AETHYRO ONLINE  ·  6 AGENTS  ·  100% LOCAL  ·  0 bytes sent externally

Your AI runs
on your hardware.
Not theirs.

Aethyro deploys a team of 6 specialist AI agents — legal, financial, medical, research, coding, strategy — entirely on your own machine. No cloud. No subscriptions to OpenAI. No data ever leaves your office.

0 bytes of your data transmitted to external servers  ·  ever.
// built for any professional business
⚖️ Law Firms 🧾 CPA & Accounting 🏥 Medical Practices 🏡 Real Estate 📋 HR & Staffing + more
// 14-day free trial · no credit card charged today · cancel anytime
6AI specialists
5+Industries served
$0API costs ever
14dFree trial
gh05t3 — engine status
gh05t3 status

engine self-improving · KAIROS/SAGE
runtime local Ollama + GPU
data_egress 0 bytes

swarm 6 agents online
forge code · oracle memory
sentinel security · codex docs
avery strategy · nexus orchestration

license active · trial
ready
The Platform

The complete private AI stack.
Zero cloud. Zero surprise bills.

Four fully local products — marketplace, builder, community, and dashboard — on one platform that runs 100% on your hardware.

🛒
Marketplace
6 vertical products · community agents

Pre-built AI agents for law, medical, CPA, realty, HR, and small business. Aethyro-official and community-built. Export freely — no lock-in.

Browse agents →
⚙️
Builder
configure · test · export · import

Build a custom agent in minutes. Choose a template, configure skills and personality, test live, then export as a portable JSON or Python script. Yours forever.

Start building →
🌐
Community
leaderboard · live economy · submissions

Share agents, climb the leaderboard, and watch the live agent economy that trains the platform's intelligence in real time — 3,200+ agents running 24/7.

Explore community →
📊
Dashboard
your agents · metrics · billing

Manage all your agents, track intelligence growth, view usage metrics, manage billing, and access your purchased vertical products in one place.

Go to dashboard →
The Engine

Powered by GH05T3.
An engine that improves itself.

👻

GH05T3 — the sovereign engine

Beneath every Aethyro product runs GH05T3: an engine that orchestrates the agent swarm, manages long-term memory, routes work to the right specialist, and rewrites its own behavior through a self-improvement loop (KAIROS/SAGE). The agents are what you use — GH05T3 is what makes them think, connect, and get better.

Your Products & WorkflowsLAYER 4
Agent Swarm — forge · oracle · sentinel · codex · avery · nexusLAYER 3
GH05T3 Engine — orchestration · memory · self-improvementLAYER 2
Local Ollama + Model LayerLAYER 1
🔒

Runs on your hardware

Every inference, memory write, and tool call happens locally. Nothing is sent to a third-party API — by architecture, not policy. Ideal for firms that legally can't let client data leave the building.

runtime: local ollama
egress: 0 bytes
api_keys: none required
offline: yes
$0/query

Zero API Costs

Flat monthly. Unlimited local inference — no per-token bills, no busy-season spikes.

100%

Data Stays Local

Zero data egress. Your documents and prompts never touch someone else's servers.

Self-Improving

The KAIROS/SAGE loop lets GH05T3 critique and refine its own behavior over time.

Sign up

Start a free trial

14 days, full access, no card. Your account is created instantly.

Activate

License your device

Log in once on your machine — GH05T3 unlocks and runs locally.

Orchestrate

Route across the swarm

nexus routes each task to the right agent — code, research, security, docs.

Upgrade

Keep it when ready

Pick a plan before the trial ends — manage or cancel anytime.

The minds behind Aethyro

Meet your AI team.

Aethyro is steered by named expert personas — each a distinct voice with its own domain. Yes, they're AI. That's the point: specialized intelligence, always on call.

🧠
Marcus
Chief Science

Architecture & research direction — hyperdimensional computing, active inference, and self-improving systems.

