Aetherra

A safety-first cognitive operating layer for agentic AI systems.

Version: 0.5.0-beta.0

Aetherra is an experimental AI operating layer for modular agents, persistent memory, policy-mediated execution, auditable decisions, and controlled self-improvement.

In the project architecture, Aetherra is the entity and Lyrixa is the persona. Lyrixa is how Aetherra presents itself to users; it is not a separate intelligence.

The project is in an internal Alpha/Foundation hardening phase. The current goal is not public release or unbounded autonomy. The current goal is to complete each core system until it is functional, testable, bounded, and safe enough to support reviewed internal alpha testing.

Maturity and Safety Notice

Aetherra is not production-ready autonomous infrastructure. Treat it as an active internal alpha-stage research and engineering codebase.

• Run it in a reviewed development environment.
• Keep Guardian and Security enforcement enabled for privileged paths.
• Do not connect it to sensitive systems without explicit review.
• Expect some systems to be functional foundations rather than final products.
• Prefer manual approval, rollback plans, and audit review for risky actions.

Current Priorities

1. Keep privileged actions mediated by Guardian and Security.
2. Prefer observation and proposal before autonomous modification.
3. Keep system behavior auditable through signed and structured records.
4. Complete one system at a time to a functional foundation standard.
5. Clean the repository so source, docs, tests, and generated artifacts are easy to tell apart.

System Status

System	Foundation status	Notes
Security	Functional complete	Capability checks, sandbox policy, signing, audit, and enforcement surfaces.
Guardian	Functional complete	Intent evaluation, risk assessment, approval, containment, audit integration, and performance policy.
Homeostasis	Functional foundation complete	Observation, diagnosis, recommendations, and Guardian-mediated controlled action paths.
Self-Improvement	Functional foundation complete	Proposal-only loop with Guardian review and execution delegation; no direct mutation in the foundation milestone.
Maintenance	Functional foundation complete	Umbrella loop for observe, diagnose, propose, review, approve, apply, verify, and learn.
Self-Incorporation, Memory, Kernel, Agent System, Aether Script, Integration Validation, Runtime UI, Lyrixa, Hub	Functional foundation complete	Current operating foundations with bounded responsibilities and documented safety limits.
Artificial Intelligence, Consciousness, Coding, Chat, AI Trainer	Functional foundation complete	Cognitive and intelligence foundations with governed readiness surfaces.

Primary system documents live in docs/, especially:

• docs/WHO_IS_AETHERRA.md
• docs/AETHERRA_SECURITY_SYSTEM.md
• docs/AETHERRA_GUARDIAN_SYSTEM.md
• docs/AETHERRA_HOMEOSTASIS_SYSTEM.md
• docs/AETHERRA_SELF-IMPROVEMENT_SYSTEM.md
• docs/AETHERRA_MAINTENANCE_SYSTEM.md
• docs/SYSTEM_INDEX.md

Architecture Shape

Aetherra is not a single chatbot, plugin runner, or automation script. It is a layered runtime with safety boundaries:

Lyrixa / Interfaces / Tools
        |
Agents, Plugins, Aether Script, Self-Improvement, Homeostasis
        |
Guardian policy and risk mediation
        |
Security enforcement, sandboxing, signing, and audit
        |
Kernel, Hub, service registry, memory, and runtime systems

Important design rule: Self-Improvement proposes. Guardian reviews. Security enforces. Execution is delegated only after policy allows it.

The core lifecycle is:

Observe -> Understand -> Propose -> Review -> Approve -> Execute -> Verify -> Learn

Quick Start

Requirements:

• Python 3.11+
• Git
• A virtual environment

Create and activate a virtual environment:

python -m venv .venv
source .venv/bin/activate

On Windows PowerShell:

python -m venv .venv
.\.venv\Scripts\Activate.ps1

Install dependencies:

python -m pip install --upgrade pip
python -m pip install -r requirements.txt

Run a basic smoke check:

python tools/os_smoke.py

Run focused safety and self-improvement tests:

python -m pytest -q -o addopts= \
  tests/capabilities/test_self_improvement_metrics.py \
  tests/capabilities/test_self_maintenance_services.py \
  tests/unit/test_hub_self_improvement_guardian.py \
  tests/unit/test_optimization_executor_guardian.py \
  tests/unit/test_selfinc_integration_guardian.py \
  tests/unit/test_selfinc_proposal_consumer.py

Run the Phase 5 validation harness quick profile:

python tools/phase5_validation_harness.py --profile quick --timeout 60

Start the local Hub:

python tools/run_hub_ai_api.py --port 3001

Then inspect:

curl http://localhost:3001/api/health
curl http://localhost:3001/metrics

Repository Map

Path	Purpose
Aetherra/	Primary package and runtime systems.
aetherra_hub/	Hub API blueprints, local services, and control-plane surfaces.
aetherra_coding/	Coding, analysis, verification, and orchestration utilities.
docs/	Active system documentation, guides, policy docs, and cleanup plans.
tests/	Unit, integration, capability, acceptance, and legacy standalone tests.
tools/	Verification, smoke, release, maintenance, and developer utilities.
scripts/	Operational and maintenance scripts.
requirements/	Dependency input files beyond the root requirements locks.
.github/workflows/	Focused CI, security, release, docs, and repository hygiene workflows.

The repository is still being cleaned. Tracked root-level files are being limited to public docs, package/config files, compatibility launchers, and current operational entry points. Generated runtime state, reports, databases, logs, coverage output, and packaged builds should stay ignored or be published as release artifacts instead of source.

Cleanup tracking is maintained in:

• docs/ACTIVE_SYSTEMS.md
• docs/AETHERRA_MASTER_MAP.md
• docs/MASTER_ROADMAP.md
• docs/ALPHA_READINESS_CHECKLIST.md
• docs/ALPHA_OPERATOR_RUNBOOK.md
• docs/ALPHA_READINESS_DECISION.md
• docs/REPOSITORY_STRUCTURE.md
• docs/REPOSITORY_CLEANUP_PLAN.md
• docs/ROOT_SCRIPT_WORKFLOW_TRIAGE.md
• docs/AETHERRA_FILE_MANIFEST.json
• docs/FILE_INDEX.md

Safety Model

Aetherra assumes powerful actions are dangerous until policy says otherwise.

Core rules:

• Self-Improvement must not directly modify the system during the foundation milestone.
• Guardian must evaluate privileged intent before execution.
• Security must enforce capabilities, sandboxing, signing, and audit policy.
• Homeostasis should recommend before taking controlled action.
• Rollback, containment, and approval paths are first-class requirements.
• Exceptions and execution payloads must be sanitized before reaching external clients.

This is an active research and engineering project. Do not run it as an unreviewed autonomous production system.

Development Rules

• Keep changes scoped to the system being completed.
• Prefer structured models and explicit policy decisions over ad hoc payloads.
• Add tests for safety boundaries, denial paths, and audit behavior.
• Keep generated logs, databases, build outputs, packaged distributions, and local state out of commits.
• Update the relevant docs/AETHERRA_*_SYSTEM.md document when a system's behavior or completion status changes.
• Avoid direct mutation loops. Proposal, review, approval, execution, and outcome learning should remain separate concerns.

License

Aetherra is licensed under the GNU General Public License v3.0 or later. See LICENSE, NOTICE, and COPYRIGHT.