Skip to content
Features

Everything it does. All of it on your machine.

Reges combines a camera that knows you, hands that move windows, a voice that answers and an agent that acts. Here is each part, exactly as it works today.

  • 316

    agent tools and actions

  • ~150

    skills in the library

  • 36

    personas

  • 27

    connectors

01Vision

It sees you.

Enrol your face once (“Reges, remember me”) and it greets you by name. Ask what it sees and your local vision model describes the frame.
  • Owner recognition from a face embedding matched against the enrolled owner, on this PC.
  • A name only appears once it wins 3 of the last 5 detections, so recognition does not flicker.
  • Frames go to the local vision model only when you ask, through an API that answers to this machine alone.
Reges vision marking a person it has not met as UNKNOWN
Vision: it finds faces in real time and only recognises its owner. Strangers stay unknown.
02Hands

It follows your hands.

Two hands, 21 landmarks each, every frame, from an ordinary webcam. Floating windows, the clock, the globe and your folders follow them.
  • Pinch to grab and drag; in a list, pinch to scroll.
  • Pinch and release to tap: open a folder, press a button.
  • Two-hand pinch to resize: spread apart to make it bigger.
  • Pinch and fling to throw a window off screen and close it.
  • Two modes: windows floating over the dashboard, or the full neural interface. Ctrl+Shift+V toggles it; the mouse works too.
The Reges vision window tracking a hand held up to the camera
Pinch to grab · pinch-tap to open · two hands to resize · fling to close.
03Voice

It talks back.

Say “Reges” and ask. Speech recognition and the voice both run on your PC, so your microphone is never streamed anywhere.
  • Wake word “Reges”, in one breath or with a pause (“Reges?” … “Yes?”).
  • After it answers, follow-ups need no wake word for 12 seconds.
  • Whisper speech-to-text runs in a background worker; replies are spoken by a local Kokoro voice.
  • Talk over it and it stops to listen. A live level meter shows the microphone really hears you.
Reges first-meeting screen over the command center: Hello, I am Reges
First meeting: Reges introduces itself, then learns its name, personality, voice and what to help with.
04Agent + tools

Then it does the work.

A chatbot answers. Reges acts, with 316 tools: its core agent tools, the actions of its own server and the tools each connector adds.
  • Email over IMAP/SMTP: triage, drafts and replies (sending waits for your yes)
  • Document search with sources, over the folders you choose
  • Knowledge graph, notes, journal and Obsidian vaults
  • Tasks and plans
  • MCP servers
  • Give it an objective and it plans the steps as tasks you can follow.
  • Every action has a risk level (read, write or confirm). Confirm actions never run without your explicit yes.
Reges Connectors panel listing services with their status and tool counts
Connectors: your store, inbox, repos and servers, each showing exactly what it still needs.
05Skills + personas

About 150 skills. 36 personas.

Skills are step-by-step playbooks Reges follows for a job, from clearing an inbox to checking a deploy. Personas change how it talks and what it focuses on.

A few of the skills

  • daily-planning
  • inbox-zero
  • meeting-notes-actions
  • research-brief
  • decision-memo
  • proposal-writing
  • invoicing-and-collections
  • cash-flow-forecast
  • customer-support-triage
  • lead-follow-up
  • gdpr-request-handling
  • data-breach-response
  • phishing-triage
  • secret-leak-response
  • deploy-verification
  • backup-restore-drill
  • local-model-selection
  • rag-ingestion
  • release-notes-draft
  • sop-writing
  • translation-locale-parity
  • weekly-business-report

A few of the personas

  • Executive assistant
  • Research librarian
  • Senior engineer
  • Operations manager
  • Finance controller
  • Customer support lead
  • Legal intake clerk
  • Security analyst
  • Tutor
  • Thinking partner
  • Copywriter
  • Front-desk concierge
06Connectors

27 connectors. Your accounts, your keys.

Each connector uses your own credentials, stored on your machine. Anything that posts, sends, deletes or deploys waits for your yes.

AI

  • Local models

    LM Studio, llama.cpp, Ollama or vLLM: list, benchmark, load and unload models, and inspect GGUF files.

  • Hugging Face Hub

    Search models and datasets, read model cards and download files into your own models folder.

  • ComfyUI

    Queue image generations on your own ComfyUI server and read the history.