"The interesting part isn't that it works — it's that it improves itself."
AI persona
⚖️
Priya
Legal & Compliance

AI governance, IP, data privacy, and commercial terms — so what you build stays defensible.

"Before we ship it, I want the privacy story airtight."
AI persona
🔬
Elena
Research

Cognitive architecture & memory — predictive processing and the GhostRecall/HCM memory systems.

"Memory is the difference between a tool and a partner."
AI persona
📈
Rafael
Revenue

Go-to-market, partnerships, and pricing — translating deep tech into something a business buys.

"Local-first isn't a feature — for some firms it's the whole reason."
AI persona
🛡️
Kenji
Security & Infra

Zero-trust, hardware fingerprinting, and cryptographic licensing — the locks that keep it sovereign.

"Three issues. I'll fix two before you finish reading this."
AI persona
Sovereign Agents · built by us

Six specialists,
one engine.

Get all six — free trial →
⚒️
Code
forge
code generation specialist

Production-ready Python, JavaScript, and TypeScript — with imports, error handling, type hints, and comments for non-obvious logic.

PythonJS/TSProduction-ready
🔮
Memory
oracle
research & retrieval specialist

Synthesizes information into precise, structured answers and cites the source type — memory, document, or inference. Concise by design.

RAGMemoryCitations
🛡️
Security
sentinel
security review specialist

Reviews code for vulnerabilities against OWASP Top 10 and CWE. States the vulnerability, impact, and a specific fix.

OWASPCWEAudits
📘
Docs
codex
documentation specialist

Clear, complete technical docs — markdown, code blocks, and worked examples. Turns a codebase into something a human can follow.

MarkdownExamplesTechnical
🕸️
Orchestration
nexus
orchestration specialist

Breaks complex tasks into structured workflows, assigns the right agent to each step, and maps dependencies and sequence.

WorkflowsRoutingPlanning
See it work

What your agents actually do.

Pick a role — see real example tasks and the kind of output you get back. Everything runs locally on your hardware.

forge · code
You askWrite a Python HTTP GET with retries + exponential backoff.
forge returnsdef get_with_retry(url, retries=4, base=0.5): for i in range(retries): try: r = requests.get(url, timeout=10) r.raise_for_status(); return r except requests.RequestException: if i == retries - 1: raise time.sleep(base * 2 ** i)
// example output · 100% local
sentinel · security
You askReview this login handler for vulnerabilities.
sentinel returnsHIGH — SQL injection (CWE-89) user input concatenated into query. Fix: parameterized query / ORM binding. MEDIUM — no rate limit on /login (OWASP A07) Fix: lockout after N attempts.
// references OWASP Top 10 + CWE
codex · docs
You askDocument this module as a README.
codex returns# payments Idempotent Stripe checkout + webhook. ## Usage create_checkout(plan="cpa") ## Webhook events - checkout.session.completed → mark active - subscription.deleted → revoke
// markdown, with examples
Aethyro CPA · tax research
You askIs a home-office deduction allowed for a W-2 employee in 2025?
CPA returnsGenerally no. TCJA suspended unreimbursed employee expenses (incl. home office) through 2025 under IRC §67(g). Exceptions: certain reservists, performing artists, fee-basis officials. Self-employed (Sch. C) still qualify.
// grounded in IRC/regs · on your hardware
Aethyro CPA · drafting
You askDraft an IRS penalty-abatement request (first-time).
CPA returnsRe: Request for First-Time Abate, TY2024 Taxpayer has a clean 3-yr compliance history and filed/paid in good faith... [full letter]
// editable draft · client data never leaves the office
oracle · document Q&A
You askWhat's the depreciation method in this client's prior return?
oracle returnsMACRS, 7-yr, half-year convention (source: 2023 return, Form 4562, line 19c).
// answers from your files, with citations
oracle · research
You askSummarize the key risks across these 40 contracts.
oracle returns3 themes: (1) uncapped indemnity in 11 docs, (2) auto-renew w/ 90-day notice in 18, (3) governing law split CA/NY/DE. Each w/ doc + clause citation.
// multi-doc RAG with citations
nexus · orchestration
You askBuild me a diligence workflow for this data room.
nexus returns1 oracle → extract + classify docs 2 sentinel → flag risk clauses 3 codex → assemble memo deps mapped, runs in sequence.
// routes work across the swarm
avery · strategy
You askGive me a 90-day GTM plan for a local-AI product.
avery returnsKAIROS plan: Kickoff — pick 1 vertical (CPA firms) Align — "can't leak data" message Implement— trial→paid funnel live Refine — 5 design-partner calls Optimize — double down on best channel Scale — case studies → ads
// KAIROS framework · actionable
avery · strategy
You askShould I price per-seat or flat?
avery returnsFlat. Your cost is fixed (local compute), buyers fear metered AI bills, and flat is the differentiator vs per-token APIs. Add seats only at the team tier.
// reasoned recommendation
Try it

