# NeuralVault support guides Setup and how-to guides for NeuralVault, a local-first desktop app by Perpetual Technologies. Generated from the same source the support pages render, so this is never out of step with them. ## Models and keys Run a model on your own machine, or point NeuralVault at one you pay for. ### Run a model on your own machine with Ollama *Page: https://perpetualtechnologies.co.uk/support/connect-ollama* *Markdown: https://perpetualtechnologies.co.uk/support/connect-ollama.md* *Last checked 2026-09-06* Install Ollama, pull a model, and NeuralVault finds it at localhost:11434 with no API key and no account. 1. Install Ollama from ollama.com and let it finish starting. It runs as a background service and listens on http://localhost:11434. 2. Pull a model. In a terminal, "ollama pull llama3.1" is a reasonable first choice; smaller models answer faster on modest hardware. 3. Open NeuralVault and go to Settings, then AI & extensions, then API keys. Ollama appears under "Providers that run locally" and asks for nothing, because there is no key to give a server on your own machine. 4. Open the model picker in Chat and choose the model you pulled. **How to tell it worked:** Send one message in Chat. If it answers, the whole path is working: nothing left your machine to do it. **If it does not work:** - *Ollama is installed but no models are listed.* Installing Ollama does not download a model. Run "ollama list" in a terminal. If it is empty, pull one. - *The app cannot reach Ollama.* Open http://localhost:11434 in a browser. If that does not respond, Ollama is not running, and no setting in NeuralVault can fix that. Restart it, then reopen the model picker. - *Answers are extremely slow.* That is a hardware answer, not a configuration one. A model larger than your available memory will be swapped to disk. Try a smaller variant. ### Connect LM Studio, or any other OpenAI-compatible server *Page: https://perpetualtechnologies.co.uk/support/connect-local-openai-compatible* *Markdown: https://perpetualtechnologies.co.uk/support/connect-local-openai-compatible.md* *Last checked 2026-09-06* Anything 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. 1. Start your server and note the address it prints. LM Studio’s local server defaults to http://localhost:1234/v1; llama.cpp and vLLM print theirs on launch. 2. Load a model in that server first. An OpenAI-compatible endpoint with no model loaded answers requests with an error, which reads in NeuralVault like a connection problem rather than an empty server. 3. In NeuralVault, go to Settings, then AI & extensions, then API keys, and find the OpenAI-compatible provider. 4. Give it the base URL, including the version path if your server uses one. Most want the /v1 on the end; the address the server itself printed is the one to trust. **How to tell it worked:** The model picker lists the models your server reports. If the list is populated, NeuralVault reached it. **If it does not work:** - *The model list is empty but the server is running.* Nearly always a missing or extra /v1 on the base URL. Open /models in a browser: if that returns JSON, the URL is right. - *It works in a browser but not from NeuralVault.* Some servers bind to 127.0.0.1 only and some to all interfaces. Use exactly the host the server printed, not a substitute you assume is equivalent. ### Add an API key for a cloud provider *Page: https://perpetualtechnologies.co.uk/support/add-provider-api-key* *Markdown: https://perpetualtechnologies.co.uk/support/add-provider-api-key.md* *Last checked 2026-09-06* Paste the key once in Settings, AI & extensions, API keys. It goes into your operating system’s credential store, not into a file in the vault. 1. Create the key in the provider’s own dashboard. NeuralVault never creates one for you and cannot see your account there. 2. Open Settings, then AI & extensions, then API keys. 3. Pick the provider from "Add a key", paste the key, and save. 4. Open the model picker in Chat. That provider’s models are now selectable. **How to tell it worked:** The provider moves into "Your keys". Send one message on one of its models to confirm the key is accepted, not merely stored. **If it does not work:** - *The key saves but every message fails.* Stored and valid are different things. Check the key has credit or an active billing method on the provider’s side, and that it has not been scoped to a project that excludes the model you picked. - *You want to know where the key actually went.* Into your OS credential store under the service name NeuralVault: Credential Manager on Windows, Keychain on macOS, the Secret Service on Linux. It is not written into your vault and it is not synced anywhere. - *You need to remove a key.* Delete it from the same screen. Removing your data wholesale is a separate control, in Settings, General, Remove my data. ### Connect Azure OpenAI, Z.ai, or another direct endpoint *Page: https://perpetualtechnologies.co.uk/support/connect-azure-or-direct-endpoint* *Markdown: https://perpetualtechnologies.co.uk/support/connect-azure-or-direct-endpoint.md* *Last checked 2026-09-06* Endpoints 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. 1. Open Settings, then AI & extensions, then API keys, and scroll to "Direct providers". 2. Give it a label you will recognise later, such as "Azure prod". The label is yours; it is not sent anywhere. 3. Enter the base URL exactly as your provider documents it. 4. Enter the model or deployment name. For Azure this is your deployment name, not the underlying model name, and using the wrong one is the single most common failure here. 5. Set the auth style and the path template if your endpoint differs from the default. Leave both alone unless your provider’s documentation tells you otherwise. 6. Save, then add its key the same way as any other provider. **How to tell it worked:** The provider appears in its own "Direct providers" list, separately from the built-in ones, and its model is selectable in Chat. **If it does not work:** - *Requests return 404.* The path template does not match your endpoint. Azure deployments in particular sit under a deployment-specific path rather than the plain /chat/completions most OpenAI-compatible services use. - *Requests return 401 with a key you know is good.* Auth style. Some endpoints want a bearer token and some want a provider-specific header. Match what the provider documents, not what another provider uses. ### Pick a model, and filter by what the vendor does with your data *Page: https://perpetualtechnologies.co.uk/support/choose-a-model-and-its-data-policy* *Markdown: https://perpetualtechnologies.co.uk/support/choose-a-model-and-its-data-policy.md* *Last checked 2026-09-06* The model picker has a data-policy filter with three settings: any policy, no training, or no training and no logging. 1. Open the model picker in Chat. 2. Choose one of the three data-policy chips. "Any data policy" shows everything. "No training" hides vendors that train on what you send. "No training or logging" is stricter again, and keeps only vendors whose published policy commits to neither. 3. Search or scroll to the model you want and select it. **How to tell it worked:** The list shortens as you tighten the filter, and the chip you chose stays selected. **If it does not work:** - *A model you expected has disappeared.* That is the filter working. Its vendor’s published policy does not meet the level you selected. Loosen the chip to see it again. - *Local models are still listed under the strictest setting.