Meet Nx IDE.
Nx IDE is a code editor with an AI assistant built in. Your developers describe a task in plain words. The assistant, called Nexgile Code, reads the project, proposes the changes, and makes them once a person approves. It runs on the AI model you choose, including one on your own machines.
The same assistant also works as an extension for VS Code and from the command line.
- 01 Build Code
- 02 Test Tester
- 03 Review Reviewer
- 04 Ship? You
- 05 Ship Shipper
An illustration of an Orchestrator run, not a screenshot.
What Nx IDE is, and why teams use it.
An AI assistant that lives in the code editor.
It reads a project, explains it, writes and changes code, runs commands and tests, and checks its own work. A person approves what it does.
Routine work eats a software team's time.
Understanding unfamiliar code, fixing reported bugs, writing tests, upgrading libraries, keeping documents current. Nx IDE takes on that work, so developers spend their time on decisions.
Developers, and the people who depend on them.
Developers and engineering teams use it day to day. Leaders get software delivered sooner, with clear limits on what the assistant may do and a view of what it costs.
One task, from request to finished change.
Here is what happens when a developer asks Nx IDE for something.
Four starting points on its home screen
- Explain what this product does
- Fix an issue customers reported
- Make sure a feature works
- Check changes before release
- 01
You describe the task.
Write it the way you would tell a colleague. For example: “Customers reported that large orders skip the approval step. Find the cause and fix it.” You can also paste a screenshot.
- 02
It studies your project.
Nx IDE keeps a map of your code, called Cortex, on your own machine: what exists and what depends on what. The assistant consults the map to find the right place to work.
- 03
It shows you a plan.
For larger work, start in Plan mode. The plan is saved as a document you can read, comment on and approve before any code changes.
- 04
You approve each action.
Every edit and every command appears as a card showing exactly what will happen, with Approve and Deny. Nothing changes on your machine until you say so.
- 05
It checks its work. You can undo it.
After a change, the assistant is asked to review what it did and to run the build and the related tests. Each step is saved as a checkpoint, so you can take your files back to any earlier point.
31 specialists, each for one kind of work.
A specialist is the assistant in a particular role, with its own instructions and its own list of permitted actions. The Reviewer, for example, is given no file-editing tools.
Architect
Plans the work before any code is written. This is Plan mode.
Code
Writes, changes and tidies up code. Every new chat starts here.
Chat
Answers questions and explains how things work. It cannot change files.
Debug
Finds the cause of a problem and confirms it with you before fixing.
Orchestrator
Hands a task to other specialists and coordinates them.
Tester
Checks a change against the original request, independently.
Reviewer
Reviews a change for correctness and quality, independently.
Shipper
Commits approved changes to a separate branch and pushes them.
Plus 23 ready-made specialists for everyday work
Need another? Teams define their own specialists, with their own instructions and limits, and keep them in the project next to the code.
From one assistant to a team of agents.
For bigger jobs, Nx IDE splits the work across several assistants, called agents, and keeps you in charge of the result.
Built by one specialist, checked by two others.
- 01 BuildA Code specialist makes the change.
- 02 CheckA Tester and a Reviewer check it at the same time.
- 03 Your decisionYou answer one question: Ship?
- 04 ShipOnly then is the change committed and pushed.
The Tester and the Reviewer are given your original request and the list of changed files, not the builder's explanation, and neither has tools for editing files. If they find problems, the work goes back, up to twice.
Agent Teams
When a job splits cleanly, such as a service and the screen that uses it, two to five agents work as a team. They message each other and share one task list, so the parts fit together.
Batch changes
For a large change, a planning agent proposes up to 30 separate tasks. Once you approve the plan, each task runs in its own isolated copy of the project, and each result can go to your team as a draft for review.
Agent Manager
One window shows every chat and every helper agent across your projects. An inbox lists whatever is waiting on you, and you can stop any agent without stopping the rest.
Work toward a goal
Set a condition such as “all tests pass”. The assistant keeps working until a second AI model agrees the condition is met.
Ask on the side
Ask a quick question while a task runs. The answer appears in a side pane and the task carries on.
It does as much as you allow, and no more.
In the editor, the assistant starts fully supervised: it asks before every action. You loosen that one step at a time, and can tighten it again at any moment.
Supervised
You review every action.
Where it startsTrust Reads
It may open and read files without asking.
Reading onlyTrust Edits
It may also create and change files.
Checkpoints save every stepTrust Execution
It may also run commands and use connected tools.
Your rules still applyAutonomous
It may also hand work to helper agents on its own.
Publishing still asksEven at the top level, a question whose “yes” would publish, deploy or delete something always waits for a person.
Approval for every action
Edits, commands and connections to other tools each appear as a card with Approve and Deny. You choose which kinds of action may run without asking.
On by defaultRules for commands and files
List the commands that may run unasked and the ones that must never run. Permission rules go further: allow, ask or deny by file or tool, such as “edit anything except this folder”. Rules kept in a project can only make things stricter.
You write themSpending limits
Give a task a cost limit and a request limit. When it reaches either one, the assistant pauses and asks whether to continue. Daily and monthly budgets warn you as you approach them.
You set itCheckpoints
A snapshot is saved before every file change. Restore your files to any earlier point, or branch the conversation to try a different approach.
