Nexoteck AI
Code with AI inside VS Code. Free with your local models (Ollama), or use the best cloud models and pay only for what you use: prepaid balance, no subscription.
En español: un agente de programación con IA para VS Code, gratis con modelos locales (Ollama) o con modelos en la nube y saldo prepago, sin mensualidad. La interfaz está en inglés; el asistente responde en el idioma en que le escribes. Hay un resumen en español al final.
What you can do
- Chat with your project's context. Ask questions, request explanations or changes, and work on your open files and selection.
- An agent that works for you. It reads your project, edits files, runs safe commands and validates the changes. You choose how much control you want: confirmation before touching files, a diff view, or applying directly. Every turn can be undone.
- Verified fixes. Before modifying a file, the extension checks the proposal and applies only the lines that change.
- Editing in the editor. Select code and press
Ctrl+K Ctrl+I (macOS: Cmd+K Cmd+I).
- Autocomplete with a model of your choice. Accept with
Tab.
- You choose the model and the effort. Each model shows its usage (low, medium or high) and the Effort selector tells you how much each reasoning level spends.
- Images and files. Attach screenshots, PDFs and more to the models that support them.
- Vague ideas become clear requirements. Ask for something new in a few words (“build me an online store”) and, before spending anything, you get a short pre-filled form (name, scope, design, stack) so the first result is closer to what you want. One click to confirm, one to skip.
- Review and audit without a big bill. Review Changes checks your uncommitted changes locally first (secrets, SQL built by concatenation,
eval, disabled TLS, open CORS…) and sends only the risky fragments to the model, within a token budget you control. Audit Project scans the whole project locally. Findings appear in the Problems panel.
- Know what a task cost. With Nexoteck models the chat shows the real cost of each task from the receipts of the service, and you can set a spending limit per task.
- Go back to any point. Before each agent request the extension saves a local copy of your project files (no git needed,
node_modules and similar folders skipped). Click ↶ Restore on any of your messages to return the project to how it was, even if the agent changed or deleted files with commands. The restore itself can be undone.
- It checks its own work. After changing code the agent runs your tests and then compiles the project (
npm run build, tsc, dotnet build, go build, cargo check) and fixes what fails before saying it is done. Passing tests alone do not prove the project builds.
- Starts new projects the right way. With local models it checks that your computer has the tools a new project needs (Node.js, Python…), tells you in the chat which generator version works with the Node you have and offers to update it, and gives the model the official generator command instead of letting it invent one.
- Copy any answer. Every finished answer has a ⧉ Copy button; for an agent task it copies the final summary.
- Your language. Write in Spanish, English, Portuguese, French, German or Italian and the assistant replies in the same language.
Get started in three steps
- Install the extension and open the Nexoteck AI panel in the side bar.
- Choose how to connect:
- Local and free: install Ollama, download a model and select it in the chat. Your code never leaves your computer.
- Cloud with balance: click Sign in, authorize from your browser and top up from US$5 with PayPal at nexotecklabs.com. You do not copy keys from anywhere.
- Type your first request in the chat.
How billing works
One-time payments, no subscription and no automatic charges. You top up a balance in dollars and each response deducts only the tokens it uses, at the model's rate (shown in your account). One balance works for every model. You can set daily, weekly or monthly spending limits, and we email you when you are getting close or the balance runs out. Local models with Ollama cost nothing.
Privacy
- With Ollama, your code and conversations never leave your computer.
- With cloud models, the request context travels to Nexoteck and to the model provider. We do not store the content of your conversations or files on our servers, and before sending it we hide the credentials we detect (API keys, tokens).
- Anonymous usage statistics, and you can turn them off. To improve the agent, the extension sends whether you use a local or a cloud model, the route (agent, direct, chat), the model name, the mode, how each task ended and the type of agent errors, with a random installation id. Never code, file names, prompts or replies. Turn them off with
nexoteck-ai.usageStats or VS Code's telemetry.telemetryLevel. Web search is off by default.
- Read the privacy policy and the terms.
Agent safety
- Commands go through a policy: common ones run, risky ones ask for confirmation and destructive ones are blocked.
- Whatever the agent deletes goes to the trash.
