DisSysLab · Micro-Course

Build an office of small specialist agents.
Each agent has one job.

They watch your sources. They brief you. While you work, sleep, travel.

Situation Room — Live Live

CRITICALFed signals surprise rate cut — markets reacting

HIGHOpenAI announces GPT-5 release date

MEDIUMPython 3.14 beta drops with JIT improvements

Sense and respond

This is not a chatbot.

There is no prompt. There is no wait. The office watches your world and acts — while you work, sleep, or travel.

Chatbot

Ask me anything…▌

Waits for you to ask

DisSysLab office

📡 14:02 · News classified

📧 14:01 · Inbox triaged

🔔 13:58 · Alert sent

Acts on its own

A running situation room

An office of specialist agents. This is what you get — live, on your laptop.

Briefings — newest first

The complete office

Here's all you wrote.

office.md

Sources

bluesky(max_posts=None, lifetime=None)

al_jazeera(max_articles=10, poll_interval=600)

bbc_world(max_articles=10, poll_interval=600)

Sinks

intelligence_display(max_items=8)

jsonl_recorder(path="situation_room.jsonl")

Agents

Alex is an analyst.

Morgan is an editor.

Connections

bluesky's destination is Alex.

al_jazeera's destination is Alex.

bbc_world's destination is Alex.

Alex's editor is Morgan.

Alex's discard is jsonl_recorder.

Morgan's situation_room are intelligence_display and jsonl_recorder.

No code. Two plain English documents.

The job descriptions

Plain English. That's it.

analyst.md

You are a news analyst who receives posts and articles from social media and news sources. Your job is to assess whether each item is relevant to current world events. If relevant, send to editor. Otherwise send to discard.

editor.md

You are an editor who receives classified articles from the analyst. Your job is to rewrite each article as a concise briefing note with a priority rating. Send to situation_room.

What is an office?

An office of specialists.
That never clocks out.

Specialist agents. Cooperating. Continuously.

You describe the agents. DisSysLab handles everything else.

Sources

🦋BlueSky

📡Al Jazeera

📺BBC World

Agents

🔍

Alex

analyst

inbox

0

Watching sources…

↓

✏️

Morgan

editor

inbox

0

Ready…

Outputs

📊Dashboard

💾Archive

Connect any agent to any agent.

Many → One

📰 News

🦋 Social

📡 Feeds

🔍

Analyst

Fanin

One → Many

✏️

Editor

📊 Dashboard

💾 Archive

📧 Alerts

Fanout

How do you describe an agent?

Each agent does one thing.
Describe it in English.

analyst.md

You are a news analyst who receives

posts and articles from social media

and news sources.

Your job is to decide if each item is

relevant to current world events.

Send to editor if relevant.

Otherwise send to discard.

Who they are

What their job is

Where to send results

A specialist's job description

Job Description.
It's that simple.

analyst.md

You are a news analyst who receives posts and articles from social media and news sources.

Your job is to assess whether each item is relevant to current world events — politics, economics, science, technology, or humanitarian issues that a senior analyst would want to know about.

For each item, rate its importance: CRITICAL for breaking news requiring immediate attention, HIGH for significant developments, MEDIUM for notable but not urgent, LOW for background information.

If the item is relevant, send to editor with your importance rating and a one-sentence summary. If not relevant, send to discard.

Two ways to specify a job

English when AI can do the job.
Python when it must be exact.

Most agents are small enough to specify in English. For deterministic work — dedupe by URL, sliding-window RMS, wrap an ML model — use Python. The framework runs both the same way.

analyst.md

You are a news analyst who decides if each item is relevant to current world events.

If relevant, send to editor.
Otherwise send to discard.

deduplicator.py

def run(msg):
  url = msg["url"]
  if url in seen:
    return
  seen.add(url)
  send("out", msg)

Roles are reusable

A role is a job description.
Not a person.

🔍 analyst.md Role

"…if relevant, send to editor…"

used in 3 offices ↓

📰News Office

🔍

Alex

analyst

"Alex's editor is Morgan"

✏️

Morgan

editor

🏆Sports Office

🔍

Jordan

analyst

"Jordan's editor is Sam"

👔

Sam

CEO

🔬Science Office

🔍

Riley

analyst

"Riley's editor is Dana"

🛡️

Dana

security

📚 The analyst role is identical in all three offices.
Only the wiring changes.

The org chart

Name your agents.
Give each one a role.

office.md — Agents

Alexis ananalyst.

Morganis aneditor.

Miais aneditor.

Samis areporter.

Wiring the office

Write one line.
Make a connection.

office.md — Connections

"Morgan's situation_room are intelligence_display and jsonl_recorder ."

✏️

Morgan

📊

intelligence_display

💾

jsonl_recorder

The complete office

The whole office.
One screen. Plain English.

office.md

Sources

bluesky(max_posts=None, lifetime=None)

al_jazeera(max_articles=10, poll_interval=600)

bbc_world(max_articles=10, poll_interval=600)

Sinks

intelligence_display(max_items=8)

jsonl_recorder(path="situation_room.jsonl")

Agents

Alexis ananalyst.

Morganis aneditor.

Connections

bluesky's destination isAlex.

al_jazeera's destination isAlex.

bbc_world's destination isAlex.

Alex's editor isMorgan.

Alex's discard isjsonl_recorder.

Morgan's situation_room areintelligence_display and jsonl_recorder.

Mix and match

Pick the AI per agent.

Cheap models for routine work. Premium models for the writing. Free local models when privacy matters. One office can use all three.

office.md

Eve's AI is ollama.🦙 free, local

Sam's AI is openrouter.🚦 cheap cloud

Riley's AI is claude.🧠 highest quality

You've got this

Three things.
That's all you write.

