The eight modules

Tap any module to read its guide — it opens right here, no new tab.

New here? Start with How it works for the plain-language tour, or jump straight to Krew — where most people begin.

What it actually is

Think about the tools you use for AI-related work today: maybe ChatGPT for writing, Zapier for automating, Copilot for code, a VPN or antivirus for safety. Each has its own login and bill, and they don't talk to each other. adris.tech is one desktop app that does all of it — and a few things that didn't have an affordable option before.

The biggest difference from any online AI tool: adris runs on your own machine. When you chat with an agent, that conversation isn't sent to a company's server. Your work, your files, your business conversations stay with you.

Who is this for?

You don't need to be technical. The people who get the most out of adris are:

1. Freelancers working aloneKrew agents handle the time-consuming parts so you focus on the work only you can do.
2. Small teams (2–15 people)Studio runs your repetitive processes, Krew writes your deliverables, Guard protects your accounts.
3. Developers & studentsCoder + Models + Mesh is the most affordable local-AI setup available — much of it free.
4. Small business ownersVault and Guard give you security that used to need an in-house IT team.

The stack it replaces

A typical solo operator's monthly AI spend, versus the same work inside adris:

The comparison changes quickly once you add up the subscriptions you're already paying. And because the heavy AI can run on your own hardware (or your own provider key at cost), there's no per-message markup.

An office, not a chatbot

Most AI is a chat box: you ask, it answers, you do the rest. Krew is a team of named workers — Arjun the chief of staff routes your task, Zara writes captions, Vera reviews code, Nyx researches companies. You talk to one; your whole office works.

The 57 agents

10 departments · all included on every plan

Each specialist is shown as Name.Role so you always know who's doing what. The line-up spans Content (6), Marketing (5), Sales (4), Support (6), Design (3), Data (6), Engineering (7), Product (8), Ops (2) and a Boss who routes everything — 48 in all. Nothing is locked behind a higher tier.

It reasons, then verifies

Every agent runs a loop — think, use a tool, look at the result, continue — before answering. It searches the web, reads your files and connected apps, and only then replies. That's the difference between a confident guess and a checked answer.

You stay in control

Every action that touches the outside world — send an email, post to Slack, create an issue — shows up as an approval card first. You see exactly what it wants to do, edit it if you like, and nothing irreversible happens until you tap approve. And when a task is done, the agent's browser window closes itself automatically.

What it is

Studio is where you build automations — sequences that happen without you clicking anything. A workflow is three parts: something that starts it, an agent that does the thinking, and an action that delivers the result.

Build it once. Watch it run.

An infinite canvas where every block is a step in your business

Drag a trigger from the sidebar, drop an agent in the middle, connect it to your tools on the right. That's a workflow — no code, no servers, no fragile glue.

1. A trigger starts itA time, a new email, a form submission, a webhook — you choose what fires the workflow.
2. An agent does the thinkingRead this, summarise that, decide, write a reply — the AI handles the judgement in the middle.
3. An action delivers the resultSend an email, update Notion, write a Sheet, post to Slack — dropped where it belongs.

Start from a template

Pre-wired, pre-tested workflows you can plug your accounts into and run: auto-reply to leads, weekly client report, invoice tracker, social scheduler, support router, daily standup digest, and more.

Human in the loop

Pause the workflow at any step, get a WhatsApp or Slack notification, and approve before anything irreversible — a client email, a payment. On approve it ships; on reject it's discarded, with the reason logged. Every run is recorded with a full step-by-step trace.

What it is

Copilot sees one file; ChatGPT doesn't know how your project connects. Coder reads your entire codebase when you open a project, so its answers are specific to your actual code — and it writes the change directly into the file, with a history you can roll back.

You pick the model

Most tools pick a model for you and bill what they like. Coder gives you a dropdown — free open-source (DeepSeek, Llama, Mistral, Qwen Coder) for the everyday 80%, premium models (Claude, GPT) at-cost with no markup only when you need them. A "free models only" toggle keeps you from spending by accident.

It edits your files — and remembers

Accept a change and it's written straight to disk. Every applied edit is snapshotted, so you can just say revert the last change and the file is restored. It remembers the conversation too — "now do the same for the other route" just works.

What it is

Hundreds of powerful models can be run for free; the hard part has always been setup. Models hub makes it two clicks — it auto-detects your RAM, downloads the right version, verifies it, and wires it into Krew and Coder automatically.

One-click setup

No terminal, no external tools. Open the Models tab, pick a model, click Download — adris detects your hardware, recommends a model that fits, verifies the download, and makes it ready in Coder.

No GPU? Use the cloud

If your laptop can't run a big model, plug your own OpenAI / Gemini / Claude key into Connect Apps. adris routes through it at the provider's price — no markup. Keys are stored encrypted in your OS keychain, never on our servers.

What it is

When you open a link, your device first looks up the address. Vault sits in that lookup and checks every request against lists of dangerous, ad-serving and tracking sites — blocking the bad ones before anything loads. Unlike a browser ad-blocker, it works system-wide.

Five protection levels

From "never break anything" to "lock it all down" — Swift, Block (recommended), Guard, Core and Family. Pick how aggressive it is; most people leave it on Block.