* Deliberate. A model running on your own machine sends nothing to a vendor, so no vendor policy applies to it and there is nothing for the filter to exclude. ### Let the assistant search the web *Page: https://perpetualtechnologies.co.uk/support/let-the-assistant-search-the-web* *Markdown: https://perpetualtechnologies.co.uk/support/let-the-assistant-search-the-web.md* *Last checked 2026-09-06* Web search uses a Brave Search API key, added on the same screen as your model keys but listed separately, because it is not a model provider. 1. Get an API key from the Brave Search API. Brave offers a free tier suitable for personal use. 2. Open Settings, then AI & extensions, then API keys. 3. Brave Search sits at the end of the list, below the model providers. Add the key there. **How to tell it worked:** Ask the assistant something that requires current information. If it can answer, the key is live. **If it does not work:** - *You cannot find Brave in the "Add a key" list.* It is intentionally kept out of that dropdown and given its own row, so that a missing search key never hides behind a missing model key. Look below the model providers. ### Model tiers: which model runs at which level *Page: https://perpetualtechnologies.co.uk/support/model-tiers* *Markdown: https://perpetualtechnologies.co.uk/support/model-tiers.md* *Last checked 2026-09-07* Three levels run at different times and cost different amounts. You can put a different model on each, or leave them on your default. 1. Open Settings, then AI & extensions, then Model tiers. 2. Cortex is the level that plans and reads every result. Agents work inside the brief it writes. Sub-agents do the narrowest jobs and run most often, so that level is where model choice moves the bill most. 3. Leave a level on Use the default to keep it on whatever provider you have configured, and set one only where you actually want something different. 4. Save. The change applies from your next turn, not to anything already running. **How to tell it worked:** The card confirms it saved and says the change applies from the next turn. **If it does not work:** - *A model you expected is not in the list for a level.* The list is built from the providers you have configured. Add the provider key first, under API keys, and the models appear here afterwards. - *The change did not seem to affect the answer you just got.* It applies from the next turn. A turn already in flight finishes on the model it started with. ## Connections Let the assistant reach a tool, a server or a service you already use. ### Connect an MCP server *Page: https://perpetualtechnologies.co.uk/support/connect-an-mcp-server* *Markdown: https://perpetualtechnologies.co.uk/support/connect-an-mcp-server.md* *Last checked 2026-09-06* A connector is an MCP server NeuralVault may run on an agent’s behalf. You test it first, then choose which of its tools an agent may call. 1. Open Settings, then AI & extensions, then Connectors and integrations. 2. Add a connector: give it a name, the command that starts the server, and its arguments. 3. If the server needs credentials, enter the environment variable NAMES here, not their values. Never type a secret into the command or arguments boxes: both are shown back to you in the connector list, and neither is treated as sensitive. 4. Press "Test connection". This runs the server once and asks it for its real tool list, so what you see next is what the server actually offers rather than something typed by hand. 5. Tick the tools you want to grant. Save is deliberately disabled until a test has succeeded, because there is nothing to save before then. 6. Save the connector. **How to tell it worked:** The connector appears in the list with the tools you granted. Its tools become available to agents. **If it does not work:** - *The test fails immediately.* The command is what a shell would run. Check the program is on your PATH, and that the arguments are separate entries rather than one string with spaces in it. - *The test succeeds but the tools do not work later.* Almost always missing credential values. Declaring the variable name is only half of it. See "Give a connector its credentials". - *You granted a tool but an agent still will not call it.* Granting and marking read-only are separate ticks. A granted tool that is not marked read-only waits for your approval before an agent may run it, which is the deliberate default. ### Give a connector its credentials *Page: https://perpetualtechnologies.co.uk/support/connector-credentials* *Markdown: https://perpetualtechnologies.co.uk/support/connector-credentials.md* *Last checked 2026-09-06* Credential values are entered per connector after it is saved, in a password field that is never shown back to you. 1. Save the connector first, declaring the environment variable names it needs. 2. Find the connector in the list. Each declared variable now has its own credential row. 3. Enter the value and save it. It goes to your operating system’s credential store, one way: the app cannot read it back to show you, only pass it to the server it belongs to. **How to tell it worked:** Test the connection again. A server that failed on authentication before will now list its tools. **If it does not work:** - *You cannot see the credential rows.* They only exist once the connector is saved with variable names declared. An unsaved form has nothing to attach a credential to. - *You want to check what you typed.* You cannot, by design. Re-enter it rather than hunting for it. Nothing in the app displays a stored credential value. ### Plugins: what they are, and how one connects *Page: https://perpetualtechnologies.co.uk/support/plugins-in-this-build* *Markdown: https://perpetualtechnologies.co.uk/support/plugins-in-this-build.md* *Last checked 2026-09-07* A plugin is a separate program that talks to NeuralVault over a local socket with narrow, revocable permissions. Personal edition or above. 1. Open Settings, then AI & extensions, then Plugins. On Free the card is not shown at all, because the plugin API is a Personal capability rather than a Free one. 2. A plugin is a program you run yourself, not something installed from a store. It connects to NeuralVault over a local socket and asks for the narrow set of capabilities it needs. 3. The screen lists every plugin currently connected, which capabilities each one was actually granted, and how long ago it registered. 4. Revoke removes a plugin’s grant, and it takes effect immediately. **How to tell it worked:** A connected plugin appears in the list with its granted capabilities beside it. With none connected the card says so, and points at the bundled example plugin and the protocol document. **If it does not work:** - *There is no Plugins card in Settings at all.* It is hidden rather than empty. The plugin API needs Personal or above, and on Free the whole card is left out instead of shown and refused. - *Your plugin does not appear in the list.* It has not connected. The list stays empty until a plugin program is running and has registered; starting NeuralVault does not start your plugin for you. - *A plugin holds a capability you did not mean to give it.* Revoke it there. What a plugin asked for and what it was granted are shown separately, so check the granted column rather than assuming the request was met in full. ### See what an agent is allowed to do *Page: https://perpetualtechnologies.co.uk/support/what-an-agent-is-allowed-to-do* *Markdown: https://perpetualtechnologies.co.uk/support/what-an-agent-is-allowed-to-do.md* *Last checked 2026-09-06* Settings, AI & extensions, Tools lists every capability an agent could call, banded by the worst thing a mistaken call could do. 1. Open Settings, then AI & extensions, then Tools. 2. Read the consequence band on each one. The bands escalate: reads only, changes your vault, leaves this machine, cannot be undone. 