Aether

Coastal Alpine Tech portfolio

Part of the Kiwi Edge AI Stack | Founder OS: NZ-Start-Up | Agent policy: .github/agent-fleet/

Sovereign hybrid edge AI for NZ farms and founders - local-first + multi-model, Te Mana Raraunga aligned - collaborating with Venture Taranaki, startups.com investors and Kotahitanga Investment Fund (HITL + cultural advisory for formal approaches).

See AI Infrastructure Leadership for our positioning as New Zealand's leader in sovereign edge AI Infrastructure.

Agents inform, draft, prepare, monitor, and remind. Humans advise, sign, file, send, and pay. Anti-hallucination policy: .github/agent-fleet/anti-hallucination.md | Congruence: CAT_CONGRUENCE.md

Problems we are solving

Aether is the sovereign agentic companion (including computer-use paths) with hard HITL ceilings for NZ builders.

1. Unconstrained agent hype - "Autonomous employees" create legal, tax, and cultural risk in NZ.
2. Tool sprawl - Founders and engineers need one companion that can draft and operate tools under policy.
3. Offshore default - Agentic stacks often assume US cloud identity and data paths.
4. Weak explainability - Operators need skill routing and human gates, not black-box actuation.

Solution we have built

Built capability	What it does
Agent orchestrator	ReAct-style loop + skill routing
Computer-use path	Desktop operation where explicitly enabled
Skill packs	Super Grok governance skills (aether-core, CAT standards, skills CI) + NZ Start-Up fleet under skills/nz-startup/
Licence	Proprietary (Coastal Alpine Tech) - internal companion across CAT products

Skills guide: docs/SKILLS.md

Policy slogan: Agents inform, draft, prepare, monitor, and remind. Humans advise, sign, file, send, and pay.

Local (Taranaki) and national (Aotearoa) economic benefits

Lever	Benefit
Regional R&D HQ	Product design and IP stay in New Plymouth / Taranaki - not only Auckland/offshore SaaS
Primary-sector productivity	On-farm and rural tools aim to cut waste, protect consents, and support export competitiveness
Skilled employment pathways	Edge install, field support, agritech ops, software, compliance, and cultural advisory roles as pilots scale
Data sovereignty	Te Mana Raraunga-aligned local custody keeps high-value operational data onshore
HITL jobs quality	Agents inform / draft / prepare / monitor / remind; humans advise / sign / file / send / pay - augment people, do not fake full autonomy

Stage honesty (pre-seed): Impact today is founder R&D, near-term contractors, and EDA/partner leverage. Permanent multi-region payroll follows paid pilots and revenue - we do not invent headcount claims.

Sovereign Agentic Development System

Aether is a culturally grounded, extensible agentic development orchestrator. It helps you plan, debug, scaffold, and execute development work using tools and reusable skills while keeping you in control through strong human oversight.

Coastal Alpine Tech Limited pre-seed startup, New Plymouth, Taranaki, Aotearoa New Zealand. Canonical edge target: Raspberry Pi 5 (16GB) with Hailo-10H NPU (40 TOPS). Local LLM via Ollama (qwen2.5-coder / Gemma 4 class models).

Architecture Overview

Diagrams: Architecture images and Mermaid maps describe the target product architecture for this pre-seed stack. They are engineering design maps not claims of large-scale commercial fleet deployment.

Aether is the sovereign agentic development orchestrator for the stack: ReAct loop over local tools and markdown skills, with HITL gates and optional Ollama (qwen2.5-coder / Gemma-class models) on developer or edge hardware.

System map

%%{init: {
 "theme": "dark",
 "themeVariables": {
 "fontSize": "15px",
 "fontFamily": "Inter, ui-sans-serif, system-ui, sans-serif",
 "primaryColor": "#0ea5e9",
 "primaryTextColor": "#f8fafc",
 "primaryBorderColor": "#38bdf8",
 "lineColor": "#67e8f9",
 "secondaryColor": "#1e293b",
 "tertiaryColor": "#0f172a",
 "clusterBkg": "#0b1220cc",
 "clusterBorder": "#38bdf880",
 "titleColor": "#e2e8f0"
 },
 "flowchart": {
 "nodeSpacing": 36,
 "rankSpacing": 44,
 "padding": 18,
 "htmlLabels": true,
 "curve": "basis",
 "useMaxWidth": true
 }
}}%%
flowchart TB

 classDef act fill:#422006,stroke:#fbbf24,stroke-width:2px,color:#fffbeb
 classDef core fill:#134e4a,stroke:#2dd4bf,stroke-width:2px,color:#f0fdfa
 classDef ai fill:#3b0764,stroke:#e879f9,stroke-width:2px,color:#fdf4ff
 classDef store fill:#1e1b4b,stroke:#a5b4fc,stroke-width:2px,color:#eef2ff
 classDef host fill:#0c4a6e,stroke:#38bdf8,stroke-width:2px,color:#f0f9ff
 classDef desk fill:#052e16,stroke:#4ade80,stroke-width:2px,color:#f0fdf4
 classDef stack fill:#312e81,stroke:#a5b4fc,stroke-width:2px,color:#eef2ff