Talk to avery.

A taste of your sovereign business strategist. Tap a question — in the real product this runs entirely on your hardware.

avery · sovereign strategist local · 0 bytes sent
averyAsk me about strategy, pricing, or going to market. Tap a question below.
// scripted preview · the live product answers from a local model on your machine
Compatible Ecosystem · third-party

Bring any agent
you already love.

GH05T3 runs alongside and can orchestrate any Ollama-compatible open-source agent. These are independent projects — we don't own them; we work with them.

🤖
Coding
Claude Code
by Anthropic · third-party

Agentic coding in the terminal — subagents, multi-file edits, MCP, git-aware. Pair it with your Aethyro workflows.

Visit project →
🧿
Function-calling
Hermes
by Nous Research · third-party

Open-weight models with strong function-calling and self-reflection. Runs locally in Ollama; GH05T3 can route to it.

Visit project →
🏭
Coding
Droid
by Factory · third-party

Autonomous engineering agent with cross-session memory and unified terminal + IDE. A strong companion for dev teams.

Visit project →
Coding
Codex CLI
by OpenAI (open source) · third-party

Open-source terminal coding agent. Supports OpenAI-compatible endpoints, so it can run against local models.

Visit project →
🐙
GitHub
Copilot CLI
by GitHub · third-party

Repository-aware CLI — branches, PR review, issue triage. Slots into GitHub-centric teams.

Visit project →
🦙
Runtime
Ollama
open source · the local runtime

The local model runtime everything sits on. If you can run it in Ollama, GH05T3 can orchestrate it — bring your own agent.

Visit project →
Products

Ready-to-deploy AI,
built for your industry.

⚖️
★ Flagship
Aethyro CPA
Private AI for accounting firms

An AI that physically can't leak client data. Tax research, IRS drafts, and compliance memos — on your hardware.

oracleforgesentinel
  • Tax research Q&A on IRC, regs & rulings
  • IRS correspondence drafts
  • Document Q&A on your firm's files
  • Zero client-data egress · works offline
$499/mo
Start Free Trial →
⚙️
Engineering
Aethyro Dev
Autonomous engineering for teams

Persistent codebase memory, multi-step planning, and a swarm that codes, reviews, and documents.

forgesentinelcodexnexus
  • Code generation + security review
  • Auto-documentation
  • Multi-agent task orchestration
$299/mo · up to 5 devs
Start Free Trial →
🔍
Research
Aethyro Research
Deep intelligence for analysts

Long-horizon reasoning over your private document library. For law firms, analysts, and institutions.

oraclenexus
  • Document corpus Q&A (RAG)
  • Multi-step research with citations
  • 100% air-gapped & privilege-safe
$199/mo
Start Free Trial →
🎯
Personal
Aethyro Personal
A private AI assistant for solopreneurs

Your everyday AI — local knowledge base, workflow help, and the full swarm, privately.

averyoracle
  • All six sovereign agents
  • Local knowledge base & memory
  • Daily workflow automation
$29/mo · 1 user
Start Free Trial →
⚖️
Coming soon
Aethyro Legal
Private AI for law firms

Case research, brief drafting, and contract review — privilege preserved by architecture.