3. The band is about damage, not frequency. A rarely used tool that cannot be undone still sits in the worst band, which is the point. **How to tell it worked:** The list matches the tools your build actually has, with a band against each. **If it does not work:** - *You want to switch a tool off from this screen.* This pane reports, it does not control. What an agent may actually reach is decided by connector grants (Settings, AI & extensions) and by the approval step before a write. ### Where your calendar events come from *Page: https://perpetualtechnologies.co.uk/support/where-calendar-events-come-from* *Markdown: https://perpetualtechnologies.co.uk/support/where-calendar-events-come-from.md* *Last checked 2026-09-06* Settings, Sync & calendar, Calendar reports which sources are producing events. It reports only: there are no controls on it. 1. Open Settings, then Sync & calendar, then Calendar. 2. It counts the events arriving from each source over a wide window, so you can see which are producing something and which are silent. 3. A source showing nothing is genuinely ambiguous: it can mean no feed is configured, or a feed that is configured and empty. The screen says so rather than guessing. **How to tell it worked:** The counts match what you expect to see on the Calendar page itself. **If it does not work:** - *You are looking for a way to add or remove a calendar source.* There is not one in the app yet. The backend publishes no command to list, add or remove a source, so a control here would be a button with nothing behind it. - *The Calendar page is not there at all.* It is a Business edition feature. Check Settings, General, Licence for which edition this account is on. ### Outbound hosts: where tools are allowed to send requests *Page: https://perpetualtechnologies.co.uk/support/outbound-hosts* *Markdown: https://perpetualtechnologies.co.uk/support/outbound-hosts.md* *Last checked 2026-09-07* An allow-list of hosts. Anything not on it stops and asks you before a single request goes out, and it starts empty. 1. Open Settings, then AI & extensions, then Outbound hosts. 2. On a new install nothing is allowed, so every outbound call will ask you first. That is the intended starting state rather than a misconfiguration. 3. Add a host by name, for example docs.rs, and press Enter. 4. Add only what you actually want reachable. Everything else keeps stopping for approval, which is the point of the list. **How to tell it worked:** The host appears in the list, and a tool call to it stops asking for approval. **If it does not work:** - *You are still asked about a host you added.* The entry has to match the host the tool actually calls. A subdomain is a different host, so allowing the parent does not cover it. - *You would rather not be asked at all.* There is no off switch for the prompt in this build, and the empty list is what makes it safe by default. Add the hosts you use and the asking stops for those. ### Sync a vault between your own machines *Page: https://perpetualtechnologies.co.uk/support/sync-across-devices* *Markdown: https://perpetualtechnologies.co.uk/support/sync-across-devices.md* *Last checked 2026-09-07* Points this vault at a sync server you run, and reconciles with it on a schedule. Personal edition or above. 1. Open Settings, then Sync & calendar, then Sync. If that whole group is missing from the Settings rail, your edition does not include it. 2. Sync reconciles this vault against a running nv-sync-server instance, which you run yourself. It is not a service we host for you. 3. Put the server URL in, then the token. The card shows a configured badge once a token is stored. 4. With sync enabled a background reconcile pass runs on an interval, and the card tells you what that interval is. **How to tell it worked:** The token row shows the configured badge, and the card states the reconcile interval it is running on. **If it does not work:** - *There is no Sync & calendar group in Settings.* The whole category is hidden when your edition includes neither device sync nor calendar, rather than shown and refused on save. - *You expected sync without running anything yourself.* There is no hosted option in this build. Sync talks to an nv-sync-server instance on your own network, so something has to be running it. ## Your vault and the app Where your notes live, the instructions the assistant follows, and how the app looks and behaves. ### Find, choose or move your vault *Page: https://perpetualtechnologies.co.uk/support/where-your-vault-lives* *Markdown: https://perpetualtechnologies.co.uk/support/where-your-vault-lives.md* *Last checked 2026-09-06* Settings, General, Vault names the actual folder on your machine, and lets you point NeuralVault at a different one. 1. Open Settings, then General, then Vault. It shows the real path and how many notes are in it. 2. To use a different folder, choose it there. To start a fresh one, create it there. 3. Your notes are ordinary Markdown files in that folder. You can open it in Explorer or Finder and read every one of them without NeuralVault. **How to tell it worked:** The path shown matches the folder you opened, and the note count matches what is in it. **If it does not work:** - *The app opens and everything is empty.* This is nearly always the vault pointing somewhere other than where you think. Read the path on the Vault screen first, before assuming notes were lost. They are files on disk; if the folder still has them, nothing is gone. - *You moved the folder yourself and now it is empty.* Point the app at the new location from the Vault screen. NeuralVault does not follow a folder you move behind its back. ### Write standing instructions the assistant always follows *Page: https://perpetualtechnologies.co.uk/support/standing-instructions* *Markdown: https://perpetualtechnologies.co.uk/support/standing-instructions.md* *Last checked 2026-09-06* An instruction file at the root of your vault is injected into every conversation. It can be called CORTEX.md, CLAUDE.md or AGENTS.md, and exactly one of them is ever in force. 1. Create a file called CORTEX.md at the top level of your vault folder. 2. Write your standing instructions in it as plain Markdown: how you want to be addressed, what the assistant should never do, conventions it should follow. 3. Open Settings, then AI & extensions, then Cortex standing instructions, to see exactly what is in force. **How to tell it worked:** The Standing instructions screen shows your text, and names which of the three filenames it read. **If it does not work:** - *You wrote CORTEX.md but the assistant ignores it.* It has to be at the ROOT of the vault. A CORTEX.md inside a subfolder is an ordinary note. - *You have both a CORTEX.md and a CLAUDE.md.* Only the first of CORTEX.md, CLAUDE.md, AGENTS.md that exists and is not blank is used. Never both. A blank CORTEX.md is skipped, so a written CLAUDE.md beside it is what takes effect. ### Skills: what they are and where they come from *Page: https://perpetualtechnologies.co.uk/support/skills* *Markdown: https://perpetualtechnologies.co.uk/support/skills.md* *Last checked 2026-09-06* The Skills screen lists every Claude Code skill found on this machine. It reports what is installed; it is not a set of switches. 1. Open Settings, then AI & extensions, then Skills. 2. Search by name to answer the usual question, which is whether a particular skill is installed. 3. To add one, install it where Claude Code keeps skills on this machine. NeuralVault discovers what is on disk rather than managing an installer of its own. **How to tell it worked:** A newly installed skill appears in the list, and the total count goes up. **If it does not work:** - *You are looking for a per-skill on/off switch.