  • Mind

    The assistant’s diary, mood, knowledge gaps and secret scrubber. Built in, nothing leaves the machine.

  • OpenRouter

    Optional, cost-capped cloud models, used only when you ask and only after personal data is scrubbed.

  • Providers

    Keeps track of every model backend Reges can call, local or cloud, so the router can pick the right one.

Business

  • Desk

    A built-in local business desk: contacts, deals, activities, orders, notifications and inbox triage.

  • Meta

    Facebook Pages and Instagram Business: posts, comments and messages. Publishing waits for your yes.

  • Discord

    Read and post through your own bot or a channel webhook. Posting and moderation wait for your yes.

  • Tebex

    Your webstore for game servers, through the store’s own Plugin API key.

Developer

  • Engineer

    Devin-style engineer mode: sandboxed git worktree, a test loop and a gated GitHub PR when it is done.

  • GitHub

    Repositories, issues, pull requests, releases and Actions runs, including GitHub Enterprise Server.

  • Vercel

    Projects, deployments and build logs; redeploy, promote and roll back.

  • Coolify

    Your self-hosted PaaS: applications, services and deployments. Env values never leave Coolify.

  • Hostinger

    Websites, domains, DNS zones and managed Node.js apps through the Hostinger API.

  • Deploy verify

    Proves a site is really up after a deploy: HTTP, TLS and DNS checks from your machine.

  • Release gate

    One call answers “can this folder ship?” for Node and Python projects.

Everyday

  • Documents

    Reads, extracts and OCRs PDFs and documents in the folders you choose.

  • Knowledge

    A local knowledge graph and notes search over your own files and Obsidian vaults.

  • Everyday

    Quick calculators and everyday lookups: unit conversion, dates and simple maths.

  • Life admin

    Public holidays and other everyday life-admin lookups.

  • Money info

    Currency conversion and financial reference lookups.

  • Weather

    Local weather and short-range forecasts for planning your day.

  • Customer help

    Customer-support triage and FAQ answers drawn from your own documents.

  • Home Assistant

    Control and query your Home Assistant smart-home devices.

Game servers

  • Cfx server

    FiveM and RedM server health, config linting and security checks.

  • Cfx natives

    The public RedM and FiveM native databases, with a checker for scripts.

07Self-learning memory

It gets better every day.

Learning here is not model training. After each conversation Reges keeps durable facts worth remembering and brings them into every request, all on your disk.
  • Facts are kept per subject, capped at 12 each: the least important or oldest is replaced, near-duplicates merge, and a contradicted fact is replaced.
  • Everything passes a secret scrubber first, so a pasted API key, token, card number or password is never remembered.
  • At 03:00 each night it consolidates, logs the day and writes a short diary entry with the knowledge gaps worth researching next.
  • Mood only changes tone. Trust only weights memory. Neither ever grants permissions.

data/learning · illustration

observe

“I prefer short replies, and I’m in Lisbon next week.”

facts · operator (max 12)

+ Prefers short replies salience 0.8

+ Travelling to Lisbon next week salience 0.6

scrub

“my key is sk-…” → [REDACTED:openai_key] · never stored

nightly · 03:00

merge · drop stale · diary entry · gaps to research next

08Lockdown

One word. Locked down.

Say “lockdown” and Reges stops taking actions. It cannot be closed; only the owner’s face or the master password lifts it.
  • Unlock by a stable owner-face match through the webcam, or with the master password.
  • The master password is stored as a salted scrypt hash, never in plain text.
  • Remote logins lock an address out for five minutes after eight wrong passwords.

Lockdown

Owner verification required. Look at the camera, or enter the master password.

Wording from the Reges lock screen

09Privacy

Nothing leaves your PC.

Local models on your own hardware: no Reges cloud, no account, no data trail. The internet is used only for what you ask for.
  • The server binds to 127.0.0.1 by default and refuses requests from other sites and unknown host names.
  • Bring your own model: LM Studio, Ollama, llama.cpp or any OpenAI-compatible endpoint.
  • When a developer connector finds a secret in your code, it reports its type and location, never the value.
  • Read the details on the security page.
The Reges command center with the live vision window detecting a face
The command center: system, agent, memory, tasks and what Reges sees, in one place.
Get Reges

See it on your desk.