Private by design

Every lookup is encrypted (DNS-over-HTTPS), so nobody on your network — not your ISP, not the café router — can log where you go. It checks every domain against malware and phishing lists too, so flagged domains never reach your device.

What it is

Most small businesses are exposed not because they don't care, but because staying safe takes expertise they don't have time to build. Guard reviews your contracts, watches for threats and tells you what to fix — in plain language, nothing technical required.

Contract scanner

Drop a PDF — even scanned, even multi-language — and get a risk score in seconds

1. Reads scanned PDFsIf the pages are images, OCR pulls the text straight off the scan.
2. Handles Hindi & regional languagesMixed-language documents are transcribed, translated and analysed in English.
3. Scans 200-page agreementsLong contracts are read section by section in one upload — bounded so a scan never drains your budget.

Threats, vulnerabilities & compliance

Connect your inbox and Guard checks incoming email for phishing and fraud, logging everything to a tamper-evident audit trail. It sends a weekly briefing of new vulnerabilities in your stack — scored by whether you're actually affected. And it checks your code, config and policies against GDPR, DPDP, PCI and ISO, giving you a one-page scorecard with specific fixes.

What it is

Better models need more memory than most laptops have. Mesh pools the RAM of devices on the same Wi-Fi, so the model runs across all of them — three 8 GB laptops become one 24 GB pool. It's live in the app today.

How it works

A large model is made of many layers. Mesh splits those layers across the devices in your group, so each only holds a slice — and they work together over the local network to run the whole model. That's why three small laptops can run something none could alone.

Ways to use it

Pool your own devices (laptop + an old PC), a team's machines (idle memory chips in for heavy tasks), or students on the same Wi-Fi (hostel, library, study room). Up to 3 devices is always free; add more pay-as-you-go.

One menu decides everything

At the top of the adris window there is a single menu for which AI the app runs on. It is not a per-screen setting — it decides the chat, all 57 agents, your automations, Coder, and the AI tools on the Shelf. Change it once and everything follows.

You have four choices: your Claude Code subscription, your Codex subscription, your own API key (NVIDIA, Gemini, OpenAI, Groq and others — billed by them, never marked up by us), a local model (free, works with no internet), or adris.tech's own AI, paid per use.

Setting it up without a terminal

You do not install anything yourself and you never paste an API key.

1. Pick it in the menu. Choose "Your Claude Code" or "Your Codex".

2. adris installs it. If the tool is not on your machine, adris fetches and sets it up in its own folder. Nothing is installed system-wide and nothing else on your computer is changed.

3. Sign in once. You sign in to Anthropic or OpenAI directly, with the account you already have. adris never sees your password.

You can disconnect at any time from the same menu. It signs the tool out and adris goes back to whichever option you pick next.

Making sure it bills as your subscription

There is a real trap here and it has cost people money: run these tools the wrong way and the work is billed as API usage rather than against the subscription you already pay for.

adris checks which one you are on and refuses to run if connecting would put your work on a metered API bill. If it cannot confirm you are on a subscription, it stops and tells you, rather than spending your money to find out.

adris also shows where you are in your 5-hour and weekly usage window, read from the tool's own records on your machine — so running out is something you see coming rather than something that happens mid-task.

What leaves your computer

When you use this, what you ask goes to Anthropic or OpenAI under your own account and their terms — the same as if you had typed it into their app yourself. It does not come to us. adris.tech stores none of your business data, on this path or any other.

If you would rather nothing left the machine at all, choose a local model instead and the app works offline.

What it is

A narrow rail down the right edge of the window. It never expands — hover and the name appears beside it, so the work never shifts sideways when you reach past it.

Behind it is a catalogue of 18 tools, each listed against the paid product it replaces: EspoCRM and Twenty instead of Salesforce; Odoo, ERPNext and Dolibarr instead of Zoho One or SAP; Akaunting instead of QuickBooks; Invoice Ninja instead of FreshBooks; Chatwoot and Zammad instead of Zendesk; Documenso instead of DocuSign; Cal.com instead of Calendly; Baserow instead of Airtable; Metabase instead of Power BI; n8n instead of Zapier.

One click installs and starts it. adris waits until it is genuinely answering before it opens, so you never land on a dead page.

These are your installations, running on your machine under their own open-source licences. adris does not host them, resell them, or redistribute them.

Docker, plainly

These tools run inside Docker, which is free software for running applications in a contained way. If it is not on your computer, adris says so in plain words and points you at it rather than failing in a way you cannot read.

Two honest notes. Docker is a real download and it asks for a restart the first time. And Docker Desktop's own licence is not free for larger companies — above roughly 250 employees or $10M revenue it requires a paid subscription from Docker. That is their rule, not ours, and you should know it before you install rather than afterwards.

Some tools need a database beside them. adris starts those together from a file it generates itself, so an ERP arrives working rather than half-installed.

What adris will and will not run

Only what is in the catalogue. adris runs a fixed list of known images at pinned versions. It never builds software from source, and nothing an agent or a model produces can add something to that list — the catalogue is the boundary, and it is checked at the moment of running, not merely at the moment of asking.

✦ means seen working. Entries adris has actually installed and watched answer are marked as such. The rest say plainly that they are untested, rather than implying everything has been checked.

The AI-native tools on the Shelf run on whichever AI you chose in the title bar — the same one menu as the rest of the app.