Large‑Scale α‑AGI Business 👁️✨ ($AGIALPHA) Demo – “Infinite Bloom 3.0”

preview

Current runnable path — 1.36.0

Mode: Supplied-input planning. Reviews supplied observations and selects an exact effort-bounded portfolio of proposed work.

Prerequisites: Python 3.11–3.13; finite computation uses the standard library.

python -m alpha_factory_v1.demos check alpha_agi_business_2_v1
python -m alpha_factory_v1.demos run alpha_agi_business_2_v1
python -m alpha_factory_v1.demos.alpha_agi_business_2_v1.planning --input scenario.json

Expected result: Finite JSON evidence review and portfolio; bundled inputs are explicitly synthetic. Use planning --input FILE for supplied data.

Scope: Value points and source labels are supplied assumptions, not verified research, revenues or model output. Legacy service and explicit optional commentary require separate setup.

Original research narratives and flowcharts below are preserved as conceptual background. They do not expand the tested scope stated here.

Supplied-input evidence planning

Run one bounded, offline computation without starting the legacy service:

python -m alpha_factory_v1.demos.alpha_agi_business_2_v1.planning --example
python -m alpha_factory_v1.demos.alpha_agi_business_2_v1.planning --input scenario.json > report.json

The shell launcher now runs the same finite planner and requires explicit input:

bash alpha_factory_v1/demos/alpha_agi_business_2_v1/run_business_2_demo.sh --example
bash alpha_factory_v1/demos/alpha_agi_business_2_v1/run_business_2_demo.sh --input scenario.json

It preserves the caller's relative input path and does not build Docker images, install packages, start a server or configure credentials. PYTHON may select an already installed Python executable. Use the separately documented service command below for legacy topics.

The original notebook imports the installed planner, loads explicit example or supplied input, computes, exports an unsigned evidence file without overwriting, and recomputes recovered input. Its Python cells are tested offline with network connections blocked. Actual Colab deployment is unqualified; no automatic clone/install, provider setup or indefinite service wait remains in the executable walkthrough.

Copy example.json to start your own scenario. Change provenance.kind to supplied and describe your sources in provenance.note. The bundled example is synthetic. The command requires exactly one input option, prints JSON once and exits. Invalid input exits nonzero without a success report. The catalog runs the synthetic example offline.

Input contract

Schema: agialpha.business2.input.v1. Unknown fields, duplicate JSON keys, non-finite numbers, booleans in numeric fields, and files above 128,000 bytes are rejected. Numbers must be JSON integers, not decimals. IDs use 1–40 lowercase letters/digits/hyphens and start with a letter. All text rejects control characters and invalid Unicode. Source/metric/unit/title text is nonempty and at most 200 characters; provenance.note is at most 500 characters.

Root: schema, id, provenance, policy, observations, opportunities (all required)
provenance: kind (constructed or supplied), note
policy: effortHours (0–100,000), maxProjects (0–16), minSupportingSources (1–64)
observations: 1–64 entries with unique id, source, metric, unit, baseline, current, direction (increase or decrease). Baseline/current are nonnegative integers up to 1,000,000,000 in the same declared unit. Fractional measurements must be converted to a named integer unit.
opportunities: 1–16 entries with unique id, title, effortHours (1–100,000), valuePoints (0–1,000,000), evidenceIds (1–64 distinct, existing observation IDs)

Research computes current minus baseline. Only a strictly positive change in the requested direction supports an opportunity; neutral/opposite changes remain visible in the report. Support counts distinct stripped source labels, so multiple rows from the same label do not multiply corroboration. Labels are case-sensitive and unverified; two labels can still refer to the same underlying source. Conflicting observations remain visible rather than being silently removed. This simple support threshold is a screening rule, not a reliability score.

Planning enumerates at most 65,536 subsets of eligible opportunities. It maximizes supplied valuePoints under the effort and project-count limits, then minimizes effort, then uses lexicographically sorted IDs to break ties. Zero-value projects may be omitted. An empty portfolio is a valid outcome, including when budget or evidence is insufficient. Opportunities are independent; no hidden dependency, monetary value, causation or revenue model is inferred.

The report includes all observation deltas, source support, selection/exclusion reasons, remaining hours, subset count, provenance and SHA-256 input/report commitments. Commitments use sorted-key compact UTF-8 JSON; they bind supplied data, not its truth. Original array order is retained in the input commitment, while report rows are sorted by ID.

Legacy service and optional commentary

The original service keeps alpha.plan, alpha.research, and alpha.insight topics and the plan, summary, and insight envelope keys. Plan and summary now contain structured computed results. Set BUSINESS2_INPUT to a scenario path before starting the service; missing or invalid input fails rather than substituting fixtures. Each cycle rereads the file and includes its inputDigest, so consumers can correlate results when the file changes. Use atomic replacement when updating input. Existing API_TOKEN/NEO4J_PASSWORD and service prerequisites still apply.

