Skip to content

Project notice

Start here — get a useful, reviewable result

MONTREAL.AI flagship: Find your Alpha. Build your SUCCESSOR Ω.

Start the flagship → · Démarrer le parcours phare →

Choose a valuable opportunity in software, research or mathematics. Prepare a Computer Use mission to investigate real public material and produce a concrete artifact. Discover and combine strategies at Level 3, then test a better problem formulation at Level 4 while preserving the required outcome and acceptance criteria. Keep the proposed value, attempts, observations, reviews and lessons in one project. Planning and the catalog example run free in your browser; model-driven execution uses a separately configured native runner. Follow the flagship guide for the guided path and current scope.

For bounded transformational discovery: open the free offline workspace or follow the native source-pilot guide. It preserves the required artifact, compares alternative methods and retains reviewable evidence. The implementation register states the structural and research limits.

See what was actually demonstrated: Read the computer-use study or open its English/French visual reader. Two repaired websites, tested recovery and fewer scripted setup controls; simulated owner, not a production benchmark.

For a maintained local example: open the catalog engineering tool. Reconcile supplied or permitted CSVs, inspect the results and retain the tool. It runs without an account, installation, model API or wallet; the flagship opportunity mission above is the main starting point. The simple planning walkthrough remains available in English and French with larger text. The instructions below are for installing and operating your own agent.

Starting a Nova-Seed project? Follow the English first-use guide to prepare, encrypt and keep a small project before choosing any wallet or paid service. The guide also explains the installed executor and the separate execution, review and payment decisions.

Vous préférez le français? Le guide de première utilisation explique comment préparer et sauvegarder un projet, puis utiliser la console locale. La préparation ne demande ni portefeuille ni service payant; chaque autorisation reste une décision distincte.

Alpha-Factory runs bounded missions against inputs you supply, retains a signed journal, and asks you to inspect the result before approval. Start with the allocation example; it needs no wallet, API key, Docker or model download. The factory guide explains all five mission types and the complete Ascension path. Browser-only examples are available in Decision Studio.

For self-generated reasoning tasks and independent solver review, open the Curriculum Lab.

For the new two-generation mission lifecycle, open the bilingual SUCCESSOR Ω workspace or follow its first-session guide. Once installed, run alpha-agent --home successor-state successor-demo --output successor-evidence. It constructs and measures real bounded challengers, preserves usable methods and tests clean restoration. Missing independent qualification remains HOLD; a software demonstration grants no production authority.

Find your Alpha. Build your SUCCESSOR Ω. Follow the guided project journey to retain discovery, bounded work, separately priced USDC / $AGIALPHA proposals and Fleet local returns in one history. The browser guides explicit file handoffs; funding and independent commissioning remain separate.

Current Alpha contracts: use the October 18 address directory and retained deployment evidence. The historical October 11 wallet profile remains for its existing obligations; a current address reference does not reroute a wallet or authorize funding.

1. Install one release

Use Python 3.11, 3.12 or 3.13. From one release, download these four assets into the same folder:

  • The alpha_factory_v1-…-py3-none-any.whl wheel.
  • requirements-agent.lock.
  • SHA256SUMS.
  • install_agent.py.

Open a terminal in that folder and run:

python3 install_agent.py --release-dir . --venv .venv-agent

On Windows use python in place of python3. If several Python versions are installed, select a supported interpreter explicitly, for example python3.12.

The installer checks the wheel and dependency lock against the release checksums, creates a new environment, installs the hashed dependencies and verifies the installation. Do not mix assets from different versions. If the environment path already exists, choose a new path; your existing state is retained. For an offline install, add --wheelhouse /path/to/compatible-wheels. Download and extract alpha-agent-v…-operator-guide.zip for all the linked guides in one folder.

To check downloaded files before installation, add --check-only. It validates checksums and paths without creating an environment. The repository guide maps the other launch paths.

Activate the environment on macOS or Linux:

source .venv-agent/bin/activate

On Windows PowerShell:

.\.venv-agent\Scripts\Activate.ps1

If PowerShell blocks activation, call .\.venv-agent\Scripts\alpha-agent.exe in place of alpha-agent; activation is not required. Confirm alpha-agent --version matches the release you downloaded. A checksum error or installation failure is a reason to stop and fix the inputs, not to reuse a partially installed environment. See first-run troubleshooting.

2. Run and inspect the example

alpha-agent examples --output my-missions
alpha-agent --home ./agent-state init
alpha-agent --home ./agent-state run my-missions/allocation.json
alpha-agent --home ./agent-state serve

Open http://127.0.0.1:8765 and use the token in agent-state/api.token. Select the returned mission. Inspect its selected projects, budget, risk and totals, then approve or reject the exact result. These example inputs are constructed scenarios; edit their assumptions before using a result for a real decision. Approval records your review and does not execute a trade or make a payment. Stop the console with Ctrl+C.

3. Continue with your own work

  • Edit the copied examples: research, allocation, scheduling, forecasting or code. Code evaluation requires explicit permission and an isolated Docker runtime.
  • Use the factory guide for reviewed native work, FusionPlans and signed delivery.
  • Explore workflow alternatives in the MATS Search Lab, then export a reproducible review bundle.
  • Use the demo walkthrough for all browser experiences and the preserved local experiments.
  • Use operations and recovery before upgrades, backup, restore or provider configuration.
  • Use the Ascension protocol guide for the local-EVM enterprise lifecycle.
  • Use the contract deployment and migration guide to inspect the deployment plan and rehearse the disposable Hardhat fixture. It is separate from installing this agent.
  • Inspect the release's release-manifest.json for exact acceptance results and its validation_evidence inventory. The original alpha-agent-v…-validation.zip retains local checks; separate public-validation packs retain deployed-site checks. Keep their manifests, parts and matching SHA256SUMS together. The source, browser, complete site and unchanged manuscript are separate release assets.

If a command fails, keep the existing environment and journal. Check the JSON error, supported Python version and selected profile; follow the recovery guide using a new destination. Never manually edit signed journal records. The maintained runtime, exact deployed-source record and later contract candidates are distinct from the broader research vision; release readiness defines the verified scope.