* There is not one, anywhere in the product. The Enabled column is a status, not a control. Opting in happens once per directory, and narrowing what a particular agent may use is done per agent. ### Change the theme and how tightly packed it is *Page: https://perpetualtechnologies.co.uk/support/theme-and-density* *Markdown: https://perpetualtechnologies.co.uk/support/theme-and-density.md* *Last checked 2026-09-06* Settings, General, Appearance. Both the theme and the density persist, and both apply the moment you pick them. 1. Open Settings, then General, then Appearance. 2. Pick a theme and a density. The change applies immediately rather than after a restart. 3. Both are written to your settings file and read back next launch, so the choice survives closing the app. **How to tell it worked:** Close and reopen the app. It comes back the way you left it. **If it does not work:** - *It looks right now but resets when you reopen.* The change applies to the screen immediately and is saved in the background, so a display that is correct proves the first half only. A reset means the write failed, and that is worth telling us about. ### Turn native desktop notifications on or off *Page: https://perpetualtechnologies.co.uk/support/notifications* *Markdown: https://perpetualtechnologies.co.uk/support/notifications.md* *Last checked 2026-09-07* One switch, under Connections, for whether NeuralVault raises your operating system’s own notifications. 1. Open Settings, then Connections, then Native OS notifications. 2. The single switch turns them on or off, and it saves as you press it. **How to tell it worked:** The switch stays where you put it after leaving Settings and coming back. **If it does not work:** - *The switch is on and you still see nothing.* Your operating system has its own permission for notifications, separate from this one. Check that NeuralVault is allowed to notify you in your system settings. - *You expected the Connections group to hold your connectors.* It does not. Connections holds notifications and hotkeys only. Connectors live under AI & extensions, which is the single most confusing thing about this Settings page. ### Keyboard shortcuts, and why you cannot change them *Page: https://perpetualtechnologies.co.uk/support/hotkeys* *Markdown: https://perpetualtechnologies.co.uk/support/hotkeys.md* *Last checked 2026-09-07* The shortcuts are listed under Connections. They work inside the app only, and they are fixed in this build. 1. Open Settings, then Connections, then Hotkeys. 2. The card lists every shortcut with the keys beside it. That list is the reference: it is generated from the shortcuts the app actually has. **How to tell it worked:** The shortcut you wanted is in the list, with its key combination beside it. **If it does not work:** - *You want to change or rebind one.* You cannot in this build. The card says so in as many words: they are fixed for now. There is no settings file that overrides them either. - *A shortcut does nothing when the app is not focused.* That is expected. These work inside the app only, so they are not global shortcuts and will not fire from another window. ## Automation, safety and backups What runs on its own, what it is allowed to do, and what is being kept. ### Background jobs, and turning one off *Page: https://perpetualtechnologies.co.uk/support/background-jobs* *Markdown: https://perpetualtechnologies.co.uk/support/background-jobs.md* *Last checked 2026-09-07* NeuralVault runs a few scheduled jobs on their own. The Background jobs screen lists each one, when it last ran, and a switch. 1. Open Settings, then System, then Background jobs. 2. Each row is one job, with a switch and the time it last ran beside it. A job that has never run says so rather than showing a blank. 3. Turning a switch off saves immediately. There is no separate save button on this card. **How to tell it worked:** The row you changed keeps its new position after you leave Settings and come back, and its last-ran time stops advancing once it is off. **If it does not work:** - *A banner says background scheduling is paused in this window.* Your switches still save. They take effect wherever the active scheduler is running, which is a different window or a different machine, so the change is not lost. - *A job says no run recorded yet and never changes.* It has not had a reason to run, or the scheduler is paused as above. The time shown is the last actual run, not a countdown to the next one. ### The Safety screen: rewrites, security events and audit logs *Page: https://perpetualtechnologies.co.uk/support/safety-queues* *Markdown: https://perpetualtechnologies.co.uk/support/safety-queues.md* *Last checked 2026-09-07* Three review queues in one place: changes the assistant wants to make to itself, security events, and a live view of every agent run. 1. Open Settings, then System, then Safety. 2. Self-rewrite review queue counts changes the assistant has proposed to its own instructions and is waiting on you for. Review opens the queue. 3. Security events counts anything open that wants a decision. Nothing open is the normal state. 4. Audit logs opens a live view of agent runs: queued, running and recent. **How to tell it worked:** Each of the three rows shows a count or an explicit nothing open, rather than staying on loading. **If it does not work:** - *A row says it could not load.* The card reports what it can and names the error on the row that failed, so the other two are still trustworthy. It is a read failure rather than a sign that something was approved without you. - *There is no Review button beside a row.* That review page is not available on your edition. The count is still shown, so you can see there is something there. ### Backups: what runs on its own, and what does not *Page: https://perpetualtechnologies.co.uk/support/vault-backups* *Markdown: https://perpetualtechnologies.co.uk/support/vault-backups.md* *Last checked 2026-09-07* Local vault backups run automatically every few hours whatever your settings say. Copying them off the machine is a separate thing that is off by default. 1. You do not have to switch local backups on. They run automatically every few hours regardless of any backup setting, and that is true on every edition including Free. 2. Copying those backups off the machine is the separate part, and it uses restic, a program NeuralVault does not install for you. 3. The Off-box backup card is deliberately not on the Settings page unless off-box backup is already enabled. On a normal install you will not find it, and that is the intended state rather than a missing screen. **How to tell it worked:** Nothing to check for local backups: they are not something you enable, so there is no switch that could be in the wrong position. **If it does not work:** - *You want off-box backup and cannot find the card.* It only renders when off-box backup is already enabled, so there is no way to turn it on from Settings in this build. It was taken off the default surface deliberately, because it can do nothing until restic is installed separately. - *You already had it on and want to change or stop it.* The card is still there in that case, on purpose: anything already running keeps its off switch. Settings, then System, and it sits under Background jobs and Safety. ### Watch & Learn, and the programs it needs first *Page: https://perpetualtechnologies.co.uk/support/watch-and-learn* *Markdown: https://perpetualtechnologies.co.uk/support/watch-and-learn.md* *Last checked 2026-09-07* A recorder that captures your screen and microphone so the assistant can learn a task by watching. It needs ffmpeg, and two optional extras. 