule. The financial detail is what proves the system works in production. The pattern is what's portable to logistics, sales, compliance, content, or any domain where a process repeats and decisions follow rules.

CLAUDE.md · identity file
Identity

Who The Agent Is

Values, judgment, ways of thinking. The first file read at every wake-up. If the agent had a soul, this would be it.

# HOW IT THINKS - defense before offense (survive first, win second) - grade every thesis A through F (no "hold and hope") - 13F > insider > Congress (confirmation hierarchy) - conviction-weighted position sizing
PORTFOLIO_STATUS.md
State

What The Agent Owns Right Now

Live positions. Rewritten at every close review so tomorrow's agent knows where it stands.

ONTO · 150 sh FOUR · 50 sh CVS · 100 sh SGOV · $100k cash park
OPEN-THESES.md
Reasoning

Why The Agent Owns What It Owns

Thesis for every position. Entry rationale, kill-conditions, what would invalidate it. Read every morning so the agent can grade theses A through F.

ONTO · semiconductors · grade A entry: after earnings beat kill-if: sector thesis breaks stop: conviction-based
CONGRESS_TRADES.md
Sense · 1 of 5

What Congress Is Buying

Pulled fresh at 4:34 AM daily. Cross-referenced against the watchlist for signal. Refreshed from House Clerk + Senate disclosures.

# 172 filings parsed # 0 matched watchlist today # signals logged for later review
13F_TRACKER.md
Sense · 2 of 5

What Institutional Managers Are Doing

Quarterly holdings of tracked managers. New positions, increased stakes, full exits. Signals with fiduciary-grade conviction.

# 7 managers tracked # next filing cycle: mid-quarter # positions diffed vs last quarter
DECISIONS.md
Continuity

What The Agent Has Decided

Append-only log of every decision. When the agent explored X and chose Y over Z, it's here. Prevents the agent from re-litigating settled questions.

FOUR stop: hold — no tighten reason: enough ATR buffer through next earnings window
A Day Of Operations

The Heartbeat Is A Schedule.

Every tick is a scheduled job firing. Tap any dot to see what happened.

12 AM
6 AM
12 PM
6 PM
12 AM
04:30 AM · weekdays
data ingestion

Source Data Pulled From Origin

Deterministic shell scripts — not AI. Each pulls from a defined source (CRM, internal database, regulatory feed, public API) and writes a markdown snapshot the agents read later. Pre-digested input means agents make decisions on consistent, version-controlled data.

04:30 source_one.sh → SHARED/DOMAIN_INPUT_1.md 04:31 source_two.sh → SHARED/DOMAIN_INPUT_2.md 04:32 context_feed.sh → SHARED/CONTEXT.md 04:33 reference.sh → SHARED/REFERENCE.md 04:34 events_log.sh → SHARED/EVENTS.md
10 dots shown · 22 total scheduled jobs across the week · running whether anyone is watching or not
Live Output

This Is What It Looks Like Running.

Tail of the agent status log — real format, unedited.

tail -f /shared/status/AGENT_STATUS.md
[04:30:01] source_one.sh shared/inputs/DOMAIN_INPUT_1.md
[04:31:03] source_two.sh shared/inputs/DOMAIN_INPUT_2.md
[04:32:06] context_feed.sh shared/context/CONTEXT.md
[04:33:04] reference.sh shared/inputs/REFERENCE.md (QUIET)
[04:34:03] events_log.sh shared/inputs/EVENTS.md
 
Domain Agent | 06:05 PT | morning-briefing | COMPLETE
Briefing → shared/briefings/am.md
State reviewed: 47 items
Flagged for attention: 3
Auto-executions: 0
 
Domain Agent | 06:35 PT | threshold-check | COMPLETE
1 alert triggered → surfaced for operator approval
 
Environment Agent | 07:00 PT | external-scan | COMPLETE
3 items surfaced · 1 time-sensitive · 2 evergreen
 
Orchestrator | 07:30 PT | summary | SENT
Briefing delivered: "Morning state clear.
1 approval pending. Today's focus: <task>"
What This Means For You

The architecture is the asset.
The identity file is what makes it yours.

This system runs autonomously, 24/7, on infrastructure the operator owns. The same pattern adapts to operations agents (logistics, inventory), sales intelligence (pipeline management, lead research), compliance (document processing, regulatory monitoring), and any domain where a process repeats and decisions follow rules.

What changes per deployment
  • The identity file (CLAUDE.md) — your business logic, your rules
  • The data feeds — your sources, your APIs, your CRM
  • The integrations — your stack, your tools
  • The schedule — your cadence
What stays the same
  • The architecture (cron + identity + shared memory + integrations)
  • The security model (input validation, escalation boundaries, audit trails)
  • The principle: agents that execute, not chatbots that answer
  • Client-owned infrastructure and IP

If you want a system like this for your business, book a discovery call. No pitch. No pressure. We'll look at your operations and tell you honestly whether agents will move the needle.

Book a Discovery Call →
or
Send a quick inquiry instead