- In an untrusted workspace the agent does not modify files or run commands.
- The settings that control confirmations are read only from your user settings: a foreign repository cannot turn them off.
- Secrets stay private. The agent does not read, search, print or change
.env files, private keys or cloud credentials, and does not dump your environment variables, unless your message explicitly says it may (“check the .env”). Templates such as .env.example are fine.
- No accidental data loss in your databases. Commands that drop or truncate tables, delete rows without
WHERE, reset migrations or remove database volumes always ask for your confirmation, even in the run-everything mode, and the scripts or .sql files they hide are inspected too.
- Your data survives updates. Conversations, profiles and restore points are versioned and migrated; if you go back to an older version, it will not overwrite what a newer one saved. See
docs/formatos-de-datos.md.
Useful settings
Open Settings and search for nexoteck-ai:
| Setting |
What it does |
nexoteck-ai.applyMode |
direct applies changes; review opens a diff view |
nexoteck-ai.confirmFileOps |
When to ask for confirmation before creating, moving or deleting files |
nexoteck-ai.agent.runCommands |
Which commands the agent may run |
nexoteck-ai.autocomplete.enabled |
Turns autocomplete on or off |
nexoteck-ai.autocomplete.model |
Model independent from the chat; DeepSeek Flash through Nexoteck by default |
nexoteck-ai.webSearch |
Lets the AI search the internet (off by default) |
nexoteck-ai.language |
Language of the messages shown outside the chat: en (default), es or auto (follows VS Code). The assistant always answers in the language you write in |
nexoteck-ai.requirements.ask |
Show the short requirements form for very vague requests for something new (on by default; costs no tokens) |
nexoteck-ai.agent.maxTaskCostUsd |
Spending limit per task in US$ (0 = no limit). The agent stops before starting another request once the confirmed cost reaches it |
nexoteck-ai.review.maxInputTokens |
Most tokens of changed code a review sends to the model (6000 by default) |
nexoteck-ai.location |
activityBar (default) or secondarySidebar: where the chat lives. Move Chat to the Other Side Bar switches it with one command |
Autocomplete appears automatically when you stop typing and is accepted with Tab. DeepSeek Flash requires a connected account and balance; each cloud suggestion uses balance. You can turn it on or off and choose its model in the settings above. Limit per suggestion: nexoteck-ai.autocomplete.maxTokens (128 by default).
Personalization separates general preferences from the ones you declare for this workspace. An instruction such as "only for this reply" applies to that turn without being saved as a preference. From the Command Palette you can use What It Learned About Me, Edit or Forget a Preference or Forget What It Learned About Me. Turning off nexoteck-ai.personalization stops learning and the use of the saved profile, including its language.
If something does not work
Run Nexoteck AI: Diagnose Connection from the Command Palette. In one step it checks the service address, whether it answers, your sign-in and balance, Ollama and the selected model, and tells you what to fix. The status bar shows the model and your balance, and turns into a warning when you are signed out or out of balance: click it to run the diagnosis. New here? Nexoteck AI: Get Started opens a short walkthrough.
Requirements
VS Code 1.140 or later. Ollama only if you want to use local models.
Support
Write to soporte@nexotecklabs.com or visit nexotecklabs.com/ai. Made by Nexoteck Labs.
Resumen en español
Nexoteck AI es un asistente de programación para VS Code: chat con el contexto de tu proyecto, un agente que edita archivos, ejecuta comandos seguros y verifica los cambios, edición en el editor y autocompletado. Es gratis con modelos locales (Ollama) o con modelos en la nube pagando solo lo que consumes, con saldo prepago y sin mensualidad. Con Ollama tu código no sale de tu equipo; con la nube se ocultan las credenciales detectadas y no guardamos el contenido de tus conversaciones. La interfaz está en inglés y el asistente te responde en el idioma en que le escribes. Antes de cada petición del agente guarda una copia local de tu proyecto para que puedas volver atrás con ↶ Restore, compila el proyecto después de las pruebas, no lee tus secretos (.env, claves) ni borra tablas de tus bases de datos sin que lo confirmes, y tus datos se conservan al actualizar. Ayuda: soporte@nexotecklabs.com.
| |