Agenaxy is a local-first AI agent workbench. Files, chats, rules, run history, and artifacts stay on your machine while you choose a cloud or local model per task.

Does a cloud model see my whole machine?

No. Pick a cloud model for a task and it gets only that one task's context — never your history, and never the rest of your machine.

Vault Mode restricts sensitive work to trusted model and tool connections. Unapproved network access is blocked at the operating-system level, and a blocked step stops and reports why.

What does a run leave behind?

Every run leaves an artifact you can edit, a table you can check, and a structured action record — plans, tool calls, and files touched — attached to the session.

Is Agenaxy available today?

Agenaxy is in private beta. The website lets people request an invite by describing one real workflow they would hand off.

Private beta · macOS

Hire the model. Own the work. The AI workbench for sensitive files. The whole AI workbench on one machine.

Q: Where do my files live when I use Agenaxy?

On your machine. Agenaxy works on your files right where they live — no copies on anyone's server, and no one else can read them.

Q: Are my original files changed?

No. Every run works on a copy of your files; whatever the agent does, the files you started with stay exactly as they were.

Agenaxy is a local-first AI agent workbench: automate real work — documents, datasets, recurring workflows — while keeping your data safe on your machine. Hand confidential files to an agent inside a vault on your machine — local models, trusted connections only — and get the finished work back. Most agent platforms upload your work to the cloud — your files, your context, your history end up on someone else's servers. Agenaxy goes the other way: the model comes to your machine, and the work stays yours.

Get early access

Your files stay on your machine

Agenaxy works on your files right where they live — no copies on anyone's server. No one else can read them, not even us.

Your originals stay untouched

Every run works on a copy of your files. Whatever the agent does, the files you started with stay exactly as they were.

You control what the cloud sees

Pick a cloud model for a task and it gets that one task's context — never your history, never the rest of your machine.

Vault Mode

Lock your work to trusted connections only — a model on this machine, or one on your own servers.

Contract-Draft.pdf

Find clauses that may restrict publication. Cite exact sections and draft source-linked suggestions.

Prepared context, Read file, Reasoned +14 steps

Prepared context

Read file·Contract-Draft.pdf

AIindexed by section in your local workspace

Reasoned

Checked result

clause-notes.md created. Source excerpts, agent observations, and suggested edits remain linked. Illustrative workflow.

orders.csv

Build the weekly report, same format as last week — group complaints by issue.

Prepared context, Ran command, Reasoned +14 steps

Prepared context

Ran command·python -c <inline>

AIprofile and clean orders.csv, refunds.csv, reviews.csv

exit 0

Reasoned

Checked result

Weekly report generated — complaints grouped by issue, saved to weekly-report.html.

Ask the agent to work on these sealed files…Ask the agent anything…

AutonomyManualTrusted

Local model

Sealed · trusted connections only

Runs locally · stays on this device

One toggle, two modes — Standard for everyday work with any model · Vault when the work can't leave. Click the tabs to preview.

You probably need Agenaxy when the work matters more than the chat.

Five moments you might recognize.

01You've started a chat by deleting names and dates. Here, your data stays intact on your machine — you choose, task by task, what any cloud model sees.

02You're using AI for work, and you need a record you can explain.

03Your company has rules about what can go into AI tools — and you've signed them.

04You need to know what an agent did, not just what it said.

05Your context is worth more than the latest model.

How a run works

Three moves. Each leaves a durable record.

STEP 1

Delegate

Describe the files, rules, and outcome — the agent draws up a plan.

STEP 2

Run with control

The agent works on a copy of your files — your originals are never modified. You can steer or cancel any step.

STEP 3

Get the artifact + action record

Outputs land in your folder; a structured record of what happened — plans, tool calls, files touched — stays attached to the session.

Two directions

In the cloud, or on your machine.

Same convenience, opposite data flow. That's the whole difference.

The platform way

Every device is a door into a hosted workspace. Convenient — and the work settles there, not with you.

