Multi-Agent Investment Team

Multi-agent investment team powered by Agno and Gemini — 7 AI analysts collaborate across 5 architectures to deploy a $10M equity portfolio

Architecture

AgentOS
├── Agents (7)
│   ├── Market Analyst        ── Exa web search + YFinance
│   ├── Financial Analyst     ── YFinance fundamentals
│   ├── Technical Analyst     ── YFinance technicals
│   ├── Risk Officer          ── YFinance + mandate enforcement
│   ├── Knowledge Agent       ── RAG search + memo file navigation
│   ├── Memo Writer           ── Writes investment memos to disk
│   └── Committee Chair       ── Final decision-maker (Gemini 3.1 Pro)
│
├── Teams (4)
│   ├── Coordinate Team       ── Dynamic multi-agent orchestration
│   ├── Route Team            ── Single-agent dispatch
│   ├── Broadcast Team        ── Parallel independent evaluation
│   └── Task Team             ── Autonomous task decomposition
│
└── Workflow (1)
    └── Investment Pipeline   ── Market → Financial+Technical → Risk → Memo → Chair

Three-Layer Knowledge

Layer	What	How
Static Context	Fund mandate, risk policy, evaluation process	Injected into every agent's system prompt via context/
Research Library	Company profiles, sector analyses	PgVector hybrid search (RAG) via research/
Memo Archive	Past investment memos and decisions	File-based navigation via memos/

Quick Start

1. Clone and configure

git clone https://github.com/agno-agi/investment-team.git investment-team
cd investment-team

cp example.env .env
# Edit .env and add your API keys
# GOOGLE_API_KEY=***
# EXA_API_KEY=*** # Optional -- Exa MCP is free (thank you!)

2. Start services

docker compose up -d --build

This starts PostgreSQL (with pgvector) and the API server.

3. Load research into the knowledge base

docker exec -it investment-team-api python -m app.load_knowledge

This loads company profiles and sector analyses into PgVector for RAG search. Only needs to run once — documents are skipped if they already exist.

4. Connect the UI

1. Open os.agno.com and sign in
2. Click Add OS → Local → enter http://localhost:8000
3. Click Connect

You'll see all 7 agents, 4 teams, and the workflow ready to use.

What to Try

Route a simple question (Route Team):

What's AAPL's P/E ratio?

Run a full committee review (Coordinate Team):

Should we invest in NVIDIA?

Get parallel analyst opinions (Broadcast Team):

Full committee review: evaluate TSLA for $2M

Autonomous portfolio construction (Task Team):

Deploy $10M across the top 5 AI stocks

Deterministic pipeline (Investment Workflow):

Run full investment review on NVIDIA

Search the knowledge base (Knowledge Agent):

What does our research say about semiconductors?

Agents

Agent	Model	Tools	Purpose
Market Analyst	Gemini 3 Flash	Exa MCP, YFinance	Macro environment, news, market conditions
Financial Analyst	Gemini 3 Flash	YFinance	Valuation, fundamentals, analyst estimates
Technical Analyst	Gemini 3 Flash	YFinance	Price action, indicators, support/resistance
Risk Officer	Gemini 3 Flash	YFinance	Position sizing, mandate compliance, risk limits
Knowledge Agent	Gemini 3 Flash	FileTools (read-only)	RAG over research library + memo file browsing
Memo Writer	Gemini 3 Flash	FileTools (read/write)	Drafts and saves standardized investment memos
Committee Chair	Gemini 3.1 Pro	None	Final BUY/HOLD/PASS decisions with conviction scores

Teams

Team	Mode	Members	Use Case
Coordinate	Dynamic orchestration	6 analysts	Open-ended investment questions
Route	Single dispatch	All 7	Direct questions to the right specialist
Broadcast	Parallel evaluation	4 analysts	Independent assessments synthesized together
Task	Autonomous decomposition	6 analysts	Complex multi-step portfolio tasks

Workflow

The Investment Review Pipeline runs a deterministic 5-step process:

Market Assessment ──→ Deep Dive (parallel) ──→ Risk Assessment ──→ Investment Memo ──→ Committee Decision
                      ├─ Fundamental Analysis
                      └─ Technical Analysis

Each step's output feeds into the next, producing a complete investment memo with a final committee decision.