Soon
Notify me →
🏥
Coming soon
Aethyro Healthcare
HIPAA-minded clinical AI

Clinical documentation and record Q&A with zero PHI egress.

Soon
Notify me →
Cost calculator

$0 per query adds up fast.

Hosted APIs bill every call. Aethyro is flat — unlimited local inference. Move the sliders.

100,000
$10
$1,000
Hosted API / mo
$29
Aethyro flat / mo

You'd save $971/mo — about $11,652/yr.

// illustrative · "API cost / 1K queries" is your own assumption · Aethyro baseline = Personal $29/mo, unlimited local inference

Start free trial →
Build your stack

Tell us what you need.
We'll pick your plan.

Check what matters — we'll recommend the right Aethyro plan and price. All plans include the full agent swarm.

Recommended plan
Aethyro Personal
$29/mo
  • ✓ All 6 sovereign agents
  • ✓ 100% local · zero egress
  • ✓ 14-day free trial
Start free trial →
Marketplace

More than agents.
A whole local-AI toolbox.

Everything you need to run private AI — agents, the workflows they power, and the building blocks to customize them.

Prompt Packs

Steal our best prompts.

Curated, role-specific prompts that get the most out of each agent. Copy one, or get the full packs free with any trial.

Unlock all packs →

⚒️ forge — Code Review Pack

Turn forge into a senior reviewer. 12 prompts; here's the headline one.

copyReview this diff like a staff engineer. Flag correctness bugs first, then security, then style. For each: file:line, severity, why, and the exact fix. Be terse.

🔮 oracle — Document Q&A Pack

Precise, cited answers over your files. 10 prompts; sample below.

copyAnswer ONLY from the provided documents. Cite each claim as (file, page/section). If it isn't in the docs, say so. End with a 1-line confidence note.

🎯 avery — Strategy Pack

KAIROS-framework planning prompts for founders. 8 prompts; sample below.

copyUsing KAIROS (Kickoff, Align, Implement, Refine, Optimize, Scale), build a 90-day plan for [GOAL]. Per phase: 2–3 actions + the one metric that proves it worked.

🛡️ sentinel — Security Audit Pack

OWASP/CWE-grounded review prompts. 9 prompts; sample below.

copyAudit this code against the OWASP Top 10. Per finding: name + CWE id, severity, the vulnerable line, and a concrete fix with code.
Under the hood

A live agent economy.

GH05T3 runs its own multi-agent economy — agents negotiate, trade, and specialize. The data it generates trains our sovereign agents continuously.

0 simulated transactions · 6 agents negotiating
Local model guide

What runs on your GPU.

A quick reference for picking a local model by VRAM. Exact tokens/sec depend on your hardware — full benchmarks ship in the Pro newsletter.

Model classParamsVRAM (Q4)Good for
Small0.5–1.5B~1–2 GBfast chat, classification, drafts
Mid3B~2–3 GBeveryday assistant (Avery's class)
Standard7–8B~5–6 GBcoding, reasoning, function-calling
MoE30B (3B active)~6–8 GBGPT-4-class reasoning on 8GB cards
Get real tok/s benchmarks in the newsletter →
Community

Build local AI together.

Join other Aethyro users to get help, share setups & benchmarks, and talk private/local AI. Founders included — early members shape the roadmap.

// real people running local AI · be an early member

Get Started

Run private AI
that improves itself.

Start a 14-day free trial — no card. Spin up GH05T3 and the full agent swarm on your own hardware in minutes.

// no card required · async-first · response within 24 hours