Flexie SAGSmart Agent Gateway

AI agents that work on systems you already run.

Your databases, your servers, the files on your own computer. A CRM or an ERP records what happened; these act on it, and check with you before they change anything. They run on your own computer or your own server, so the work stays where you are.

Two editions: Personal, free on your own computer, and Enterprise, one installation your whole company works in.

Download for Mac Download for Windows

Free to use on your own computer · no account needed · 144 MB

Runs on your own machine or your own server
Use the AI providers you already pay for, or your own models
You choose what it may reach, and what it must ask about
Fully open source — read every line of it

Ask it about your customers. It goes and checks.

SAG Personal

Which customers slipped this month, and draft a follow-up for the three worth the most.

Thought for 6 seconds

Asked the customer database answered

select name, mrr, last_seen from customers where last_seen < now() - interval 30 day

Northwind Ltd£4,18041 days

Halcyon Group£3,05037 days

Six accounts have gone quiet past thirty days. Three carry most of it, together £9,400 a month.

I have drafted the follow-ups. Say the word and I will send them.

How can I help you today?

Attach files Ask to approve

AI can make mistakes. Check your AI vendor's terms & conditions.

What it can reach

Knowledge Runbooks Skills Files & search Terminal PowerShell Web browser API requests MCP servers SSH Client Gmail Google Workspace Microsoft 365 iCloud IMAP & SMTP CalDAV & iCal MySQL PostgreSQL SQL Server Oracle Local network link

Which customers slipped this month, and draft a follow-up for the three worth the most.

Asked the customer database

select name, mrr, last_seen from customers
where last_seen < now() - interval 30 day

Northwind Ltd£4,18041 days

Halcyon Group£3,05037 days

Six accounts have gone quiet past thirty days. Three carry most of it, together £9,400 a month. I have drafted the follow-ups.

What it does

It can reach the work, not just talk about it.

It works inside the systems you already run, and everything it can reach is something you switched on, for the people you chose.

Answers from your own data

Connect a database and ask questions of it in plain language.

You decide which tables it may see and which fields come back empty. Every query is checked against those rules before it runs.

It asks before it acts

Anything that changes something stops and shows you exactly what it is about to do.

You approve it or you do not. Nothing happens in between, and approving never fails afterwards.

Your files and your servers

It works on the computer you are sitting at, and on machines only you can reach.

Read and edit files, search a project, run a command. Each of those is a permission you grant or withhold.

Knowledge it can add to

Give it your policies, your product notes, your way of doing things.

It finds the one page it needs instead of re-reading everything, and can write back what it learns.

Any model, including yours

Use the providers you already pay for, or run models on your own hardware.

Both look the same to everyone using it, so work that should not leave the building can run on your own machines without anybody changing how they work.

You decide who gets what

People, groups and teams, each with their own models, tools and knowledge.

Permissions are checked every single time, so taking something away takes effect at once.

You decide what it sees, and who gets to see it.

Use the AI providers you already have an agreement with, or run models on your own machines so the work stays in the building. Either way the rules are yours: which systems it may reach, which fields come back empty, and what it has to ask you about first.

Running it

One place to decide what it may do.

Add people, connect the model providers you already use, switch capabilities on one at a time, and see what it is doing right now.

Get it

Two editions, the same product underneath.

Start on your own computer: download it and it works, with no server to set up, no account to make and nothing to configure before you can ask it something. When your company needs one installation everybody works in, that is Enterprise.

Flexie SAG Personal

Everything on one computer, answering only to you. Free.

Ready to use in about a minute
Your conversations stay on your disk
Reaches your own files and folders
Keeps itself up to date

Download for Mac Download for Windows

Flexie SAG Enterprise

One installation your whole company works in, on your own servers.

People, teams and who may do what
Shared models, tools and knowledge
Every permission checked on every request
Runs on your own servers

Talk to us

sales@flexie.io · taking early customers

For whoever sets it up

Run the models on a machine you rent or own.

Models can come from a provider's API, or on hardware you control. This is the second kind: a service you install on a machine with an NVIDIA graphics card. Once it has joined, those models appear in SAG like any other.

Installing

Open the console, go to Inference and add a machine. It hands you a single line to paste into a terminal on that machine, already filled in: the address of your own installation, and a token that stands for an hour. There is nothing to look up and nothing to type.

Paste it and the machine does the rest. It reads the graphics card, fetches the build that matches it and starts a service that comes back after a reboot. Nothing is compiled on your machine, and if the card is one we do not publish for, it says so and writes nothing at all.

The machine then appears in the console on its own, and the models on it can be given to whichever workspaces should have them.

Which graphics cards work

Any NVIDIA card from the Ampere generation onwards, which means 2020 and later. The engine needs bfloat16 in hardware, and cards older than that do not have it.

Not supported: Tesla T4, V100, and the RTX 20 series or older. These are common on the cheapest cloud instances, so it is worth checking before you rent one. The least expensive cards that do work well are the L4 and the A10.

Builds

The installer picks the right one by itself. They are listed here so you can fetch one by hand for a machine with no route to the internet, and pass it with --from.

Build	For	CUDA	Size
cpu	No graphics card. Works anywhere, and is a great deal slower.	—	21 MB
cuda80	A100, A30	12.8	51 MB
cuda86	A10, A10G, A40, RTX 3090, RTX A6000	12.8	50 MB
cuda89	L4, L40, L40S, RTX 4090, RTX 6000 Ada	12.8	49 MB
cuda90	H100, H200, GH200	12.8	134 MB
cuda100	B200, B100, GB200	12.8	54 MB
cuda103	B300, GB300 (Blackwell Ultra)	13.0	52 MB
cuda120	RTX 5090, RTX PRO 6000 Blackwell	12.8	53 MB

You do not need any of this to install: the installer picks the right build and checks it. It is here because a download you cannot verify is a download you are trusting on faith. Beside each file is a .sha256, a short fingerprint of it, so you can confirm the file you received is the file we published and not something altered on the way. And builds.txt is the plain list of which builds exist right now, for anyone mirroring them or checking before they start.

Checking a machine can be reached

A node listens on 19443, and SAG connects to it. The machine cannot tell you whether that works, because a firewall, a security group or a missing port forward all look like a healthy install from the inside. So ask from out here instead:

# the address the internet sees you at
curl https://sag-repo.flexie.io/v1/ip

# and whether your node answers on it
curl https://sag-repo.flexie.io/v1/reachable?port=19443

The installer does this for you at the end and prints the result. It only ever connects back to the address the request came from, so it cannot be aimed at anything else.

Adding it to SAG

The machine registers itself when the installer runs, using the token.
It appears in the console under Inference.
Download a model onto it there, then give that model to a team.