Borealis-Github-Replica/AGENTS.md

# Borealis Agents

## Overview
Borealis pairs a no-code workflow canvas with a rapidly evolving remote management stack. The long-term goal is to orchestrate scripts, schedules, and workflows against distributed agents while keeping everything self-contained and portable.

Today the stable core focuses on workflow-driven API and automation scenarios. RMM-level inventory, patching, and fleet coordination exist in early form; the server orchestrator and agent heartbeat are the critical pieces Codex should prioritize.

## Architecture At A Glance
- `Borealis.ps1` is the entry point for every component. It bootstraps dependencies, clones bundled virtual environments, and spins up server, agent, Vite, or Flask modes on demand.
- Bundled assets live under `Data/Agent`, `Data/Server`, and `Dependencies`. Launching installs copies into sibling `Agent/` and `Server/` directories so the development tree stays clean and the runtime stays portable.
- The server stack spans NodeJS + Vite for live development and Flask (`Data/Server/server.py`) for production APIs, backed by Python helpers (`Data/Server/Python_API_Endpoints`) for OCR, scripting, and other services.
- Agents run inside the packaged Python venv (`Data/Agent` mirrored to `Agent/`). `borealis-agent.py` handles the primary connection, with `agent_supervisor.py` and the PowerShell watchdog managing SYSTEM-level operations and resilience.

## Dependencies & Packaging
`Dependencies/` holds the installers/download payloads Borealis bootstraps on first launch: Python, 7-Zip, AutoHotkey, and NodeJS. Versions are hard-pinned in `Borealis.ps1`; upgrading any runtime requires updating those version constants before repackaging. Nothing self-updates, so Codex should coordinate dependency bumps carefully and test both server and agent bootstrap paths.

## Agent Responsibilities

### Communication Channels
Agents establish REST calls to the Flask backend on port 5000 and keep a WebSocket session for interactive features such as screenshot capture. Future plans include WebRTC for higher-performance remote desktop. No authentication or enrollment handshake exists yet, so agents are implicitly trusted once launched.

### Execution Contexts
`agent_supervisor.py` runs as `NT AUTHORITY\SYSTEM` via a scheduled task created during installation. `Scripts/watchdog.ps1` checks every five minutes that the supervisor stays alive and restarts it when needed. The primary `borealis-agent.py` process runs as the interactive user to cover foreground automation while delegating privileged work to the supervisor.

### Logging & State
All runtime logs live under `Logs/<ServiceName>` relative to the project root (`Logs/Agent` for the agent family). The project avoids writing to `%ProgramData%`, `%AppData%`, or other system directories so the entire footprint stays under the Borealis folder. Log rotation is not yet implemented; contributions should consider a built-in retention strategy. Configuration and state currently live alongside the agent code.

## Roles & Extensibility
- Role modules follow a `role_<purpose>.py` naming convention and should implement a configurable expiration window so the agent can abandon long-running work when the server signals a timeout.
- At present, roles aggregate inside `Data/Agent/agent_roles.py`. The desired end state is a `Agent/Borealis/Roles/` directory where each role lives in its own file and is auto-discovered or explicitly registered on startup.
- New roles should expose clear hooks for initialization, execution, cancellation, and cleanup. They must tolerate the agent being restarted, handle both Windows and (eventual) Linux paths, and avoid blocking the main event loop.
- Planned script execution split examples: `role_ScriptExec_SYSTEM.py` for privileged script or task execution and `role_ScriptExec_CURRENTUSER.py` for interactive scripts and tasks, keeping core orchestration in `borealis-agent.py`.

## Operational Guidance
- Launch or test a single agent locally with `.\Borealis.ps1 -Agent` (or combine with `-AgentAction install|repair|launch|remove` as needed). The same entry point manages the server (`-Server`) with either Vite or Flask flags.
- When debugging, tail files under `Logs/Agent` and inspect the watchdog output to confirm the supervisor is running. Use the PowerShell packaging scripts in `Data/Agent/Scripts` to reinstall scheduled tasks if they drift.
- Updates today require manually stopping related processes (`taskkill /IM "node.exe" /IM "pythonw.exe" /IM "python.exe" /F`) followed by a fresh run of `Borealis.ps1 -Agent`. This is a known limitation; future work should automate graceful agent restarts and remote updates without collateral downtime.
- Known stability gaps include suspected Python memory leaks in both the server and agents under multi-day workloads, occasional heartbeat mismatches, and the flashing watchdog console window. A more robust keepalive should eventually remove the watchdog dependency.
- Expect the agent to remain running for days or weeks; contributions should focus on reconnect logic, light resource usage, and graceful shutdown/restart semantics.

## State & Persistence
`database.db` currently stores device inventory, runtime facts, and job history. Workflow and scheduling metadata are not yet persisted, and no internal scheduler exists beyond WebUI prototypes. Planned scheduling work will need schema updates and migration guidance once implemented.

## Platform Parity
Windows is the reference environment today. `Borealis.ps1` owns the full deployment story, while `Borealis.sh` lags significantly and lacks the same packaging logic. Linux support needs feature parity (virtual environments, supervisor equivalents, and role loading) before macOS work resumes.

## Roadmap & Priorities
- Harden the agent core: modular role loading, reliable reconnect/keepalive, and watchdog replacement.
- Build inventory on demand (process lists, installed software, update metadata) and prepare for patch management workflows similar to commercial RMM tooling.
- Deliver the advanced scheduling matrix: workflows that trigger on timers or external API states, evaluate conditions, and fan out to script roles running as SYSTEM or the interactive user.
- Design a first-class update mechanism that can stage new agent builds, restart gracefully, and hot-detect new roles once they land on disk.
- Clean up deployment ergonomics so agents tolerate weeks of uptime without manual intervention and can accept hot-loaded role updates.

## Security Outlook
Security and authentication are intentionally deferred. There is currently no agent/server handshake, credential model, or ACL on powerful endpoints, so deployments must remain in controlled environments. A future milestone will introduce mutual registration, scoped API tokens, and hardened remote execution surfaces; until then, prioritize resilience and modularity while acknowledging the risk.