Alpha Agi Business V1

preview

Current runnable path — 1.14.0

Mode: Offline sample. Ranks bundled business opportunities; service agents publish illustrative business events.

Prerequisites: Python 3.11–3.13; installed project dependencies.

python -m alpha_factory_v1.demos check alpha_agi_business_v1
python -m alpha_factory_v1.demos run alpha_agi_business_v1

The catalog command uses bundled inputs and explicit offline defaults.

Expected result: Best alpha opportunity and its sample score.

Scope: No company incorporation, registration, funding or actual trade is performed.

The catalog explains installation, stopping, backups and recovery. Browser charts for legacy demos are labeled sample replays. Original research narratives and advanced scripts below are preserved; they do not expand the tested scope stated here.

This repository is a conceptual research prototype. References to "AGI" and "superintelligence" describe aspirational goals and do not indicate the presence of a real general intelligence. Use at your own risk. Nothing herein constitutes financial advice. MontrealAI and the maintainers accept no liability for losses incurred from using this software. Each demo package exposes its own __version__ constant. The value marks the revision of that demo only and does not reflect the overall Alpha‑Factory release version.

Large‑Scale α‑AGI Business 👁️✨ ^$AGIALPHA

Proof‑of‑Alpha 🚀 — an autonomous business entity that finds, exploits & compounds live market alpha
using Alpha‑Factory v1 multi‑agent stack, on‑chain incentives & antifragile safety‑loops.

Important: This is a research demonstration. It simulates how an autonomous business could operate using Alpha‑Factory v1, but it is not a real or operational company. Use it only for experimentation and educational purposes.

✨ Executive Summary

Mission 🎯 Continuously harvest alpha across equities • commodities • crypto • supply‑chains • life‑sciences and convert it into compounding value — automatically, transparently, safely.
Engine ⚙️ Alpha‑Factory v1 👁️✨ → six specialised agents orchestrated via A2A message‑bus (see §4).
Vehicle 🏛️ A legally‑shielded α‑AGI Business (x.alpha.agi.eth) governed & financed by scarce utility token $AGIALPHA.
Result 📈 A self‑reinforcing fly‑wheel that Out‑learn • Out‑think • Out‑design • Out‑strategise • Out‑execute the market, round‑after‑round.

Quick Start bash python start_alpha_business.py # launch the orchestrator python openai_agents_bridge.py # expose via OpenAI Agents python gradio_dashboard.py # interactive dashboard See the Quick Start and Deployment Recipes sections for advanced options.

1 An α‑AGI Business? 🌐

Open financial & industrial alpha is shrinking 📉 — yet trillions in inefficiencies remain:

Mis‑priced risk in frontier markets
Latent capacity in global logistics
Undiscovered IP in public patent corpora
Cross‑asset statistical edges invisible to siloed desks

Hypothesis 🧩 Alpha‑Factory v1 already demonstrates general skill‑acquisition & real‑time orchestration. Pointed at live, multi‑modal data it surfaces & arbitrages real‑world inefficiencies continuously.

On-chain as <name>.alpha.agi.eth, an α-AGI Business 👁️✨ unleashes a self-improving α-AGI Agent 👁️✨ (<name>.alpha.agent.agi.eth) swarm to hunt inefficiencies and transmute them into $AGIALPHA.

2 System Blueprint 🛠️

flowchart LR
  subgraph "α‑AGI Business (x.alpha.agi.eth) 👁️✨"
    direction LR
    P(PlanningAgent)
    R(ResearchAgent)
    S(StrategyAgent)
    M(MarketAnalysisAgent)
    T(MemoryAgent)
    F(SafetyAgent)
    P --> S
    R --> S
    S --> M
    M -->|PnL + risk| F
    S --> T
    R --> T
  end

  subgraph Broker["Exchange / DeFi DEX 🏦"]
    E[Order Router]
  end

  Client((Problem Owner))
  Treasury(($AGIALPHA\nTreasury))

  Client -. post α‑job .-> P
  S -->|Orders| E
  E -->|Fills & Market Data| M
  F -->|Audit hash| Treasury
  Treasury -->|Reward release| Client

3 Role Architecture – Businesses & Agents 🏛️

α‑AGI Business - ENS: <sub>.alpha.agi.eth - Treasury: wallet holds $AGIALPHA; can issue bounties - Responsibilities: curate job portfolios, pool data/IP, enforce constraints - Value: captures upside from solved quests and reinvests

α‑AGI Agent - ENS: <sub>.alpha.agent.agi.eth - Treasury: personal stake (reputation + escrow) - Responsibilities: detect, plan & execute α‑jobs published by any Business - Value: earns $AGIALPHA rewards, boosts reputation, stores reusable templates

Legal & Conceptual Shield 🛡️ Both layers inherit the 2017 Multi‑Agent AI DAO prior‑art — a publicly timestamped blueprint for on‑chain, autonomous, self‑learning agent swarms, blocking trivial patents and providing a DAO‑native wrapper for fractional ownership.

