Local AI, without the friction
CereVault
A beautifully designed interface for working with local language models. Turn on Agent mode and your model can search the web, use your files and call tools, with every action shown and every sensitive step confirmed. Your chats stay local.
Screen recording of CereVault answering a question: the question is typed and sent, the app searches the web, writes an answer and lists the pages it used as sources.
Pick the right model for the job.
Switch mid-conversation. CereVault keeps the thread.
Screen recording: a question is answered by one model, the model is changed in the sidebar picker, and a follow-up question is answered by the new model in the same conversation.
It stops before your data leaves.
Emails, numbers and keys are flagged before a search is sent.
Screen recording: a question containing an email address is typed and sent, and CereVault shows a Send this search? confirmation that warns the email will be sent to DuckDuckGo.
Finds your models automatically.
Ollama, LM Studio and CereVault Serve show up on their own.
Screen recording of the Manage Models screen listing models discovered from CereVault Serve, Ollama and LM Studio on the local network.
A private AI assistant that runs on your own models.
Finds your models automatically
CereVault discovers AI servers on your network, including Ollama, LM Studio, and CereVault Serve, and connects in seconds.
Switch models mid-conversation
Change models at any time, even mid-conversation, without losing your flow.
Work with text and images
Attach screenshots, photos, and files. Vision-capable models can read, summarize, and extract information so you move faster.
Clean, readable output
Responses render with full Markdown, including tables, lists, and code blocks, and formatting is cleaned automatically so results are easy to use.
On-device voice dictation
Dictate messages with Apple's on-device speech recognition where supported, so your audio stays on your machine.
Private by default
Your chats stay on your device. No accounts. No tracking. No cloud required. Everything is already private.
Agent mode with web search
Your local model can search the web with no setup or API key, read files from folders you grant, and remember facts about you. Every tool action is shown inline.
Web search with no setup
Agent mode searches the web out of the box with no API key. Add a Brave Search key if you want a different provider. Keys and tokens move to the system Keychain.
Sources under answers
When a reply drew on web results, the sources it actually used are listed beneath it as tappable links.
Privacy check before data leaves
If a search or tool call contains an email address, phone number, card number, or API key, CereVault tells you and asks first.
Workspaces
Separate workspaces each keep their own tools, memory, folders, permissions, model, and chat history. Switching is instant.
MCP servers
Connect Model Context Protocol servers and their tools join the agent. Every third-party tool call asks for your confirmation first.
Automations
Schedule a saved prompt to run every few minutes, hourly, or daily. Results are saved as new chats, and runs are read-only by default.
Bring your own local model.
CereVault is the interface. You choose where the models run. Point it at a local provider like Ollama, LM Studio, or CereVault Serve, pick a model, and start chatting. Nothing leaves your device or your network unless you ask it to.
What you'll need
- macOS 15.2+, iOS / iPadOS 17+, or visionOS 1+
- A local model provider: Ollama, LM Studio, or CereVault Serve
- At least one downloaded model (browse the library)
- New here? Follow the setup guide
Get CereVault
Free until Nov 1. Available on the App Store for Mac, iPad, iPhone, and Vision.