 G["Goal / CI signal"] --> ORCH["AetherOrchestrator<br/>ReAct loop"]
 ORCH --> LLM["Ollama client<br/>JSON action contract"]
 ORCH --> SK["Skill loader<br/>skills/*/SKILL.md"]
 ORCH --> TL["File tools<br/>read | search | write | memory"]
 ORCH --> CU["Computer use hybrid<br/>screenshot | click | type | shell"]
 ORCH --> GR["Guardrails + threat model<br/>HITL gates"]
 LLM --> DEC["Validated decision"]
 DEC --> GR
 GR --> | approved | TL
 GR --> | approved | CU
 GR --> | halt / approve | HITL["Human approval"]
 TL --> MEM["JSONL memory / audit"]
 CU --> MEM
 SK --> ORCH

 subgraph HOSTS["Hybrid hosts one code path"]
 WIN["Windows 10/11<br/>install.ps1 | pyautogui"]
 LIN["Linux / RPi OS<br/>install.sh | X11/Wayland"]
 MAC["macOS optional"]
 end

 subgraph STACK["Kiwi Edge hybridisation"]
 CORE["Coastal-Alpine-Core skills"]
 WEA["Weaver multi-tenant mesh"]
 CAS["coastal-alpine-stack monorepo"]
 end

 ORCH -.-> HOSTS
 SK -.-> CORE & WEA & CAS
 CU -.-> WIN & LIN

 class G,HITL act
 class ORCH,SK,TL,CU core
 class LLM,DEC ai
 class GR,MEM store
 class WIN,LIN,MAC host
 class CORE,WEA,CAS stack

Loading

Layer	Components	Role
Loop	ReAct + tools + computer use	One action per step (files or desktop)
Skills	Markdown packs + Kiwi Edge skills	Domain procedures + stack architecture
Safety	Guardrails + skill HITL	Writes / desktop actuation gated by default
LLM	Ollama local (text + vision)	Offline-capable on Windows, Linux, RPi
Hosts	install.ps1	install.sh
Hybrid stack	Core	Weaver

Full detail: docs/ARCHITECTURE.md | docs/GETTING_STARTED.md

Quick Start

One-line install (recommended)

Linux / macOS

curl -fsSL https://raw.githubusercontent.com/fivepanelhat/Aether/main/install.sh | bash
aether doctor

Windows (PowerShell)

irm https://raw.githubusercontent.com/fivepanelhat/Aether/main/install.ps1 | iex
aether doctor

Note: If script execution is blocked: Set-ExecutionPolicy -Scope CurrentUser RemoteSigned

From a clone

Linux / macOS

git clone https://github.com/fivepanelhat/Aether.git
cd Aether
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[computer]" # or: pip install -e . for CLI only
aether init
aether --help
aether skills
aether run "Audit the API routes for security issues"

Windows (PowerShell)

git clone https://github.com/fivepanelhat/Aether.git
cd Aether
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -e ".[computer]" # or: pip install -e . for CLI only
aether init
aether --help
aether skills
aether run "Audit the API routes for security issues"

Prerequisites (both platforms): Python 3.10+, Git, Ollama for local models. On Linux desktop control also needs a display server (X11/Wayland) and often python3-tk / scrot depending on distro.

• Getting Started Guide: A complete guide on how to use Aether's ReAct loop and tools.
• CLI Reference: A detailed reference of all available CLI commands.

aether remediate

Trigger the error remediation workflow on a specific error or CI failure.

aether remediate "CI failed on main branch with test error in user.test.ts"

Computer Use Edge AI that operates your desktop

Aether now hybridises sovereign edge AI with computer use: a local (Ollama) vision model looks at screenshots and drives the real mouse, keyboard, and shell to accomplish goals entirely on-device. No screenshots or keystrokes leave the machine. Works on Windows and Linux (and macOS) from one code path.

Download & install (terminal, cross-platform)

Linux / macOS

curl -fsSL https://raw.githubusercontent.com/fivepanelhat/Aether/main/install.sh | bash

Windows (PowerShell)

irm https://raw.githubusercontent.com/fivepanelhat/Aether/main/install.ps1 | iex
aether doctor

Or from a clone, install the desktop extras alongside the base package:

pip install -e ".[computer]" # adds pyautogui + Pillow
aether doctor # verify Ollama + display + backend

The installers create an isolated virtualenv, expose the aether command on your PATH, install the bundled skills, and check for the local Ollama runtime. Pull a vision model once:

ollama pull qwen2.5-vl:7b # vision model for the agentic loop
ollama pull qwen2.5-coder:7b # text model for `aether run`

Agentic desktop control

# Vision loop: Aether observes the screen and acts to reach the goal.
aether computer run "Open the calculator and compute 12 * 9"
aether computer run "Rename the selected file to report_final.txt" --max-steps 15

# Authorize actuation without per-step prompts (batch / trusted contexts):
aether computer run "Tidy my downloads folder" --auto-approve

# Rehearse without touching the mouse/keyboard:
aether computer run "..." --dry-run

By default every actuating step (click, type, key, scroll, shell) pauses for human approval on a TTY, routed through the same guardrails/HITL layer that gates Aether's file writes. screenshot and screen_info are read-only and never gated.

