Skip to content
Hydra

Your company’s AI agents, on your own servers.

Staff chat, make images and designs. Engineers run agents in terminals. Every request passes one set of permissions and lands in one audit log that stays on your hardware.

Request a pilot

Answer
A chat answering a question.
Make
An image being made.
Act
A terminal running an agent.

audit log · your server

  1. 09:41nora@kestrel approved ERP Sales Invoice SINV-0042
  2. 09:40ilse@kestrel ran claude in seat routing-2
  3. 09:38sam@kestrel generated 4 images in the Spring campaign kit
  4. 09:36mai@kestrel asked which shipments are late this week
  5. 09:35tomas@kestrel exported Q3 freight cost report (xlsx)
  6. 09:33dev@kestrel revoked API key ci-nightly
Many heads, one body. Each line is a real kind of event: who asked what.

Many heads, no body.

Your people already use AI. It just has no shape: every person and every team on their own, and nobody can say what happened.

Hydra gives the heads one body: your server, your people’s accounts, one set of rules, one record.

What it does

serene · answers

Answer

Staff ask in a chat. Hydra reads their Drive, the company ERP, and web pages through their own signed-in browser, and says where each answer came from.

Chat for staffInbox

A chat answering a question about open shipments.

laughing · makes

Make

Product pictures, kits and social designs follow your brand library. Everything made is kept, with who made it.

Images and kitsDesigns

The image library.

sly · acts

Act

Engineers run Claude and Codex in real terminals on the server. Before an agent changes company data, a named person approves it.

Terminals and agentsApprovalsDevices and companionAI API

Terminal seats with an agent running.

One Hydra, every team

What people type on an ordinary Tuesday, and what happens next.

All use cases →

Everyone brings their own account.

Each person brings their own Claude or ChatGPT plan. Hydra spreads the company’s work fairly across every connected account, so one team’s busy afternoon never stalls another’s, and no plan sits idle.

One account per person
An admin adds your plan in a minute and you approve it on your own device. Hydra never sees the password.
Fair share, by default
Everyone gets their slice. When your account is busy, a quiet one picks up the work.
More people, more room
Each plan someone brings adds room for the whole company.
Grow into API keys
When a team outgrows plans, switch it to API keys. Nobody changes how they work.

connected accounts · fair share now

  • nora Claude 72%
  • dev ChatGPT 58%
  • mai Claude 41%
  • sam Claude 66%
  • ilse ChatGPT 35%
  • tomas Claude 24%

6 accounts · 3 teams busy · nobody waiting

Sample data: Kestrel Freight, an invented company.

How model accounts work →

Where it runs

Hydra installs on a server you own. Your files, chats, images and audit log are stored there and nowhere else.

Stays on your server

  • Chats, files and the image library
  • Users, roles and approvals
  • The audit log

Leaves only through your accounts

  • Prompts, sent to the model provider you chose, under your own account
  • Nothing to us. We hold no copy of your data

Read the data flow →

One body

Roles decide who may do what. A named person approves before an agent writes to company data. The audit log records who asked what, and the owner can look at Hydra exactly as a colleague sees it, read-only, with that colleague told.

About the audit log →

The audit log.

A pilot

We install Hydra on your hardware and run it with a small group of your people. You keep what you set up.

  1. Your hardware

    One Linux server you own or rent. We tell you the size before we start.

  2. About two weeks

    Install, sign-in, roles, then real work with a few staff and engineers.

  3. You keep it

    The data stays on your server. A licence per server, priced with the pilot.

What a pilot involves →

Run it on your own servers.

Tell us how many people, and which servers. We reply with a plan for a pilot.

Request a pilot