4 Featured Alpha‑Factory Agents 🤖

Featured Alpha‑Factory Agents - PlanningAgent – MuZero++ task graph search; decomposes jobs and allocates resources (planning_agent.py). - ResearchAgent – Tool-former LLM with web and DB taps (research_agent.py). - StrategyAgent – Game-theoretic optimiser; crafts risk-adjusted playbooks (strategy_agent.py). - MarketAnalysisAgent – 5M ticks/s ingest; benchmarks edge vs baseline (market_analysis_agent.py). - MemoryAgent – Retrieval-augmented vector store (memory_agent.py). - SafetyAgent – Constitutional-AI and seccomp sandbox (safety_agent.py). - ExecutionAgent – Order routing and trade settlement (execution). - AlphaComplianceAgent – Regulatory checklist (alpha_agi_business_v1.py). - AlphaPortfolioAgent – Portfolio snapshot (alpha_agi_business_v1.py). All agents speak A2A protobuf, run on OpenAI Agents SDK or Google ADK, auto‑fallback to offline GGUF models — no API key required.

5 End‑to‑End Alpha Walk‑through 📖

ResearchAgent scrapes SEC 13‑F deltas, maritime AIS pings & macro calendars.
MarketAnalysisAgent detects anomalous spread widening in copper vs renewable‑ETF flows.
PlanningAgent forks tasks → StrategyAgent crafts hedged LME‑COMEX pair‑trade + FX overlay.
SafetyAgent signs‑off compliance pack (Dodd‑Frank §716, EMIR RTS 6).
ExecutionAgent routes orders to venues; fills + k‑sigs hashed on‑chain; escrow releases $AGIALPHA; live PnL feeds Grafana.
Best Alpha Example Using the bundled sample opportunities the top ranked item is “gene therapy patent undervalued by market” (score 88). Launching the demo with --submit-best automatically queues this opportunity for execution. Wall clock: 4 min 18 s on a CPU‑only laptop.

6 Quick Start 🚀

For a concise walkthrough see QUICK_START.md. For a deployment checklist aimed at production environments consult PRODUCTION_GUIDE.md.

git clone https://github.com/MontrealAI/AGI-Alpha-Agent-v0.git
cd AGI-Alpha-Agent-v0/alpha_factory_v1/demos/alpha_agi_business_v1

# easiest path – auto-installs dependencies and opens the docs

python start_alpha_business.py

# automatically queue the highest scoring demo opportunity
python start_alpha_business.py --submit-best

# Docker-based run (add --pull to use GHCR, --gpu for NVIDIA)
./run_business_v1_demo.sh [--pull] [--gpu]
# REST docs → http://localhost:8000/docs
# or run directly without Docker
python run_business_v1_local.py --bridge --auto-install
# expose orchestrator on a custom port
python run_business_v1_local.py --bridge --port 9000
# expose the Agents runtime on a custom port
python run_business_v1_local.py --bridge --runtime-port 6001
# automatically open the REST docs in your browser
python run_business_v1_local.py --bridge --open-ui
# Set `ALPHA_OPPS_FILE` to use a custom opportunity list
# ALPHA_OPPS_FILE=examples/my_alpha.json python run_business_v1_local.py --bridge

# Optional configuration
python scripts/setup_config.py
# Edit the `config.env` file to set variables such as:
#   - OPENAI_API_KEY
#   - YFINANCE_SYMBOL
#   - ALPHA_BEST_ONLY
#   - API_TOKEN (REST auth token, defaults to "demo-token" — change for production)
#   - MCP_ENDPOINT (optional Model Context Protocol URL)
#   - MCP_TIMEOUT_SEC (optional timeout in seconds for MCP network requests)
#   - AUTO_INSTALL_MISSING=1 to let `check_env.py` install any missing packages
#   - WHEELHOUSE=/path/to/wheels for offline package installs
# The launcher automatically picks up these settings.

> **Security Note:** `API_TOKEN` defaults to `demo-token` for quick demos. Replace it with a strong, unique value before
  any production deployment.

By default this launcher restricts `ALPHA_ENABLED_AGENTS` to the five
lightweight demo stubs so the orchestrator runs even on minimal setups.
Set the variable yourself to customise the agent list.

