Continuous Astra V5

Continuous Astra V5

A Python harness for a continuous AI assistant, with scheduled work, email, optional Discord, shared tools and parallel subagents. Includes the Aegis script runtime, reusable skills, a local prompt editor and an API spending guard.

Uses the hosted gpt-6-luna API. This repository contains the harness source; model weights are not included. Requires Python 3.11+ and your own service credentials.

Quick start

python -m pip install -r requirements.txt

Copy config.example.py to config.py and fill in your OpenAI and AgentMail settings. Keep config.py private. The original setting name AGENT_MAIL_ADRESS is retained for compatibility; environment variables can also supply credentials.

python astra.py

The package starts fresh: empty configuration, no contacts, memories, messages, logs or saved work. CONTACTS in astra.py is empty. Add only your intended contacts; configured contacts receive first-wake email confirmations. Ctrl+C stops the agent.

Optional features

  • Discord: add your bot token and two account usernames to config.py. Enable Message Content Intent for your bot. The second username is the operator; their messages trigger immediate wakes. Other approved messages wait for a wake.
  • Budget: python astra.py --api-budget 1 --no-prompt enables the local USD guard. Its pricing constants are dated October 4, 2026; update them for current provider pricing. This guard estimates usage and is not an account billing limit.
  • Prompt editor: python operator_dashboard.py opens a local editing service. Open the private link saved in tmp/operator_dashboard_url.txt; restart the agent after edits.
  • Docker: after configuring config.py, run docker compose up -d --build. Persistent runtime data lives in data/; use one deployment per inbox.

Scheduled wakes default to 20 minutes and can be set from 20 to 120 minutes. Email and Discord tools perform real actions when configured. Generic communication guidance is included; the distribution contains no private conversation examples.

Development

python -B -m unittest discover -p "test_*.py"

The offline suite uses fake services and temporary stores. See AEGIS.md for scripting syntax and PUBLIC_RELEASE.md for package changes and validation. SHA256SUMS.txt records the released file checksums.

The YAML header follows the Hugging Face card format. Upload this folder's contents as the repository root to keep the banner link working.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support