Overview

A reference AgentOS with 14 agents, 3 teams, and 5 workflows in one FastAPI service.

The Demo OS is a single AgentOS application that exercises every feature in the Runtime section. Use it as a working reference when building your own.

1git clone https://github.com/kern-agi/demo-os && cd demo-os
2cp example.env .env # set OPENAI_API_KEY
3docker compose up -d --build

Open os.kern.ndx.rocksAdd OSLocalhttp://localhost:8000.

What's inside

TypeCountExamples
Agents14Helpdesk (HITL + guardrails), Approvals (compliance), Contacts (memory), Studio (multimodal), Reasoner (reasoning + fallback)
Teams3Dash (text-to-SQL), Research (4 team modes), Investment (7-agent committee)
Workflows5Morning Brief, AI Research, Content Pipeline, Repo Walkthrough, Support Triage

All running inside one FastAPI service, sharing one Postgres for sessions, memory, knowledge, traces, and schedules.

Map: capability → demo

Each row points at the demo agent that exercises that capability.

CapabilityDemo agentPage
Hybrid search RAGDashRAG
MCP tool / context providerMCP agentMCP
Dependency injectionInjectorDependencies
Persistent learning across sessionsDash, Contacts, InvestmentLearning
Knowledge base with chunking & rerankingDashKnowledge
Entity memory + user profile + session contextContactsMemory
Session state with agentic updatesTaskboardSession State
HITL confirmation, user input, external executionHelpdesk, Approvals, FeedbackHuman-in-the-Loop
Multi-agent text-to-SQL with dual schemaDash teamText-to-SQL
Multi-agent investment committee with GeminiInvestment teamInvestment

Why a reference implementation matters

The docs explain what's possible; Demo OS shows it running. When the docs say "you can wire pre-hooks for PII detection," the Helpdesk agent shows the four-line config that does it. When the docs say "agents can update their own session state," the Taskboard agent is a 30-line working example. Every page in this section links straight to the source.

Run it locally

1git clone https://github.com/kern-agi/demo-os && cd demo-os
2cp example.env .env # OPENAI_API_KEY required, others optional
3docker compose up -d --build
4
5# Load Dash's SaaS dataset and knowledge
6docker exec -it demo-os-api python -m agents.dash.scripts.load_data
7docker exec -it demo-os-api python -m agents.dash.scripts.load_knowledge

Confirm at http://localhost:8000/health. Then open os.kern.ndx.rocksAdd OSLocalhttp://localhost:8000.

Deploy it to Railway

1railway login
2./scripts/railway_up.sh

The script provisions Postgres, configures env vars, and ships the app. Same pattern as the Scout, Dash, and Coda deploys.

Note

📷 Add screenshot: Demo OS connected to AgentOS UI showing the agent picker with all 14 agents listed.

Next

RAG →