Meet Nx IDE.
A code editor built on VS Code, with the Nexgile Code AI agent built in. Describe a task in plain English — the agent reads your code, plans the change, writes it, runs your tests and reports back. It asks before it acts, works with the AI model you choose, and keeps every change reversible.
You
“Limit repeated login attempts, and add a test.”
Illustration of the agent working on a task.
What is Nx IDE?
Think of it as a capable colleague who sits in your code editor: you describe the job, it does the work, and you stay in charge of every change.
The editor developers already know
Nx IDE is built on VS Code, one of the most widely used code editors. Everything works the way developers expect, and they can add extensions from an open marketplace.
An AI agent that does the work
The Nexgile Code agent is built in. It reads your project, plans the change, writes the code, runs your commands and tests, and tells you exactly what it changed.
Rules you control
Out of the box it asks before it edits a file or runs a command. You decide how much to delegate, which actions are allowed, and which files are off-limits.
Developers
Hand off routine work like tests, fixes, clean-ups and upgrades, then review the result.
Tech leads
Set how the agent works for the whole team, with shared rules, specialists and checks kept in the project.
Modernization teams
Understand old code before you change it, then move it to a newer language or version with every difference marked.
Security & compliance
Approvals, rules for what the agent may do, records of what it did, and an AI choice that fits your policy.
One task, from request to reviewed change.
Example: “Limit repeated login attempts, and add a test.” Here is what happens.
Describe
Type the task in plain English — or start from a ready-made command such as Generate Tests or Translate Code.
Plan
For bigger work, the Architect writes a plan file you can read, comment on and approve before any code changes.
Build
The agent edits files and shows each change side by side for you to accept or reject. Restore points let you roll back at any time.
Check
It runs your build and tests, and fixes what fails. Give it a goal, and a separate AI judge decides when the work is really done.
Review
Separate Tester and Reviewer agents, which cannot edit code, test and review the change before anyone ships it.
Ship
A summary lists every file changed and every command with its result. The Shipper saves the change to your code history only when you say so.
Watch the agent work a task.
The same task, step by step — reading the code, writing a plan you approve, making each change only with your say-so, and running the tests until they pass.
To-dos: 0 of 4 done
- Read the login code
- Write the plan
- Add the limit and a test
- Run the tests
Delegation
Supervised
Review every action
Restore points
A save point after every step — roll back to any of them.
Illustration using the agent's own labels and steps. The project, code and cost are examples; the cost of a real task depends on your AI provider and model.
Recently added.
The biggest additions to Nx IDE — each marked with where it stands today.
Cortex: a map of your code
A map of how your code fits together, built on your own computer. It shows what depends on what, what a change could break, which tests it affects, and how risky your edit is.
Agent Manager
One screen to start, watch, answer and stop many agents at once — across several projects, even while the panel is closed.
Big changes, split up (/batch)
Splits a big change into up to 30 smaller tasks. Each one runs as its own agent on its own copy of the code, is checked, and can be sent for review on its own.
Agent Teams
Two to five agents that talk to each other and share a to-do list. For example, one can build the screens while another builds what runs behind them, with testing kept separate.
Plans you can edit
Plans are saved as files. Highlight any part, say what should change, and the AI rewrites just that part — with undo.
Finish lines and reviews (/goal, /review)
Set a finish line that a separate AI judge checks before the agent may stop — and get an independent review that rates each finding critical, warning or suggestion.
Project memory
Tell it once, for example “always add a test for new code”, and later sessions remember. It can also search your earlier chats. Notes stay on your machine.
Usage & cost dashboard
See what you spend by day, AI model, project and specialist, with daily and monthly budget alerts and a spreadsheet export. The data stays on your computer.
Your rules, written down
Decide which actions are allowed, which must ask and which are blocked — and add your own checks, like running the tests before the agent may say it's done.
Built in, and off until you switch them on:
See what a change will affect — before it's made.
Cortex is a map of your code that Nx IDE builds on your own computer — nothing is uploaded. It knows what depends on what, which tests cover it and how risky a change is. The agent checks it before editing, and you can explore it too.
- Impact: everything that depends on a piece of code, and how risky changing it is
- Changes: what your edits could affect, and which tests to run
- Flows and hotspots: how work moves through the code, and where it's most complex
- Rename something everywhere at once — previewed first, and undone in one step
Order total calculation
Illustration of Cortex's Impact view. The views listed are the panel's own.
Big changes, many agents at once.
Type /batch and describe a big change. The agent splits it into up to 30 smaller tasks. Once you approve the plan, each task runs as its own agent on its own copy of the code, up to eight at a time, and is checked before it's saved. Your own copy is never touched.
1 · One request
/batch translate every page of the app into Spanish
→ a plan of 12 tasks, each with its own checks, for you to edit and approve
2 · Agents at work, each on its own copy of the code
+ 8 more, up to eight at a time
3 · Ready to review
A change ready for review, for every task
— sent as a draft for your team to approve, or combined into one. You review each piece on its own.