1. Open Settings, then System, then Watch & Learn. 2. Install ffmpeg first. Screen and microphone capture will not work without it, and NeuralVault does not install it for you. The card has an install guide open by default. 3. Point the ffmpeg path at your install if it is not already found, then use the status line on the card to confirm it was detected: it reports the version and the path. 4. Transcription with whisper.cpp and keyframe descriptions with a local Ollama vision model are both optional. Leave them empty and the recorder still records. **How to tell it worked:** The ffmpeg status line says found, with a version and a path, rather than not checked yet. **If it does not work:** - *ffmpeg status stays on not checked yet.* Nothing has asked it to look. Save the card once, which triggers the check, rather than expecting it to poll on its own. - *The microphone device field does not match your hardware.* It takes the device name your system reports, not a friendly one you choose. Copy the name exactly as your operating system lists it. ### Index a repository so an agent can answer questions about it *Page: https://perpetualtechnologies.co.uk/support/code-graph* *Markdown: https://perpetualtechnologies.co.uk/support/code-graph.md* *Last checked 2026-09-07* The code graph turns a repository into symbols and call paths, so an agent can answer where something is defined instead of grepping. 1. Open Settings, then AI & extensions, then Code graph. 2. Put the path of the repository you want indexed into the field, for example the folder your project lives in. 3. Build the index. When it is done the card reports how many symbols it holds, and the commit it was built from. **How to tell it worked:** The card shows a symbol count, and names the commit the index was built from. **If it does not work:** - *The card says no code graph is configured.* That is the state before you point it at anything. Add a repository path and build it. - *The symbol count looks stale after you changed the code.* The commit shown beside the count is the one it was built from. If that is behind your working tree, the index is stale rather than wrong, and it needs rebuilding. ## Account and data Signing in, what your edition includes, and removing everything. ### Which screen answers a plan, licence or billing question *Page: https://perpetualtechnologies.co.uk/support/billing-usage-or-plan* *Markdown: https://perpetualtechnologies.co.uk/support/billing-usage-or-plan.md* *Last checked 2026-09-07* The app answers what your edition is. The website answers what you pay. There is no usage screen in this build. 1. In the app, open Settings, then General, then Licence. It names the edition this install is running, says whether the licence was signature-checked, and shows the licence ID and expiry date when there is one. 2. That same card is where you sign in. Signing in is the ordinary route to a paid edition: the licence arrives and then renews itself. There is a paste box below it for a signed licence file if you were sent one directly. 3. Money questions are answered on this website rather than in the app. Invoices, the payment method and cancellation all live under your account. 4. If a renewal is late, press Renew now on that card rather than waiting for the next scheduled check. **How to tell it worked:** Settings, General, Licence names your edition, and the badge beside it reads verified rather than not verified. **If it does not work:** - *You are looking for a Usage or a Managed usage screen.* Neither exists in this build. Model spend is billed by whichever provider your API key belongs to, on their own account, and the app does not total it for you. - *The badge says not verified.* The licence file was read without checking its signature. Development builds honour those and released builds ignore them, so it will stop granting anything. Sign in on that same card to replace it with a signed one. - *You want to cancel, or change a payment method.* Neither happens in the app. Both are on this website under your account, which hands off to the payment provider’s own portal. ### Paying for AI through NeuralVault instead of your own key *Page: https://perpetualtechnologies.co.uk/support/pay-for-ai-through-us* *Markdown: https://perpetualtechnologies.co.uk/support/pay-for-ai-through-us.md* *Last checked 2026-09-07* Not in the build you can download. Today you bring your own provider key, or run a model locally for nothing. 1. There is nothing to set up. The published version has no credit balance, no top-up and no Managed usage screen, so searching Settings for one will not find it. 2. Use your own provider key instead: Settings, then AI & extensions, then API keys, and add a key for whichever provider you want to pay. You are then billed by that provider directly. 3. If you would rather not open an account with a model provider at all, run a model on your own machine. That costs nothing, needs no key, and the Ollama guide covers it end to end. **How to tell it worked:** Settings, AI & extensions, API keys shows a provider with your key against it, and a message in Chat gets an answer. **If it does not work:** - *You read that you could buy AI credit through us.* It is not in this download. This page will not walk you through a screen you cannot open, so treat it as absent until a release says otherwise. - *You have no key and no local model, and Chat will not answer.* That is the expected state rather than a fault: nothing can run a model for you yet. Either add a provider key or install Ollama. ### Sign in, and see what your edition includes *Page: https://perpetualtechnologies.co.uk/support/sign-in-and-edition* *Markdown: https://perpetualtechnologies.co.uk/support/sign-in-and-edition.md* *Last checked 2026-09-06* Your edition comes from your account. You sign in; there is no licence file to install. 1. Open Settings, then General, then Licence. 2. Sign in with the email you used to buy. The screen then names your edition and lists what it includes. 3. Renewal happens from the same screen and keeps the install current. **How to tell it worked:** The screen names your edition, and the capability list matches what you paid for. **If it does not work:** - *You are looking for somewhere to paste a licence key.* There is nowhere, deliberately. Editions are tied to the account, not to a file you install. - *You are signed in but on the wrong edition.* Check the email matches the one that made the purchase. If it does, that is a billing question rather than an app one, and worth writing to us about. - *You are offline.* That is fine. Once signed in, the install knows what it may do with no network at all. Signing in is the only step that needs a connection. ### You paid, but the app still says Free *Page: https://perpetualtechnologies.co.uk/support/paid-but-still-says-free* *Markdown: https://perpetualtechnologies.co.uk/support/paid-but-still-says-free.md* *Last checked 2026-09-06* The app checks your plan on a schedule rather than the instant a payment clears, so force the check instead of waiting for it. 1. Open Settings, then General, then Licence, and use the renewal control. That forces the check rather than waiting for the next scheduled one. 2. If it still reports Free, sign out and back in. Signing in forces the same check. 3. Open your account on the website and confirm the subscription is actually shown as active there. **How to tell it worked:** The Plan and sign-in screen names your paid edition, and the capability list grows to match it. **If it does not work:** - *The website shows the subscription but the app does not.