A personal AI system that runs on your machine.

VIRA is an open-source framework for building a personal AI. It watches your system and workspace, remembers your conversations, and lets agents use tools only with permission. Local models run through Ollama, and nothing leaves your computer unless you set it up that way.

VIRA is in early development. Linux is the main supported platform today.

Example chat Illustrative output

YouWhich directory are you running in, and is the repo clean?

VIRAI’ll check with two approved system tools.

  1. Read the working directory and platform Permitted
  2. Check the repository status Permitted

VIRAI’m running from ~/VIRA on Linux, and the working tree has no uncommitted changes.

2 of 2 tool calls permitted

What happens when you ask something

  1. You send a request

    Conversations and their messages are saved in a local database, so they survive restarts. You can rename, archive or delete them, or start an incognito chat that is never used as a memory source.

  2. VIRA assembles context

    Sensors watch the system, hardware, processes, network, workspace, projects and your activity. VIRA combines what they report with your saved memories into a bounded amount of context for the model.

  3. An agent plans and acts

    Agents work in a think, plan, act, reflect loop, and they can also react to events from the sensors.

  4. Tools run only with permission

    VIRA checks permissions before each tool call and records the request, the result, the decision and how long it took.

What VIRA is built around

Local first
There is no hidden telemetry and no cloud dependency. VIRA works with local models through Ollama. OpenAI and Anthropic adapters exist, but you have to configure them yourself.
Context at the center
The kernel is built around a constantly updated model of your machine, workspace, conversations and memories, not around a single prompt.
Permissioned tools
Local and MCP tools are checked before they run. For example, the shell tool only answers approved system questions such as the date, the working directory or repository status. It does not run arbitrary commands.
Open and modular
VIRA is MIT licensed. Agents, sensors, tools, model providers and storage are separate modules, so you can extend one without rewriting the rest.

Run it locally

You need Python 3.11 and Ollama running. A Docker setup is described in the README.

git clone https://github.com/Vishalsng112/VIRA.git
cd VIRA
pip install -r requirements.txt
python run.py

Then open http://localhost:8000. The dashboard shows runtime health, metrics, modules and the live event stream, and /chat opens the chat workspace.

Where the project is

VIRA is developed in the open. The commit history and project board show what changed recently and what’s planned.

Now
The runtime foundation. The kernel, sensors, conversations, scoped memory, agents, permissioned tools and a local dashboard work end to end. Sensors mostly poll today.
Next
Personalization and hardening. Modeling your routines, better memory search, durable and resumable runs, event-driven sensing, sandboxed agents, and Windows and macOS support through containers.
Later
Tools and agents from a description. Describe what you need, and VIRA would generate an agent or tool, verify it, and show its permissions for your approval before it runs. This is planned, not built yet.

Build on VIRA

Write an agent, add a sensor, fix a bug or improve the docs. Each layer is its own module, so you don’t need to understand the whole system to improve one part. If you’re unsure where to start, open an issue with the question label.