BUSINESS2_INPUT=/absolute/path/scenario.json \
  python -m alpha_factory_v1.demos.alpha_agi_business_2_v1.alpha_agi_business_2_v1

Commentary is disabled unless BUSINESS2_COMMENTARY=1. The existing optional OpenAIAgent path is preserved (configured cloud key, otherwise its local Ollama endpoint), but the repository's compatibility stub is explicitly rejected. A separately installed genuine compatible SDK and provider are required. Failures propagate to the service's error handling; no fallback insight is published. Supplied input is sent to that provider only in this explicit mode. Commentary is labeled provider_commentary and is not independently verified or used in allocation. No live provider call is part of offline acceptance.

Historical concept narrative

The preserved narrative and flowchart below describe research ambitions. Claims about live alpha, automatic model upgrades, autonomous execution and monetary outcomes are not features of the tested supplied-input workflow above.

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.

Global markets seep trillions in latent opportunity — alpha in the broadest sense: pricing dislocations • supply‑chain inefficiencies • novel drug targets • policy loopholes • unexplored material designs.
Alpha‑Factory v1 turns that raw potential into deployable breakthroughs, autonomously.

⚡ TL;DR

# one‑liner demo ‑ works on any laptop
docker run -p 7860:7860 ghcr.io/montrealai/alpha-asi:latest
# open http://localhost:7860 → fully‑functional α‑AGI Business
# (offline‑first – no API key required, automatically upgrades if $OPENAI_API_KEY is present)

A single command launches the Orchestrator, seven specialised α‑AGI Agents, an antifragile safety shell and a friendly web dashboard.
Out‑of‑the‑box it discovers, validates and executes live alpha in any data‑rich domain.

Prefer a quick tour? Run the self-contained Colab Notebook to launch the orchestrator and try a live API call in minutes.

🏗️ System Overview

flowchart LR
  subgraph Alpha_AGI_Business
    direction TB
    MA[📊 MarketAnalysisAgent] --> O(Orchestrator)
    ST[🧠 StrategyAgent] --> O
    RE[🔍 ResearchAgent] --> O
    CG[🛠️ CodeGenAgent] --> O
    PL[🧮 PlanningAgent] --> O
    SA[🛡️ SafetyAgent] --> O
    ME[💾 MemoryAgent] --> O
    O -->|alpha insights| B(("α-AGI Business<br/>(*.alpha.agi.eth)"))
  end
  B -->|$AGIALPHA rewards| Users((Stake‑holders))

The orchestrator natively speaks A2A & OpenAI Agents SDK. If no external key is detected it seamlessly switches to local Llama‑3 GGUF weights (no outbound traffic). A lightweight LLMCommentAgent publishes a market comment each cycle when the SDK is available.

🏛️ Role Architecture – Businesses & Agents 🏛️

Entity	ENS Convention	Treasury	Responsibilities	How it Creates Value
α‑AGI Business	`<sub>.alpha.agi.eth`	Wallet holds $AGIALPHA & bounties	Defines Problem Portfolios (α‑jobs), pools data/rights, sets domain constraints	Aggregates high‑value problems, captures upside from solved portfolios, reinvests in new quests
α‑AGI Agent	`<sub>.alpha.agent.agi.eth`	Personal stake (reputation + escrow)	Detects, plans & executes α‑jobs issued by any Business	Earns $AGIALPHA rewards, grows reputation, evolves reusable alpha recipes

Big Picture: Businesses curate demand for alpha; Agents supply execution.
Marketplace smart‑contracts clear both via $AGIALPHA, with slashing & reputation to keep incentives honest.

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.

An α‑AGI Business 👁️✨ is a fully autonomous enterprise that identifies latent “alpha” opportunities across all industries and converts them into compounding value by out‑learning, out‑thinking, out‑designing, out‑strategizing, and out‑executing all competition.

🛡️ Legal & Conceptual Shield — both layers inherit the 2017 Multi‑Agent AI DAO public prior‑art, blocking trivial patents and providing a DAO‑first wrapper for fractional resource ownership.

🤖 Featured Agents (this demo)

Agent	Core Skill	Example Live Contribution
PlanningAgent	Goal‑decomposition & critical‑path search	Maps 12‑step route from raw SEC filings → trading strategy → executed orders
ResearchAgent	Web / doc intelligence & summarisation	Surfaces obscure FDA ruling that shifts biotech valuations
StrategyAgent	Game‑theoretic scenario planner	Monte‑Carlo sims to price carbon‑credit policy options
MarketAnalysisAgent	Real‑time quantitative signal miner	Spots 42 bp ETH‑perp funding mis‑price
CodeGenAgent	Secure tool execution & infra scaffolding	Auto‑generates production‑ready ETL with tests
SafetyAgent	Alignment, sandbox & red‑team	KL‑shield, seccomp jail, chaos‑testing
MemoryAgent	Retrieval‑augmented long‑term store	Surfaces best alpha playbooks on demand
LLMCommentAgent	Short market commentary via OpenAI Agents	Posts a one‑sentence outlook every cycle