* That is worth telling us about, and it is the one case where the detail matters: say what the account page shows and what the app shows. Those two disagreeing is a fault on our side, not something you did wrong. - *You paid with a different email than you signed in with.* The edition follows the account that made the purchase. Sign in with that email. ### An update did not apply, or the app cannot find one *Page: https://perpetualtechnologies.co.uk/support/update-did-not-apply* *Markdown: https://perpetualtechnologies.co.uk/support/update-did-not-apply.md* *Last checked 2026-09-06* Check for updates in Settings, About. Installs from before 30 August 2026 ran a different underlying build and cannot update themselves in place. 1. Open Settings, then General, then About, and check for updates there first. 2. If nothing is offered and your version predates 30 August 2026, uninstall and install the current version by hand. That older build has no path to update itself. 3. Your vault is not touched by an uninstall. Your notes are your own files in your own folder. **How to tell it worked:** The About screen reports the new version number after restarting. **If it does not work:** - *After updating, your theme, vault folder or provider choices are gone.* They should all survive an update exactly as you left them. If they did not, that is a bug rather than something you did, and worth telling us. ### Anonymous usage reporting: there is none in this build *Page: https://perpetualtechnologies.co.uk/support/usage-reporting* *Markdown: https://perpetualtechnologies.co.uk/support/usage-reporting.md* *Last checked 2026-09-07* The version you can download collects no telemetry and no analytics, so there is no switch to find and nothing to turn off. 1. There is nothing to do, and that is the whole answer. This build contains no usage reporting and no switch for it anywhere in Settings. 2. If you would rather check than take our word for it, the privacy notice on this site states the same thing for this exact version, and the app repeats it in its own onboarding. 3. The only thing that leaves your machine on purpose is a bug report you fill in yourself, and the app shows you that text in full before it opens anything. **How to tell it worked:** There is no privacy, telemetry or usage toggle anywhere in Settings, because there is nothing for one to control. **If it does not work:** - *You are looking for a Privacy choices screen.* There is not one in this build. What you may be after is Settings, General, Remove my data, which is a different thing: it deletes what NeuralVault has written on this machine. - *A later version offers to turn usage reporting on.* Then it is a new feature in that version rather than something that was hidden here. If it is ever added it is off on a fresh install, off after an update, and stays off until you turn it on. ### Remove everything NeuralVault has put on this machine *Page: https://perpetualtechnologies.co.uk/support/remove-my-data* *Markdown: https://perpetualtechnologies.co.uk/support/remove-my-data.md* *Last checked 2026-09-06* Settings, General, Remove my data lists what exists first, and removes nothing until you type the confirmation exactly. 1. Open Settings, then General, then Remove my data. 2. Ask it to list what is there. Nothing is read until you press it, because this screen touches real files and real credentials. 3. Read the list. It covers app state and stored credentials, not your vault: your notes are your files, and removing them is something you do yourself, in Explorer or Finder. 4. Type the confirmation phrase exactly to proceed. **How to tell it worked:** Re-run the listing afterwards. What was removed is gone from it. **If it does not work:** - *You want your notes deleted too.* Delete the vault folder yourself. The app deliberately will not delete the folder it does not own, because a local-first product that silently removes your documents is the worst possible failure mode. - *The confirmation is not accepted.* It has to match exactly, including case. That is the safety catch: there is no boolean a mis-wired button could pass instead. --- # Symptom index Every problem above, listed by what the person actually sees. 65 entries. Browsable at https://perpetualtechnologies.co.uk/support/questions - **Ollama is installed but no models are listed.** Installing Ollama does not download a model. Run "ollama list" in a terminal. If it is empty, pull one. Full guide: Run a model on your own machine with Ollama — /support/connect-ollama - **The app cannot reach Ollama.** Open http://localhost:11434 in a browser. If that does not respond, Ollama is not running, and no setting in NeuralVault can fix that. Restart it, then reopen the model picker. Full guide: Run a model on your own machine with Ollama — /support/connect-ollama - **Answers are extremely slow.** That is a hardware answer, not a configuration one. A model larger than your available memory will be swapped to disk. Try a smaller variant. Full guide: Run a model on your own machine with Ollama — /support/connect-ollama - **The model list is empty but the server is running.** Nearly always a missing or extra /v1 on the base URL. Open /models in a browser: if that returns JSON, the URL is right. Full guide: Connect LM Studio, or any other OpenAI-compatible server — /support/connect-local-openai-compatible - **It works in a browser but not from NeuralVault.** Some servers bind to 127.0.0.1 only and some to all interfaces. Use exactly the host the server printed, not a substitute you assume is equivalent. Full guide: Connect LM Studio, or any other OpenAI-compatible server — /support/connect-local-openai-compatible - **The key saves but every message fails.** Stored and valid are different things. Check the key has credit or an active billing method on the provider’s side, and that it has not been scoped to a project that excludes the model you picked. Full guide: Add an API key for a cloud provider — /support/add-provider-api-key - **You want to know where the key actually went.** Into your OS credential store under the service name NeuralVault: Credential Manager on Windows, Keychain on macOS, the Secret Service on Linux. It is not written into your vault and it is not synced anywhere. Full guide: Add an API key for a cloud provider — /support/add-provider-api-key - **You need to remove a key.** Delete it from the same screen. Removing your data wholesale is a separate control, in Settings, General, Remove my data. Full guide: Add an API key for a cloud provider — /support/add-provider-api-key - **Requests return 404.** The path template does not match your endpoint. Azure deployments in particular sit under a deployment-specific path rather than the plain /chat/completions most OpenAI-compatible services use. Full guide: Connect Azure OpenAI, Z.ai, or another direct endpoint — /support/connect-azure-or-direct-endpoint - **Requests return 401 with a key you know is good.** Auth style. Some endpoints want a bearer token and some want a provider-specific header. Match what the provider documents, not what another provider uses. Full guide: Connect Azure OpenAI, Z.ai, or another direct endpoint — /support/connect-azure-or-direct-endpoint - **A model you expected has disappeared.** That is the filter working. Its vendor’s published policy does not meet the level you selected. Loosen the chip to see it again. Full guide: Pick a model, and filter by what the vendor does with your data — /support/choose-a-model-and-its-data-policy - **Local models are still listed under the strictest setting.** Deliberate. A model running on your own machine sends nothing to a vendor, so no vendor policy applies to it and there is nothing for the filter to exclude. Full guide: Pick a model, and filter by what the vendor does with your data — /support/choose-a-model-and-its-data-policy - **You cannot find Brave in the "Add a key" list.