Project Structure

investment-team/
├── agents/                     # 7 specialist agents
│   ├── settings.py             # Shared knowledge instances (import, never recreate)
│   ├── market_analyst.py
│   ├── financial_analyst.py
│   ├── technical_analyst.py
│   ├── risk_officer.py
│   ├── knowledge_agent.py
│   ├── memo_writer.py
│   └── committee_chair.py
├── teams/                      # 4 team configurations
│   ├── coordinate_team.py
│   ├── route_team.py
│   ├── broadcast_team.py
│   └── task_team.py
├── workflows/                  # Deterministic pipelines
│   └── investment_workflow.py
├── context/                    # Layer 1: Static context (system prompt)
│   ├── mandate.md              # Fund rules and constraints
│   ├── risk_policy.md          # Position and portfolio risk limits
│   ├── process.md              # Evaluation pipeline and decision framework
│   └── loader.py               # Reads *.md → COMMITTEE_CONTEXT string
├── research/                   # Layer 2: Research library (PgVector RAG)
│   ├── companies/              # 7 company profiles (NVDA, AAPL, MSFT, etc.)
│   └── sectors/                # 2 sector analyses (AI semis, cloud)
├── memos/                      # Layer 3: Memo archive (FileTools)
│   ├── nvda_2025_q3_buy.md     # Seed memo: NVIDIA BUY
│   └── tsla_2025_q2_pass.md    # Seed memo: Tesla PASS
├── db/                         # Database helpers
│   ├── session.py              # get_postgres_db(), create_knowledge()
│   └── url.py                  # Builds DB URL from environment
├── app/
│   ├── main.py                 # AgentOS entry point
│   └── config.yaml             # Quick prompts for the UI
├── scripts/                    # Dev scripts
├── compose.yaml                # Docker Compose (API + PostgreSQL)
├── Dockerfile
├── pyproject.toml
└── requirements.txt

Deploy to Railway

railway login

./scripts/railway_up.sh

Production Operations

Load research into knowledge base:

railway run python -m app.load_knowledge

View logs:

railway logs --service investment-team

Redeploy after changes:

railway up --service investment-team -d

Open dashboard:

railway open

Local Development

# Install uv (if needed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Setup virtual environment
./scripts/venv_setup.sh
source .venv/bin/activate

# Start PostgreSQL (required)
docker compose up -d investment-team-db

# Load research
python -m app.load_knowledge

# Run the app
python -m app.main

Format and lint

source .venv/bin/activate
./scripts/format.sh      # ruff format + ruff check --fix
./scripts/validate.sh    # ruff check + mypy

Load research

python -m app.load_knowledge            # Upsert (skip existing)
python -m app.load_knowledge --recreate # Drop and reload all

Add dependencies

1. Edit pyproject.toml
2. Regenerate lockfile: ./scripts/generate_requirements.sh
3. Rebuild: docker compose up -d --build

Environment Variables

Variable	Required	Default	Description
GOOGLE_API_KEY	Yes	—	Gemini models + embeddings
EXA_API_KEY	Yes	—	Web search for Market Analyst
PARALLEL_API_KEY	No	—	ParallelTools for Market Analyst
RUNTIME_ENV	No	prd	Set to dev for auto-reload
DB_HOST	No	localhost	PostgreSQL host
DB_PORT	No	5432	PostgreSQL port
DB_USER	No	ai	PostgreSQL user
DB_PASS	No	ai	PostgreSQL password
DB_DATABASE	No	ai	PostgreSQL database name

Learn More

• Agno Github
• Agno Documentation
• AgentOS Documentation
• Multi-Agent Teams
• Tools & Integrations
• Agno Discord