🌸 “Infinite Bloom” – Unicorn‑Level Walk‑Through

Phase	Autonomous Agent Actions
1 Visionary Spark	investor.agent drafts yield objective → PlanningAgent explodes into 9 sub‑goals
2 Predictive Brilliance	MarketAnalysisAgent detects ETH‑staking basis; ResearchAgent validates macro context
3 Deal‑Making & Build	StrategyAgent designs delta‑neutral hedge; CodeGenAgent ships audited ERC‑4626 GardenVault
4 Adaptive Dynamics	TradingAgent auto‑rebalances; MemoryAgent archives best PnL shards; SafetyAgent chaos‑tests liquidity drain
5 Launch Blitz	marketing.agent triggers on‑chain airdrop; meme.agent seeds virality – TVL +200 % / 48 h
6 Self‑Improvement	Signal drift? Agents spin new strategy, vote on‑chain; upgrade shipped zero‑downtime
7 Unicorn & Beyond	Valuation crosses \$1 B inside weeks, illustrating how α‑AGI Businesses break conventional curves

🔥 Live Alpha Example – Today’s Opportunity

Detected by MarketAnalysisAgent during CI run (UTC‑now):

Cross‑venue ETH‑stETH basis = 38 bp (2 σ) → delta‑neutral carry trade.

Action Plan (autonomous):

buy stETH on Curve (0 bp fee via Negotiator rebate)
short ETH‑PERP on dYdX (funding –22 bp)
Hedge drift via 4‑hourly re‑balancer
Net yield ≈ 5.9 % APR, chain‑audited every epoch

(Full agent transcript in dashboard → Alpha tab)

🚀 Quick Start

Scenario	Command
Laptop (online)	`docker run -e OPENAI_API_KEY=$OPENAI_API_KEY -p 7860:7860 ghcr.io/montrealai/alpha-asi:latest`
Air‑gapped / offline	`docker run --env OFFLINE=1 -p 7860:7860 ghcr.io/montrealai/alpha-asi:latest`
GPU cluster (Helm)	`helm repo add montrealai https://ghcr.io/montrealai/charts && helm install agialpha montrealai/alpha-asi`

No key? No problem — the stack auto‑loads Llama‑3 8‑B GGUF weights and stays entirely offline.

git clone https://github.com/MontrealAI/AGI-Alpha-Agent-v0.git
cd AGI-Alpha-Agent-v0/alpha_factory_v1/demos/alpha_agi_business_2_v1
./run_business_2_demo.sh       # builds & launches locally

📦 Deployment Recipes

Target	How‑to	Notes
Laptop demo	`docker compose up`	CPU‑only, 4 GB RAM
Prod K8s	`helm upgrade --install agialpha montrealai/alpha-asi`	HPA on GPU > 70 %
Singularity	`singularity run alpha_asi_offline.sif --offline`	No network, checksums included

🔐 Safety & Compliance Highlights

KL‑Shield – policy stays within ε = 0.02 of constitutional baseline
seccomp‑jail – 4‑syscall allow‑list; 0/10 k fuzz escapes
Antifragile chaos‑testing – latency spikes, reward flips, gradient dropout → 82 % absorbed inside 4 M steps
EU AI‑Act Art 52 traceability – every A2A envelope hashed hourly to Solana testnet

💎 Tokenomics (excerpt)

Parameter	Value	Purpose
Supply	1 B `$AGIALPHA` (fixed)	Aligns Agents ↔ Businesses ↔ Users
Perf Fee	2 %	Funds R&D + Safety
Burn	0.5 % per tx	Anti‑inflation
Safety Fund	5 % of burns	Black‑swan cover

🛣️ Roadmap

Q2‑2025 — zk‑roll‑up micro‑harvests & real‑time DAO votes
Q3‑2025 — RWA corporate notes & carbon yields
2026+ — Regional blooms (APAC, LATAM) & VR garden shows

🖥️ Dev & Ops

GitHub Actions matrix (CPU / CUDA / ROCm) – build + test < 18 min
Signed container (cosign + in‑toto) – SLSA‑3 provenance
Prometheus / Grafana dashboards & OTEL traces included
Helm chart auto‑scales learner pod on GPU utilisation

📂 Repository Layout

alpha_factory_v1/
 ├─ backend/
 │   ├─ orchestrator.py
 │   └─ agents/            # PlanningAgent, ResearchAgent, ...
 └─ demos/
     └─ alpha_agi_business_2_v1/
         ├─ colab_alpha_agi_business_2_demo.ipynb
         └─ README.md      # ← YOU ARE HERE

📝 License

“Outlearn · Outthink · Outdesign · Outstrategise · Outexecute.”
Welcome to the era of Large‑Scale α‑AGI Businesses 🌸 — where autonomous swarms turn friction into alpha at planetary scale. 🚀

View README on GitHub

Large‑Scale α‑AGI Business 👁️✨ ($AGIALPHA) Demo – “Infinite Bloom 3.0”