# the demo starts several stub agents:
#   • **IncorporatorAgent** registers the business
#   • **AlphaDiscoveryAgent** generates a short opportunity via the LLM provider
#     (logged via MCP when `MCP_ENDPOINT` is set)
#   • **AlphaOpportunityAgent** emits market inefficiencies from `examples/alpha_opportunities.json`
#     (override with `ALPHA_OPPS_FILE=/path/to/custom.json`)
#     set `ALPHA_TOP_N=N` to broadcast the top-N entries or
#     set `ALPHA_BEST_ONLY=1` to only emit the single highest-scoring one
#     and optionally `YFINANCE_SYMBOL=SPY` to pull a live price via `yfinance`
#     set `ALPHA_TOP_N=3` to publish the top 3 opportunities each cycle
#     or run `python examples/find_best_alpha.py` to print the current highest-scoring entry
#   • **AlphaExecutionAgent** converts an opportunity into an executed trade
#   • **AlphaRiskAgent** performs a trivial risk assessment
#   • **AlphaComplianceAgent** validates regulatory compliance
#   • **AlphaPortfolioAgent** summarises portfolio state
#   • **PlanningAgent**, **ResearchAgent**, **StrategyAgent**, **MarketAnalysisAgent**,
#     **MemoryAgent** and **SafetyAgent** emit placeholder events to illustrate the
#     full role architecture

open http://localhost:7860      # Dashboard SPA
./scripts/post_alpha_job.sh examples/job_copper_spread.json
# or
./scripts/post_alpha_job.sh examples/job_supply_chain_alpha.json
# or
./scripts/post_alpha_job.sh examples/job_forex_alpha.json
# or
./scripts/post_alpha_job.sh examples/job_execute_alpha.json

If dependencies are missing, pass --auto-install (and optionally --wheelhouse /path/to/wheels) to the local launcher:

python run_business_v1_local.py --auto-install --wheelhouse /path/to/wheels

Or open colab_alpha_agi_business_v1_demo.ipynb to run everything in Colab. Open in Colab The notebook now includes an optional Gradio dashboard (step 5b) so you can interact with the agents without writing any code. To drive the orchestrator via the OpenAI Agents SDK run python openai_agents_bridge.py (see step 5 in the notebook). Use --host http://<host>:<port> when the orchestrator is exposed elsewhere. If the script complains about a missing openai_agents package, install it with:

pip install openai-agents

In fully offline environments provide a local wheel via the WHEELHOUSE environment variable and run check_env.py --auto-install before launching the bridge.

💾 Offline wheel install

Create a wheelhouse on a machine with internet access:

mkdir -p /media/wheels
pip wheel -r requirements.txt -w /media/wheels
pip wheel -r requirements-dev.txt -w /media/wheels

Set WHEELHOUSE=/media/wheels and run the environment check to install from these local wheels. Use the same variable when running pre-commit or tests:

python check_env.py --auto-install --wheelhouse /media/wheels

For a concise reference see docs/OFFLINE_SETUP.md.

🎛️ Local Gradio Dashboard

For a quick interactive UI run python gradio_dashboard.py after the orchestrator starts. The dashboard exposes buttons to trigger each demo agent and fetch recent alpha opportunities without writing any code. It now also supports searching the orchestrator memory and fetching recent log lines for quick troubleshooting.

python gradio_dashboard.py  # visits http://localhost:7860

Set GRADIO_PORT to use a different port. The dashboard communicates with the orchestrator via its REST API (BUSINESS_HOST environment variable). Use --token YOUR_TOKEN or set API_TOKEN to authenticate requests.

🤖 OpenAI Agents bridge

Expose the business demo via the OpenAI Agents SDK (specify --host if the orchestrator runs elsewhere and --port to change the runtime port):

# default port 5001; customise via `--port` or `AGENTS_RUNTIME_PORT`
# wait up to 10s for the orchestrator (override with --wait-secs)
python openai_agents_bridge.py --host http://localhost:8000 --port 6001 --wait-secs 10
# → http://localhost:6001/v1/agents

Pass --open-ui to automatically open the runtime URL in your browser. Use --token YOUR_TOKEN or set API_TOKEN when the orchestrator requires authentication. When the optional google-adk dependency is installed and ALPHA_FACTORY_ENABLE_ADK=true is set, the same helper agent is also exposed via an ADK gateway for A2A messaging. Visit http://localhost:9000/docs to explore the gateway when enabled (default port: 9000). To use a custom port, set the GATEWAY_PORT environment variable accordingly.

Air‑gapped setup

The bridge requires the openai-agents package and optionally google-adk when ADK federation is enabled. Build wheels on a machine with internet access:

pip wheel openai-agents google-adk -w /media/wheels

Install from this wheelhouse and verify the environment before launching the bridge:

python check_env.py --auto-install --wheelhouse /media/wheels
WHEELHOUSE=/media/wheels python openai_agents_bridge.py --host http://localhost:8000

See PRODUCTION_GUIDE.md for detailed deployment tips.