Agent Manager
One screen to start, watch, answer and stop many agents at once — across several projects, even with the panel closed.
Agent Teams
Two to five agents that message each other and share a task list — while testing and review stay independent.
Finish lines (/goal)
Set a finish line, such as “all the tests pass”, and a separate AI judge decides when it's really met.
31 built-in specialists.
Each specialist focuses on one part of the job and gets only the tools it needs. The ones that test, review and ship can't edit code at all — the product enforces it. Pick one yourself, or let the Orchestrator hand work between them.
Code
Writes, changes and tidies up code. Every new chat starts here.
Architect
Plans and designs first, and saves the plan as a file for you to review.
Chat
Answers questions and explains how things work — without changing anything.
Debug
Tracks a problem down step by step, and asks before applying the fix.
Orchestrator
Coordinates the team: hands work to Code, Tester, Reviewer and Shipper.
Tester
Writes its own test plan, runs it and gives a verdict.
Reviewer
Reviews the change independently, with severity levels.
Shipper
Saves approved changes to your code history and shares them.
Spec Bootstrap
Asks you about the project and writes a project guide (a wiki), then runs nine checks on it.
Spec Sync
Keeps that wiki up to date as the code changes, after you approve the plan.
Plus 21 specialists for everyday engineering work
Need something else? Write your own specialist as a single file, add your team's rules and reusable skills, or install more from built-in marketplaces.
Control built into every step.
The checks your team already relies on — built into the agent itself.
One slider decides how much it does on its own.
Out of the box the agent is supervised: it asks before every action. Move the dial to hand over more: first reading, then editing, then running commands. Move it back any time. Try it.
Delegation dial
Supervised
Review every action
- Read filesasks you
- Edit filesasks you
- Answer its own follow-up questionsasks you
- Run terminal commandsasks you
- Use connected toolsasks you
- Switch specialistasks you
- Start sub-tasksasks you
At every level, blocked actions stay blocked, protected files still ask first, and reading web pages, copying other code projects and running custom tools always ask first.
Out of the box, it asks before every edit, command and tool call.
One slider, from “review everything” to “autonomous”. You choose.
Rules for which actions are allowed, must ask, or are blocked. A project can only make them stricter.
The agents that test, review and ship cannot edit code — enforced by the product.
A save point after every step, so you can roll back at any time.
Run your own checks at key moments — and stop the task finishing if they fail.
Changes to its own rule files and your workspace settings always ask first.
A per-task cost or request limit pauses the agent and asks before it goes on.
Who approved that change?
You did. Approval is on by default, your rules decide what may run without asking, and every automatic approval shows which rule allowed it. Your own checks keep logs too.
What did it cost?
It's on the dashboard, by day, AI model, project and specialist — and every reply shows what it cost. Spending and request limits pause a task and ask you; budgets send you an alert.
Who was allowed to touch the code?
Only what your rules allow. Blocked actions stay blocked, protected files always ask before a change, and the agents that test, review and ship can't edit code.
Every guardrail is a setting or a file you can read. Your security team can map each one to the controls they already run.
Visit the Trust CenterUse the AI you already trust.
Bring your organization's own account with any of 26 AI providers — or run models on your own hardware. Switch any time; nothing is locked in.
Straight to your provider
Your code goes from your computer to the AI provider you chose — there's no Nexgile server in between.
No tracking
The agent sends no usage data anywhere. Its cost records, memory and code map stay on your computer.
Keys kept safe
Your AI account keys are kept in your computer's secure password store, not in plain files.
Your own computers, if you want
Run AI models on your own computers, with tools such as Ollama or LM Studio.
A trade-compliance edition
A separate edition for buyers with trade-compliance rules (TAA). It leaves out China-based AI providers and models entirely.
License checked on your computer
Your license key is checked on your own computer — nothing calls a license server.
Ask us the hard questions.
Short answers, no hedging.
Will it act without my approval?
Not unless you let it. Out of the box it asks before every edit, command and tool call — and one slider sets how much you delegate.
Can it run up a bill?
You see the cost of every reply and every day. A per-task limit pauses it and asks, and daily or monthly budgets warn you as you approach them.
Which models can we use?
Your choice of 26 providers — or AI models that run on your own computers, so your code never has to leave your building.
Does it work with our existing extensions?
Yes. Nx IDE is built on VS Code, and installs extensions from an open marketplace (Open VSX).
Can it help with old code?
Yes. It can translate code into another language file by file, marking every difference, and its code map shows what a change will affect before you make it. Results depend on the AI model you choose.
What do we need to get started?
A Nexgile license key and an account with an AI provider, or an AI model running on your own computer. Builds are available for Windows, macOS and Linux.
Start with the workforce. Extend with the IDE.
Nx·Studio designs the digital workers that run your business and hands you their code. Nx IDE is what your engineers reach for to build and extend the software around them.