The Agenaxy way

The model is hired into your machine, task by task.

Inside Vault Mode

What Vault Mode blocks.

Turn on Vault Mode for sensitive work that must stay on machines you control. Unapproved network access is blocked at the operating-system level.

Tools that send data out are blocked.Web search, uploads, and remote connectors are disabled. Other tools cannot connect to destinations outside the allowlist.

Only approved models can connect.Use a model running on this machine or on a server you control. All other model connections are blocked.

Enforced by the operating system.Scripts and commands run in an OS sandbox that denies network access — an OS-level restriction, not merely an app setting.

Blocked connections fail visibly.If a step requires a connection outside the allowlist, it stops and tells you why.

Vault chats carry a mint Vault badge, and the composer shows Sealed · trusted connections only.

Read the full Vault Mode guide →

Real handoffs

Four moments where the trail matters.

Each starts with real source material, ends in a visible Artifact, and keeps the model's contribution separate from the person's decision.

E-commerce ops

Four exports. One Monday brief.

It is 8:42. The 9:00 meeting needs three answers: what changed, what needs investigation, and who acts next. The evidence is split across four exports.

Calculate agreed metrics, surface unmatched rows, and let the model draft—not invent—the interpretation. The operator approves every action. Use a local model, Vault, and local tools when customer rows should stay on this machine.

orders.csvrefunds.csvreviews.csvsupport.csv→weekly-ops-brief.mdexceptions.csv

CalculatedModel-draftedHuman-approvedIllustrative private-beta workflow

Read the e-commerce operations workflow →

Independent creator

The idea you remember. The source you don't.

A half-remembered sentence could unlock the next piece. It is somewhere across notes, voice-memo transcripts, and rejected drafts—but the original context matters.

Give one bounded archive to a local or explicitly trusted model. Ask for unexpected connections with source links, approve the useful idea cards, then draft separate angles in your own voice.

materials/→idea-cards.mddraft-angles.md

Source-linkedCreator-approvedIllustrative private-beta workflow

Read the private archive workflow →

Finance lead

A close pack you can trace.

A reviewer asks why a balance moved. The formula, unmatched rows, possible explanation, and accountable owner are scattered across different tabs and messages.

Use deterministic operations to reconcile approved exports, then let the selected model draft possible explanations without inventing a cause. A finance owner reviews and signs off; nothing is posted.

approved exports→close-review.mdexceptions.csv

CalculatedReviewer sign-offIllustrative private-beta workflow

Read the finance month-end workflow →

Lawyer

A first pass that points back to the clause.

A confidential contract arrives late in the day. The lawyer needs to narrow one issue without losing the exact language or surrounding context.

Supply the question and source set. Agenaxy enforces the chosen local or trusted connection boundary, records the run, and keeps source-linked observations and suggested edits beside the contract copy.

Contract-Draft.pdf→clause-notes.md

SourceAgent observationSuggested editIllustrative private-beta workflow

Read the confidential document workflow →

Each workflow separates: source material · model work · human decision · an inspectable run record.

Keep the receipts.

A record you can inspect

Every run keeps a structured action record — plans, tool calls, file reads and writes — attached to the session, so you can explain what happened.

Logged per run. Reviewable any time.

A vault you own

Your whole working site — files, memory, rules, chats, artifacts — lives on this machine at full fidelity, and never in an Agenaxy cloud account. Vault Mode locks your work to trusted connections only; pair it with a local model and nothing needs to leave at all.

Raw at rest on your disk. Sealed by the OS, not by a promise.

Models you choose

Pick the model per task — a leading cloud model, an open model, or one you run yourself. Open models now cost a fraction of the big-name rates (as of July 2026).

Per-task routing. Your workspace doesn't move when your model does.

Your AI workspace. Not theirs. Your files never had to leave. Runs here. Stays here.

We're opening a small private beta.

Tell us one workflow you'd hand off.

Get early access

Leave your email. Add a workflow note if you want.

Don't paste confidential content into this form — just describe the workflow.