** It is intentionally kept out of that dropdown and given its own row, so that a missing search key never hides behind a missing model key. Look below the model providers. Full guide: Let the assistant search the web — /support/let-the-assistant-search-the-web - **A model you expected is not in the list for a level.** The list is built from the providers you have configured. Add the provider key first, under API keys, and the models appear here afterwards. Full guide: Model tiers: which model runs at which level — /support/model-tiers - **The change did not seem to affect the answer you just got.** It applies from the next turn. A turn already in flight finishes on the model it started with. Full guide: Model tiers: which model runs at which level — /support/model-tiers - **The test fails immediately.** The command is what a shell would run. Check the program is on your PATH, and that the arguments are separate entries rather than one string with spaces in it. Full guide: Connect an MCP server — /support/connect-an-mcp-server - **The test succeeds but the tools do not work later.** Almost always missing credential values. Declaring the variable name is only half of it. See "Give a connector its credentials". Full guide: Connect an MCP server — /support/connect-an-mcp-server - **You granted a tool but an agent still will not call it.** Granting and marking read-only are separate ticks. A granted tool that is not marked read-only waits for your approval before an agent may run it, which is the deliberate default. Full guide: Connect an MCP server — /support/connect-an-mcp-server - **You cannot see the credential rows.** They only exist once the connector is saved with variable names declared. An unsaved form has nothing to attach a credential to. Full guide: Give a connector its credentials — /support/connector-credentials - **You want to check what you typed.** You cannot, by design. Re-enter it rather than hunting for it. Nothing in the app displays a stored credential value. Full guide: Give a connector its credentials — /support/connector-credentials - **There is no Plugins card in Settings at all.** It is hidden rather than empty. The plugin API needs Personal or above, and on Free the whole card is left out instead of shown and refused. Full guide: Plugins: what they are, and how one connects — /support/plugins-in-this-build - **Your plugin does not appear in the list.** It has not connected. The list stays empty until a plugin program is running and has registered; starting NeuralVault does not start your plugin for you. Full guide: Plugins: what they are, and how one connects — /support/plugins-in-this-build - **A plugin holds a capability you did not mean to give it.** Revoke it there. What a plugin asked for and what it was granted are shown separately, so check the granted column rather than assuming the request was met in full. Full guide: Plugins: what they are, and how one connects — /support/plugins-in-this-build - **You want to switch a tool off from this screen.** This pane reports, it does not control. What an agent may actually reach is decided by connector grants (Settings, AI & extensions) and by the approval step before a write. Full guide: See what an agent is allowed to do — /support/what-an-agent-is-allowed-to-do - **You are looking for a way to add or remove a calendar source.** There is not one in the app yet. The backend publishes no command to list, add or remove a source, so a control here would be a button with nothing behind it. Full guide: Where your calendar events come from — /support/where-calendar-events-come-from - **The Calendar page is not there at all.** It is a Business edition feature. Check Settings, General, Licence for which edition this account is on. Full guide: Where your calendar events come from — /support/where-calendar-events-come-from - **You are still asked about a host you added.** The entry has to match the host the tool actually calls. A subdomain is a different host, so allowing the parent does not cover it. Full guide: Outbound hosts: where tools are allowed to send requests — /support/outbound-hosts - **You would rather not be asked at all.** There is no off switch for the prompt in this build, and the empty list is what makes it safe by default. Add the hosts you use and the asking stops for those. Full guide: Outbound hosts: where tools are allowed to send requests — /support/outbound-hosts - **There is no Sync & calendar group in Settings.** The whole category is hidden when your edition includes neither device sync nor calendar, rather than shown and refused on save. Full guide: Sync a vault between your own machines — /support/sync-across-devices - **You expected sync without running anything yourself.** There is no hosted option in this build. Sync talks to an nv-sync-server instance on your own network, so something has to be running it. Full guide: Sync a vault between your own machines — /support/sync-across-devices - **The app opens and everything is empty.** This is nearly always the vault pointing somewhere other than where you think. Read the path on the Vault screen first, before assuming notes were lost. They are files on disk; if the folder still has them, nothing is gone. Full guide: Find, choose or move your vault — /support/where-your-vault-lives - **You moved the folder yourself and now it is empty.** Point the app at the new location from the Vault screen. NeuralVault does not follow a folder you move behind its back. Full guide: Find, choose or move your vault — /support/where-your-vault-lives - **You wrote CORTEX.md but the assistant ignores it.** It has to be at the ROOT of the vault. A CORTEX.md inside a subfolder is an ordinary note. Full guide: Write standing instructions the assistant always follows — /support/standing-instructions - **You have both a CORTEX.md and a CLAUDE.md.** Only the first of CORTEX.md, CLAUDE.md, AGENTS.md that exists and is not blank is used. Never both. A blank CORTEX.md is skipped, so a written CLAUDE.md beside it is what takes effect. Full guide: Write standing instructions the assistant always follows — /support/standing-instructions - **You are looking for a per-skill on/off switch.** There is not one, anywhere in the product. The Enabled column is a status, not a control. Opting in happens once per directory, and narrowing what a particular agent may use is done per agent. Full guide: Skills: what they are and where they come from — /support/skills - **It looks right now but resets when you reopen.** The change applies to the screen immediately and is saved in the background, so a display that is correct proves the first half only. A reset means the write failed, and that is worth telling us about. Full guide: Change the theme and how tightly packed it is — /support/theme-and-density - **The switch is on and you still see nothing.** Your operating system has its own permission for notifications, separate from this one. Check that NeuralVault is allowed to notify you in your system settings. Full guide: Turn native desktop notifications on or off — /support/notifications - **You expected the Connections group to hold your connectors.** It does not. Connections holds notifications and hotkeys only. Connectors live under AI & extensions, which is the single most confusing thing about this Settings page. Full guide: Turn native desktop notifications on or off — /support/notifications - **You want to change or rebind one.** You cannot in this build. The card says so in as many words: they are fixed for now. There is no settings file that overrides them either. Full guide: Keyboard shortcuts, and why you cannot change them — /support/hotkeys - **A shortcut does nothing when the app is not focused.