On by defaultYour own checks as gates
Run your own scripts at set points in a task. For example, the assistant cannot report a task as done until your tests pass.
You add themA sandbox for commands
Confines the commands the assistant runs so they can write only inside the project. It can also block their access to the network.
macOS and Linux · you switch it onHidden and guarded files
List the files the assistant must never see, such as secrets. Changes to its own rules and settings ask for your approval unless you decide otherwise.
A guard against loops
If the assistant starts repeating itself, it stops and asks for help instead of running on.
On by defaultEvery control is a setting or a file you can read. Your security team can check each one against the controls they already run.
Visit the Trust CenterIt learns the project before it changes it.
Nx IDE builds a map of your code on your own machine and keeps it current. The assistant uses the map to find the right place, and you can use it to see what a change would affect.
Cortex, a map of your code
Functions, files and screens, and what depends on what. Built on your machine; it needs no account and sends nothing over the network.
On by defaultSee the impact first
Before you change something, ask what else relies on it. The map shows what would be affected and how risky the change is.
Search by meaning
Optionally add a search that finds code by what it does, not only by its wording. It needs a search service and a database that you provide.
You set it upRemembers the project
Facts you state once are kept outside the code and recalled in later sessions. Long conversations are summarized automatically so the work can continue.
On by defaultAny AI model. Your code stays yours.
Connect the AI service your company already uses, or run a model on your own machines. Requests go straight from your computer to the provider you chose.
26 AI providers
Use the AI service you already trust, and switch whenever you like.
Fully local, if you choose
Run the model on your own machine. Then no code is sent to an outside AI service.
A model for each job
Give planning a stronger model and routine work a cheaper one.
Nothing reported back
Nexgile does not sit between you and your provider and does not receive your code. Access keys stay in your operating system's secure store, and the assistant sends no usage data to us.
It works with the tools your team already has.
The assistant runs your builds and tests, looks at the app it just changed, and connects to your other systems through an open standard.
Runs and reads commands
It runs builds and tests in the terminal and reads the results, with your approval.
Sees the app it built
It opens your locally running app in a browser, takes a screenshot and reads the errors. Click an element to point at it.
You switch it onConnects to your systems
Link issue trackers, databases and more. A built-in catalog installs connections without editing settings files, each with its own permissions.
Skills, shortcuts and rules
Write your team's know-how once and reuse it. A marketplace offers ready-made skills, previewed before anything is installed.
You can see what it costs, as it happens.
Every task shows its running cost. A dashboard on your machine shows where the money goes, and budgets warn you before spending gets out of hand.
Costs are estimates based on what each provider reports. Your provider's invoice is the final word.
Live cost on every task
The cost appears next to the work as it happens, helper agents included.
Usage and cost dashboard
Spending by AI model, by project and by specialist, with export to a spreadsheet.
Budgets that warn you
Set a daily or monthly budget and get a warning as you approach it.
You set itA limit per task
Give a task a cost limit. At the limit the assistant pauses and asks before spending more.
You set itWhat the last four months have added.
Nx IDE changes quickly. These are the main additions, and all of them are in the product today.
August
4 additions- Cortex, the local map of your code
- Agent Manager
- Project memory across sessions
- License keys checked on your own machine
September
12 additions- Agent Teams and batch changes
- Usage and cost dashboard with budgets
- Permission rules, and your own checks as gates
- Plans saved as documents you can comment on
- Catalogs for connections and for skills
- Code review on demand
- A browser panel for your local app
- A sandbox for commands on macOS and Linux
- Reading web pages, PDF, Word and Excel files
- Goal-driven work and side questions
- Voice input and suggestions as you type
- The chat in a web browser tab
October
4 additions- Helper agents and teams survive a restart
- Cortex and code search inside isolated copies
- Cortex reads more languages, older ones included
- A finished task must show the checks it claims to have run
Security scanning that checks every change on your own machine, with no network needed, is being built in. It is not available yet.
Nx·Studio designs and generates digital workers from a description of your business. Nx IDE is for your engineers: they use it to build, maintain and extend software. The two need no special connection. What Nx·Studio delivers is ordinary source code, so your engineers can open it in Nx IDE like any other project.
Start with the flagship: Nx·StudioAsk us the hard questions.
Short answers, with the conditions stated.
Will it act without my approval?
Not in the editor: approval is on by default and you choose what to delegate. The command-line version is made for automation, so it runs without prompts unless you start it in approval mode.
Can it run up a bill?
You can cap it. Give a task a cost limit and the assistant pauses for your decision when it gets there, and budgets warn you by day or month. No limit applies until you set one.
Which AI models can we use?
Your choice from 26 providers, or a model running on your own machines.
Does our code leave the company?
Requests to the AI model go directly from your machine to the provider you choose; Nexgile does not receive or store your code. With a local model, no code goes to an outside AI service. Optional features such as web search contact only the services you set up.
Do our developers have to change editors?
No. Nx IDE is a desktop editor with the assistant built in, and the same assistant is available as an extension for VS Code.
What do we need to start?
A license key, and either an account with an AI provider or a model on your own machine. The key is checked on your machine, without contacting Nexgile.
See it work on a real task. Then decide.
Bring a task from your own backlog. We will show Nx IDE plan it, build it, check it and hand it back for your approval.