Sources and sinks — where data comes from, where results go — you pick from a library. No writing needed.

1

📋

Roles

"You are a news analyst who receives posts and decides if they're relevant…"

2

👥

Agents

"Alex is an analyst.
Morgan is an editor."

3

🔗

Connections

"Alex's editor is Morgan."

That's everything.

Not a DAG

Loops welcome.

Many frameworks force a one-way pipeline. DisSysLab does not. An agent can feed its upstream, feed itself, or wait for a downstream verdict before continuing.

Reviewer can send work back to Drafter. A DAG can't.

Checkpoint and resume

Snapshot. Restart. Continue.

Take a checkpoint while the office is running. After a crash, a reboot, or a code fix, restart from the last snapshot. No work is lost.

Your turn

Install it.
Two minutes.

terminal

$ python -m venv venv $ source venv/bin/activate $ pip install dissyslab Collecting dissyslab Downloading dissyslab-1.1.1-py3-none-any.whl ... Successfully installed dissyslab-1.1.1 anthropic-0.39.0 ... $ dsl --help Usage: dsl [COMMAND] Commands: list, init, run

One pip install and every office in the gallery is yours. Works on Mac, Linux, and Windows.

Pick one

See what's already built.

terminal — dsl list

$ dsl list Available offices (use dsl init <name> <folder> to copy one): periodic_brief Morning HTML brief: news + weather + tickers situation_room Multi-source news → enrichment → digest arxiv_radar Daily arXiv papers → LLM rater → digest kalshi_market_watch Prediction markets → LLM briefing backyard_birds Audio → BirdNET classifier → species labels wildlife_watcher Images → ML classifier → labeled species weather_monitor Hourly plain-English weather briefing stocks_monitor One-line read of a ticker's movement recovery_demo Monte Carlo π estimator with checkpoint-recovery

▶ GIF slot — capture dsl list live in a fresh venv

Copy it

Make a folder that's yours.

terminal

$ dsl init weather_monitor my_weather Copied weather_monitor → ./my_weather Ready. Next: cd my_weather && dsl run . $ cd my_weather

my_weather/

my_weather/ ├── office.md # the wiring ├── README.md # what it does └── roles/ ├── analyst.md # job description └── editor.md # job description

This folder is yours. Edit any file. Run it when you want.

One-time setup

Add your API key.

my_weather / .env

ANTHROPIC_API_KEY=sk-ant-api03-xxxxxxxxxxxxxxxxxxxxxx # Paste yours on the line above — no quotes.

Get a key (free trial)

Sign up at console.anthropic.com and create an API key. A few test runs cost under $1.

Keep it private

The .env file is gitignored. Don't paste your key into shared chat, code, or screenshots.

Press play

Run it.
Watch it work.

my_weather — dsl run .

$ dsl run . Starting office: my_weather Compiled network: 2 agents, 1 source, 1 sink Running… press Ctrl+C to stop.

MEDIUMforecast.openweather07:02

Clear skies, 58°F / 14°C. Warming through mid-70s by afternoon.

Morgan: Pleasant day. Light jacket in the morning, short sleeves after noon.

LOWforecast.openweather07:07

No precipitation expected in the next 24 hours.

Morgan: Dry day — good for outdoor plans. UV index 6 (high) from 11am–3pm.

HIGHforecast.openweather07:12

Wind gusts up to 32 mph expected tomorrow afternoon.

Morgan: Secure loose outdoor items. Driving conditions may worsen near 3pm.

▶ GIF slot — capture dsl run . streaming for 30 sec in a fresh venv

Make it yours

Change a prompt. Re-run.

Edit a role file in plain English. Run the office again. Watch the output change.

Before — default role

roles/analyst.md

You are a weather analyst.

Write a short, neutral forecast summary.

Send to editor.

MEDIUMClear, 58°F. Warming to mid-70s.

After — you rewrote one line

roles/analyst.md

You are a weather analyst.

Write a cheerful forecast a runner would appreciate.

Send to editor.

HIGHPerfect running weather — 58°F and clear. Cool start, warming by mile 3.

You're ready.

Build something that actually works for you — an office that watches for exactly what you care about.

📈

Stocks you ownprice swings + one-line take

🔬

New research papersarXiv filtered to your topic

📬

Important emailstriage the inbox while you sleep

🌐

A page you care aboutupdates when something changes

📦 Install from PyPI ⭐ GitHub 📖 Docs

Build an office of small specialist agents.Each agent has one job.

This is not a chatbot.

Here's all you wrote.

Plain English. That's it.

An office of specialists.That never clocks out.

Connect any agent to any agent.

Each agent does one thing.Describe it in English.

Job Description.It's that simple.

English when AI can do the job.Python when it must be exact.

A role is a job description.Not a person.

Name your agents.Give each one a role.

Write one line.Make a connection.

The whole office.One screen. Plain English.

Pick the AI per agent.

Three things.That's all you write.

Loops welcome.

Snapshot. Restart. Continue.

Install it.Two minutes.

See what's already built.

Make a folder that's yours.

Add your API key.

Run it.Watch it work.

Change a prompt. Re-run.

You're ready.

Build an office of small specialist agents.
Each agent has one job.

An office of specialists.
That never clocks out.

Each agent does one thing.
Describe it in English.

Job Description.
It's that simple.

English when AI can do the job.
Python when it must be exact.

A role is a job description.
Not a person.

Name your agents.
Give each one a role.

Write one line.
Make a connection.

The whole office.
One screen. Plain English.

Three things.
That's all you write.

Install it.
Two minutes.

Run it.
Watch it work.