** That is expected. These work inside the app only, so they are not global shortcuts and will not fire from another window. Full guide: Keyboard shortcuts, and why you cannot change them — /support/hotkeys - **A banner says background scheduling is paused in this window.** Your switches still save. They take effect wherever the active scheduler is running, which is a different window or a different machine, so the change is not lost. Full guide: Background jobs, and turning one off — /support/background-jobs - **A job says no run recorded yet and never changes.** It has not had a reason to run, or the scheduler is paused as above. The time shown is the last actual run, not a countdown to the next one. Full guide: Background jobs, and turning one off — /support/background-jobs - **A row says it could not load.** The card reports what it can and names the error on the row that failed, so the other two are still trustworthy. It is a read failure rather than a sign that something was approved without you. Full guide: The Safety screen: rewrites, security events and audit logs — /support/safety-queues - **There is no Review button beside a row.** That review page is not available on your edition. The count is still shown, so you can see there is something there. Full guide: The Safety screen: rewrites, security events and audit logs — /support/safety-queues - **You want off-box backup and cannot find the card.** It only renders when off-box backup is already enabled, so there is no way to turn it on from Settings in this build. It was taken off the default surface deliberately, because it can do nothing until restic is installed separately. Full guide: Backups: what runs on its own, and what does not — /support/vault-backups - **You already had it on and want to change or stop it.** The card is still there in that case, on purpose: anything already running keeps its off switch. Settings, then System, and it sits under Background jobs and Safety. Full guide: Backups: what runs on its own, and what does not — /support/vault-backups - **ffmpeg status stays on not checked yet.** Nothing has asked it to look. Save the card once, which triggers the check, rather than expecting it to poll on its own. Full guide: Watch & Learn, and the programs it needs first — /support/watch-and-learn - **The microphone device field does not match your hardware.** It takes the device name your system reports, not a friendly one you choose. Copy the name exactly as your operating system lists it. Full guide: Watch & Learn, and the programs it needs first — /support/watch-and-learn - **The card says no code graph is configured.** That is the state before you point it at anything. Add a repository path and build it. Full guide: Index a repository so an agent can answer questions about it — /support/code-graph - **The symbol count looks stale after you changed the code.** The commit shown beside the count is the one it was built from. If that is behind your working tree, the index is stale rather than wrong, and it needs rebuilding. Full guide: Index a repository so an agent can answer questions about it — /support/code-graph - **You are looking for a Usage or a Managed usage screen.** Neither exists in this build. Model spend is billed by whichever provider your API key belongs to, on their own account, and the app does not total it for you. Full guide: Which screen answers a plan, licence or billing question — /support/billing-usage-or-plan - **The badge says not verified.** The licence file was read without checking its signature. Development builds honour those and released builds ignore them, so it will stop granting anything. Sign in on that same card to replace it with a signed one. Full guide: Which screen answers a plan, licence or billing question — /support/billing-usage-or-plan - **You want to cancel, or change a payment method.** Neither happens in the app. Both are on this website under your account, which hands off to the payment provider’s own portal. Full guide: Which screen answers a plan, licence or billing question — /support/billing-usage-or-plan - **You read that you could buy AI credit through us.** It is not in this download. This page will not walk you through a screen you cannot open, so treat it as absent until a release says otherwise. Full guide: Paying for AI through NeuralVault instead of your own key — /support/pay-for-ai-through-us - **You have no key and no local model, and Chat will not answer.** That is the expected state rather than a fault: nothing can run a model for you yet. Either add a provider key or install Ollama. Full guide: Paying for AI through NeuralVault instead of your own key — /support/pay-for-ai-through-us - **You are looking for somewhere to paste a licence key.** There is nowhere, deliberately. Editions are tied to the account, not to a file you install. Full guide: Sign in, and see what your edition includes — /support/sign-in-and-edition - **You are signed in but on the wrong edition.** Check the email matches the one that made the purchase. If it does, that is a billing question rather than an app one, and worth writing to us about. Full guide: Sign in, and see what your edition includes — /support/sign-in-and-edition - **You are offline.** That is fine. Once signed in, the install knows what it may do with no network at all. Signing in is the only step that needs a connection. Full guide: Sign in, and see what your edition includes — /support/sign-in-and-edition - **The website shows the subscription but the app does not.** That is worth telling us about, and it is the one case where the detail matters: say what the account page shows and what the app shows. Those two disagreeing is a fault on our side, not something you did wrong. Full guide: You paid, but the app still says Free — /support/paid-but-still-says-free - **You paid with a different email than you signed in with.** The edition follows the account that made the purchase. Sign in with that email. Full guide: You paid, but the app still says Free — /support/paid-but-still-says-free - **After updating, your theme, vault folder or provider choices are gone.** They should all survive an update exactly as you left them. If they did not, that is a bug rather than something you did, and worth telling us. Full guide: An update did not apply, or the app cannot find one — /support/update-did-not-apply - **You are looking for a Privacy choices screen.** There is not one in this build. What you may be after is Settings, General, Remove my data, which is a different thing: it deletes what NeuralVault has written on this machine. Full guide: Anonymous usage reporting: there is none in this build — /support/usage-reporting - **A later version offers to turn usage reporting on.** Then it is a new feature in that version rather than something that was hidden here. If it is ever added it is off on a fresh install, off after an update, and stays off until you turn it on. Full guide: Anonymous usage reporting: there is none in this build — /support/usage-reporting - **You want your notes deleted too.** Delete the vault folder yourself. The app deliberately will not delete the folder it does not own, because a local-first product that silently removes your documents is the worst possible failure mode. Full guide: Remove everything NeuralVault has put on this machine — /support/remove-my-data - **The confirmation is not accepted.** It has to match exactly, including case. That is the safety catch: there is no boolean a mis-wired button could pass instead. Full guide: Remove everything NeuralVault has put on this machine — /support/remove-my-data