Build a first prototype
Initialize a separate AxiomAO instance and demonstrate one bounded organizational cycle.
#Work in a separate copy
Use an uninitialized copy of the complete template, never a running organization or its state directory. The documented native-session profile needs Linux, Python 3.11 or later, Git and an installed, authorized client. The commands below assume an uninitialized Git template named axiomao-template exists in your current directory. They create a different directory for the experiment. Doctor checks local prerequisites; a positive result does not prove model access or successful delegation.
git clone ./axiomao-template ./research-ao
cd ./research-ao
python3 --version
git --version
python3 -m runtime doctor#Initialize only the experiment
Read the copied constitution, instructions and mandate before initialization. Set a narrow local purpose and keep publication, payments and external communication outside the experiment. The following commands create identity and state only inside research-ao, provided you remain in the copied directory. Initialization creates organizational records; it does not install a service or start a model. Do not reinitialize an existing organization to resolve an operational error.
python3 -m runtime init --name "Research AO" --goal "Review local research notes" --engine codex
python3 -m runtime status#Demonstrate one real work cycle
Open an authorized client session in the copied instance and request a small report from local material. The manager follows the instance instructions, obtains management ownership and delegates to real native workers and reviewers when available. Require an artifact, a separate verdict, acceptance evidence and a reply to the request. If the client cannot delegate, record that limitation. Creating worker rows in SQLite is not an alternative demonstration of native execution.
#Test continuity separately
End the first session with a checkpoint and the appropriate ownership release, then start another session in the same experimental instance. Check that identity, the accepted task, artifact version and next obligation remain consistent. This small test demonstrates a specific continuity path, not unattended operation, container isolation or general production readiness. Record the source revision, client version, exact observations and failures so someone else can distinguish your experiment from historical validation reports.
Sources & context
Reference edition 0.1 · 2026-10-14. Read the editorial method.