Foxora OS · Ember 4.2 · Open Beta

The operating system
for AI agents.

A real Linux machine where she can see the screen, click, type, and remember — signed in next to you, like any other user.

Free preview · 2.1 GB ISO · x86_64 + Apple Silicon · Signed installer

Foxora Desktop
FileEditViewAgentsWindowHelp
Tue 28 Apr · 02:09
Foxora Desktop— foxora-ai · auth.ts
coder · editing
Explorer
foxora-ai/
agents/
researcher.nix
coder.nix
deployer.nix
src/
auth.ts
agent.ts
runtime.ts
flake.nix
README.md
auth.ts
agent.tsruntime.ts
HMR
1
2
3
4
5
6
7
export async function authenticate(agent: Agent) { cons
Output
pnpm test auth
✓ 132 tests passed (132ms)
Vixens— 3 agents · active
RCD
M
ship v4.1 patch and notify the team
R
researcher
analyzing changes in auth.ts, runtime.ts…
C
ask your agents…
The shift

For fifty years, one user moved the mouse.
Now there are two.

Agents broke the old contract. We stapled them to sidebars and called it AI. But an agent is not a feature — an agent is a user. And every user needs an operating system.

Human
Vixen
youIdle
Vixen· Automating
2/4
Open Chrome
Read 12 papers
Save to research den~/dens/research/
Draft summary
one machine · two users
The pain

Your agents are users without a computer.

Today's agent stack is still glue code. Half-blind, forgetful, and trapped in runtimes built for servers, not users. Most teams spend more time wiring the machine than shipping outcomes.

Blind by default.

You bolt on a browser, a file system, a screen reader — one painful integration at a time. The agent still can't really see a screen.

> agent.see_screen()
AttributeError: agent has no display
> npm i playwright puppeteer browser-use
... 3 weeks of glue code later ...

No memory of yesterday.

Every run starts from zero. No persistent session. No learned shortcuts. No shared workspace between agents on the same task.

session #847 ▓▓▓▓ archived
session #848 ▓▓▓▓ archived
session #849 ▓▓▓▓ archived
// no memory between runs

Living in someone else's box.

Docker. Lambda. ECS. Sandbox APIs. All built for servers — not for users. Every action is a fight with the runtime.

runtime : Docker 24.0.7
filesystem : read-only
display : none
shell : bash (no AI)
There's a different way to do this
The answer

So we built the OS she was missing.

Two halves of one brain. Memory that persists. Motion that executes. Not a chatbox on top of a cloud.

Foxora Engine

She can see.

Foxora Engine reads the real repo graph, filesystem, and screen — not a chat transcript. Vixen understands the project the way you do.

vision · scan
4 detected
$184.2k+12.4%
Export →
header99%
metric97%
chart94%
button99%
How Engine sees
Shell + fur

She can act.

Shell and fur drive a real machine with native keystrokes and real processes. No sandbox API roleplay.

Foxora Shell — vixen
executing
$fox book-flight --to SFO --window morning
searched 47 fares · 0.8s
compared price + duration · ranked 3 options
booked UA 234 · seat 14F · 8:45 AM
added to den/calendar · synced
actions27
time2.4s
accuracy100%
See Shell in action
Foxora Den

She remembers.

Foxora Den is shared, local, auditable memory. Every Vixen and every human reads the same context — across sessions, machines, and teammates.

~/dens/vixen/memory
3 recalled
Sarah's PDF2h ago
agentic-uis.pdf · 4.2 MB
0.98
Project context3h ago
typescript · dark · jetbrains mono
0.94
Last conversation1d ago
"remind me to review the spec"
0.89
Enter the Den
Voice shell

She speaks.

Voice in, intent out. Shell turns speech into real commands with full provenance — every word traceable back to an action.

Foxora Shell · voice
listening
intenttype:pdf·from:sarah·last:7d
$scent --type pdf --from sarah --age 7d
found · agentic-uis-spec.pdf
Speak to Shell
One stack. One shared memory. Every layer built for Vixen.
The loop

Here's the loop.

Queue a bug, a feature, or a migration. Vixen picks it up from there .

queue

Developer queues work

Queue from fur and assign to Vixen.

$ fur queue bug "checkout throws NPE"

pick

Vixen picks it up

Reads Den context and asks Engine where to edit safely.

den recall + engine graph

stack

Vixen drives the stack

Shell commands, Desktop edits, fur tests, Engine checks.

shell + desktop + fur + engine + den

pr

PR lands with audit trail

Changes, tests, and memory references are linked end-to-end.

PR #1247 · provenance attached

Every PR carries its full provenance DAG — nothing is a black box .

Proof

Three example workflows. Show me it's real.

Specific, plausible runs that show how Vixen uses the full stack .

Bug fix
Example workflow

Queue null-pointer bug. PR in 7 minutes.

Developer queues a checkout NPE. Vixen reads Foxora Den for prior cart fixes, uses Engine to map callers, runs Shell commands, edits in Desktop, runs affected tests with fur, and opens a small PR with traceable decisions.

1 bug · 1 PR · full provenance trail
vixenFoxora Shellgithub
vixen/eng · pr queue8 ready
LIN-2174 · stripe webhook handler+184/-12ready
LIN-2189 · race in token refresh+47/-31ready
LIN-2191 · bump 12 deps · auto-testedready
LIN-2202 · /me endpoint cachewriting
247 tests passing across 4 PRsgreen
8 PRs this week · in your code style 0 reverts
Migration
Example workflow

Queue Express → Fastify. 34 tasks parallelized.

Team queues migration scope. Vixen breaks it into sub-tasks, coordinates dependencies through Engine, shares context through Den, executes and validates each chunk, then opens a reviewable set of PRs with linked reasoning.

34 tasks · parallel execution · review-ready output
vixenscentfoxora-files
brief · agentic-uis · 1-pager.mdwriting
papers · 47 from arxiv + ssrnread
customer interviews · 28 transcriptsread
post-mortems · 14 internal docsread
§ through-lines · 3 patterns surfacedwritten
§ decision frame · 1-pager readylive
200 sources · 1 page ready 6 AM
Feature
Example workflow

Queue Stripe refunds webhook. Vixen ships it.

PM queues feature intent. Vixen installs the right kit from Foxora Kits, wires endpoint + tests, documents behavior, and ships a PR with exact commands, file diffs, and memory references that produced the result.

feature queued · kit installed · tested PR opened
vixenzendesk24/7
support · zendesk · 247 todaytriaging
#8412 · refund · stripe charge dup'dresolved
#8413 · “site won't load” · IP-blockresolved
#8414 · auth bug · 12 users hit itto eng
#8415 · feature ask · added to backlogresolved
#8416 · enterprise contract · escalatedto you
247 today · 61% resolved 8 escalated · with context
Real testimonials plug in here automatically — never fabricated.

Meet your new user.

Boot Ember 4.1 on the machine you already have. She's waiting on the other side.

System requirements
Memory
4 GB · rec 16 GB
Disk
20 GB · rec 50 GB
CPU
64-bit · rec multi-core
Boot
UEFI · rec Secure Boot
Inside Ember 4.1
Linux6.8 LTS
GNOME47 + Ember UI
ShellFoxora Shell 1.4
Runtimevixen agent
Kits100+ pre-installed

Free preview · Signed installer · 12-month LTS · Verified by xBesh Labs

Built for the agents you'll write tomorrow.