The bridge exposes several helper tools:
list_agents
trigger_discovery
trigger_opportunity
trigger_best_alpha (send the highest scoring demo opportunity)
trigger_execution
trigger_risk
trigger_compliance
trigger_portfolio
trigger_planning
trigger_research
trigger_strategy
trigger_market_analysis
trigger_memory
trigger_safety
recent_alpha (retrieve latest opportunities)
search_memory (search stored alpha by keyword; parameters: query (string, required) and limit (integer, optional)) Example usage: bash curl -X POST http://localhost:6001/v1/agents/search_memory \ -H "Content-Type: application/json" \ -d '{"query": "market trend", "limit": 5}'
fetch_logs (return recent orchestrator log lines)
check_health (orchestrator health status)
submit_job (to post a custom job payload to any orchestrator agent)

For a programmatic example see examples/openai_agent_client.py:

python examples/openai_agent_client.py --action recent_alpha

No Docker? bash <(curl -sL https://get.alpha-factory.ai/business_demo.sh) boots an ephemeral VM (CPU‑only mode).

7 Deployment Recipes 📦

Target	Command	Notes
Laptop (single‑GPU)	`docker compose --profile business up -d`	≈ 250 FPS on RTX 3060
Kubernetes	`helm install business oci://ghcr.io/montrealai/charts/agi-business`	HPA on queue depth
Air‑gapped	`singularity run alpha-agi-business_offline.sif`	Includes 8‑B GGUF models

CI: GitHub Actions → Cosign‑signed OCI → SLSA‑3 attestation.

8 Security • Compliance • Legal Shield 🔐

Layer	Defence
Smart Contracts	OpenZeppelin 5.x · 100 % branch tests · ToB audit scheduled
Agent Sandbox	`minijail` seccomp‑bpf (read/write/mmap/futex)
Sybil Guard	zk‑license proof + stake slashing
Data Guard	Diff & ML filter vs PII/IP
Chaos Suite	Latency spikes, reward flips, gradient nulls
Audit Trail	BLAKE3 log → Solana testnet hourly
Legal Shield	2017 Multi‑Agent AI DAO prior‑art

Full checklist lives in docs/compliance_checklist_v1.md (17 items, pass‑rated).

9 Tokenomics 💎

Parameter	Value	Purpose
Total Supply	1 B `$AGIALPHA`	Fixed, zero inflation
Burn	1 % of each Business payout	Progressive deflation
Safety Fund	5 % of burns	Finances red‑team
Min Bounty	10 k tokens	Anti‑spam
Governance	Quadratic vote (√‑stake)	Curb plutocracy

Full econ model → docs/tokenomics_business_v1.pdf.

10 Antifragility & Self‑Improvement 💪

Alpha-Factory injects stochastic stressors (latency spikes, reward flips, gradient dropouts) at random intervals. The SafetyAgent & PlanningAgent collaborate to absorb shocks; metrics show ↑ robustness over time (see Grafana Antifragility panel).

Outcome: the Business benefits from volatility — the more chaos, the sharper its edge.

11 Roadmap 🛣️

Q2‑25 — Auto‑generated MiFID II & CFTC reports
Q3‑25 — Secure MPC plug‑in for dark‑pool nets
Q4‑25 — Industry‑agnostic “Alpha‑as‑API” gateway
2026+ — Autonomous DAO treasury & community forks

12 FAQ ❓

Do I need an OPENAI_API_KEY?

No. Offline mode auto‑loads GGUF models. If a key is present the Business upgrades itself to GPT‑4o tooling.

Can humans execute α‑jobs?

Yes, but agents usually outperform on cost & latency. Manual overrides possible via the dashboard.

Is $AGIALPHA a security token?

Utility token for staking, escrow & governance. No revenue share. Legal opinion in docs/legal_opinion_business.pdf.

13 License 📜

14 Resources 📚

OpenAI Agents SDK documentation
A practical guide to building agents
Google Agent Development Kit docs
Agent‑to‑Agent protocol
Model Context Protocol
Conceptual Framework
Best Alpha Workflow alpha_factory_v1/demos/alpha_agi_business_v1/colab_alpha_agi_business_v1_demo.ipynb

15 Local Checks

Run the standard checks from this folder before committing:

python ../../check_env.py --auto-install   # verify optional packages
pre-commit run --files <paths>             # format only the staged files
pytest -q ../../../tests                   # execute the root test suite

View README on GitHub