Environment switches: AETHER_COMPUTER_DRY_RUN=1 rehearses without actuating; the vision model/host are overridable with --model / --base-url.

Skills

Skills are markdown packs under skills/*/SKILL.md, auto-loaded at runtime. Full catalogue: docs/SKILLS_CATALOG.md.

python -m aether.cli skills
# or, after install:
aether skills

Architecture & sovereignty (Kiwi Edge companion)

Skill	Role
kiwi-edge-architecture	System map: field -> MQTT -> Core -> Weaver -> portals -> Ollama/Hailo on RPi 5 16GB + Hailo-10H
security-notifications-triage	Dependabot / GHSA / CodeQL / audit response (HITL for high-impact)
te-mana-raraunga-sovereignty	Te Mana Raraunga 2018 data-sovereignty constraints

Error remediation

Skill	Role
error-remediation-orchestrator	Analyze failures and propose/apply fixes (HITL)
git-workflow	Branch, commit, push, PR (HITL)
ci-failure-parser	Structure CI / Actions logs
notification-responder	Status updates and approval requests

Trigger manually or via GitHub webhook remediation:

aether run "Apply kiwi-edge-architecture and Te Mana Raraunga checks to this Core PR"
aether remediate "CI failed on main with test error in user.test.ts"

Note: Git operations, code writes, and high-impact security changes require human approval by default.

Required Setup for Advanced Features

Some features (especially error remediation and git operations) require additional configuration:

GitHub Integration (Recommended)

To use the git-workflow skill effectively, you should have:

• A GitHub Personal Access Token (classic) with the following scopes:
• repo (Full control of private repositories)
• workflow (Update GitHub Action workflows)

How to create a token:

1. Go to GitHub -> Settings -> Developer settings -> Personal access tokens -> Tokens (classic)
2. Generate new token
3. Select the scopes listed above
4. Store the token securely (e.g. in a .env file or password manager)

Note: Aether currently expects you to handle git authentication via your local environment (SSH keys or credential manager). Token support can be added later.

GitHub Webhook Integration (CI Failure Auto-Trigger)

Aether can start investigation when your CI fails. Default is propose-only (read tools + plan; high-risk writes halt for HITL). Opt in to unattended writes with AETHER_WEBHOOK_AUTO_REMEDIATE=1.

1. Start the webhook server

python run_webhook.py
# or
aether webhook
# or on a custom port
aether webhook --host 0.0.0.0 --port 9000

2. Set your webhook secret (required verification fails closed without it)

export GITHUB_WEBHOOK_SECRET=your-secure-secret
# Optional local-dev only bypass (never in production):
# export AETHER_WEBHOOK_INSECURE=1
# Optional: authorize high-risk tool actions from webhooks (default off):
# export AETHER_WEBHOOK_AUTO_REMEDIATE=1

1. Expose the server using ngrok or deploy to a server

ngrok http 8000

4. Register the webhook in GitHub

• Go to your repo -> Settings -> Webhooks -> Add webhook
• Payload URL: https://your-url/webhook/github
• Content type: application/json
• Secret: your GITHUB_WEBHOOK_SECRET value
• Events: Select Workflow runs and Check runs

Webhook Retry Behavior

When a CI failure is received, Aether will attempt to trigger remediation up to 4 times using exponential backoff (2s -> 4s -> 8s -> 16s).

If all retry attempts fail, the error is logged but no further automatic action is taken. You can still trigger remediation manually:

aether run "Investigate CI failure in <repo> on branch <branch>"

Retry parameters are configurable via environment variables:

Variable	Default	Description
WEBHOOK_MAX_RETRIES	4	Maximum number of retry attempts
WEBHOOK_MIN_WAIT	2	Minimum wait between retries (s)
WEBHOOK_MAX_WAIT	30	Maximum wait between retries (s)

Future Integrations

• Email parsing (for inbox-based remediation)
• Slack / Discord notifications
• GitHub App (for deeper integration without PAT)

Project Structure

Aether/
|-- aether/
| |-- webhooks/ # GitHub webhook handler (FastAPI)
| |-- tools/ # Core tools (file_writer, codebase_search, etc.)
| -- orchestrator.py # ReAct loop + skill routing
|-- skills/ # Reusable skills (add your own here)
|-- docs/ # Documentation
|-- examples/ # Usage examples
|-- run_webhook.py # Start the webhook server
|-- pyproject.toml # Packaging configuration
-- README.md

Known Limitations

• Builder capabilities are still early (only the Project Scaffolder exists so far).
• Auto-remediation requires human approval at multiple steps for safety.
• Some advanced features (full auto-remediation, email triggers, etc.) are still in development.

We are actively working on improving generative (builder) capabilities.

License

Proprietary — © 2026 Coastal Alpine Tech Limited. All rights reserved. No open-source grant is implied by access to this repository; use is governed solely by LICENSE.

---

Built with focus on data sovereignty and edge intelligence. Coastal Alpine Tech Limited New Plymouth, Taranaki, New Zealand.