24 guides
Chat and models
Connecting a model, choosing one, and using a chat day to day.
Connecting a provider or key
Add a key for a cloud model, or your own endpoint.
- The "How would you like to chat?" screenIn NeuralVault 2.0 the app asks once per account on this machine how you want to chat: your NeuralVault balance, your own API key, or a model on this machine, and you can tick more than one.
- Paste an API key and how the provider is detectedIn NeuralVault 2.0 you paste a key into one box and the app works out the provider from the start of the key; a key it cannot recognise is not saved.
- Add an API key for a cloud providerPaste the key once in Settings, Intelligence, API keys. It goes into your operating system’s credential store, not into a file in the vault.
- Connect a cloud provider with your own API keyPaste one key in Settings, Intelligence, API keys and NeuralVault checks it with the provider before it keeps it; nine providers are built in and any other OpenAI-compatible service goes in the Direct providers form.
- Connect Azure OpenAI, Z.ai, or another direct endpointEndpoints that are not in the built-in list get their own form, where you supply the label, the base URL, the model, the auth style and the path yourself.
Chat will not answer
The key is refused, the model is silent, or the model you want is missing.
- Chat will not answer: what to checkA model that never replies is usually Ollama not running, a model not pulled, a key that was not accepted, or an image sent to a model that cannot see images.
- My API key is rejected, or the provider says there is no money leftThree different messages mean three different fixes: the provider refused the key, no key is saved for that provider, or the provider’s own account has run out of funds.
- A model is missing from the picker, and what its labels meanThe picker leaves out any model its provider has confirmed cannot use tools, and filters by provider, by data policy and by what is usable; a model the provider said nothing about is still listed.
- Chat gave an error: what each NV-ROUTE code meansIn the Beta build, a failed chat reply shows an NV-ROUTE code that says whether the key was refused, the account is out of funds, you are being slowed down, the model is gone, or the model is not set up.
Choosing a model
The model picker, data policy, reasoning effort and Orchestra.
- Choose which model answers in ChatIn NeuralVault 2.0 the model button on the chat screen opens a picker with search, filters and a Runs on column that says where each model runs.
- Pick a model, and filter by what the vendor does with your dataThe model picker has a data-policy filter with three settings: any policy, no training, or no training and no logging.
- Set the reasoning effort for a conversationIn NeuralVault 2.0 a chip in the chat footer opens a slider from Low to NV Ultra; the chip is absent when the app cannot confirm the current model supports it.
- Orchestra: split one task across several modelsOrchestra splits one task into subtasks, sends each to a model suited to its size, and has one model oversee the rest. It quotes the run in model calls before you start. It needs Personal or Business and is only partly built.
Models on your own machine
Ollama, LM Studio, another computer or an AI box on your network.
- Run a model on your own machine with OllamaInstall Ollama, pull a model, and NeuralVault finds it at localhost:11434 with no API key and no account.
- Connect LM Studio, or any other OpenAI-compatible serverAnything that speaks the OpenAI API on a URL you control connects through the "OpenAI-compatible" provider, and needs a base URL rather than a key.
- Connect an NVIDIA DGX Spark, or any AI box on your networkA Spark, a second PC, or any machine on your network running Ollama or another OpenAI-compatible server connects through the same "OpenAI-compatible" provider as a local one, once you give it a network address instead of localhost.
- Use Ollama running on another computer, such as a Mac or a Linux boxNeuralVault runs on Windows only, but the models can run on any machine on your network that runs Ollama. Make that machine listen on the network, then point NeuralVault’s Ollama address at it in one small file.
- Before you let other computers reach Ollama: a security noteMaking Ollama listen on the network lets any device that can reach that port use your models. Ollama’s documentation describes no sign-in for it, so keep it on a network you trust.
- Local model problems: Ollama not found, an empty list, a machine that is not reachableThe Ollama row in Settings, Intelligence, API keys tells you which of five states you are in and what to do next, whether Ollama is on this PC or another computer.
- Local and custom models ask before every command and outbound callAlso hereIn NeuralVault 2.0 a model you run yourself or reach through a custom endpoint is not trusted by default, so it asks before every shell command and every outbound call, and you can mark one trusted.Home: Tools, approvals and safety / How much it may do without asking
Using a chat
Organise chats, attach files, slash commands.
- Pin, archive, rename and group your chats in the sidebarIn NeuralVault 2.0 the sidebar lists your chats under Working, Pinned, Recent and Archived, and lets you group them into folders and sidebar workspaces.
- Select several chats and act on them togetherIn NeuralVault 2.0, Ctrl-click, Shift-click or a hover checkbox selects chats in the sidebar, and a bar offers Archive, Move to, Pin, Unpin and Delete.
- Attach an image or a text file to a messageIn NeuralVault 2.0 the message box takes images and text files by button, drag and drop or paste, with limits on size and count; an image needs a model that can see images.
- Slash commands: /new, /compact, /model, /plan and the restIn NeuralVault 2.0, typing / in the message box opens a menu of short commands: some do something at once, and some write a sentence into the box for you to read, change and send.
- A chat is listed under "Could not be read"In the Beta build, a chat whose saved file NeuralVault cannot read is listed in the sidebar under "Could not be read" with the code NV-UI-008; nothing was changed or deleted.
- Give the assistant a folder to work inAlso hereIn NeuralVault 2.0 you attach folders the assistant may read and change; it asks before each change and cannot reach anything outside them.Home: Tools, approvals and safety / Commands and folders