Skip to content
    NewAnvira Notes 1.2.20: You get the number of questions you asked for

    Modular research and infrastructure for intelligence that belongs to you.

    AI will reshape how we work and live. LocalHouseLLM is building open systems that keep intelligence modular, persistent, and on hardware you control — not locked in someone else's cloud.

    Get the free apps Explore the stack

    Anvira

    Local-first workspace for chat, agents, and knowledge — on your machine.

    Read more
    • 0/21agent workflow checks passed by a 7B model in our own testing
    • 0research papers and technical reports, free to read
    • 0releases shipped in 37 days, each with public notes
    • 0notes or prompts sent to a cloud AI by default — models run on your PC

    Latest releases

    Learn more about our mission

    The idea

    Intelligence shouldn't live inside a single model.

    Models are only one part of an intelligent system. Ours separates the rest into layers — here is what happens to a single request, step by step.

    1. 01

      You ask

      Anvira or Anvira Notes sends the request to Anvira Runtime — a background service on your own PC, not a server in someone else’s cloud.

    2. 02

      The runtime checks permissions

      Each application only sees the files, memory and tools you have allowed it to use. Applications connect to one shared runtime instead of each embedding its own AI stack.

    3. 03

      Components talk in AICL

      Every hop between components is a structured AICL message, so a part can be inspected, replaced or upgraded without rewiring the rest.

    4. 04

      ORCHA plans the work

      ORCHA decides what to do next and which tools to use — reading and editing files, running commands, searching — and recovers when a smaller model returns a malformed tool call.

    5. 05

      Nomi brings the context

      Nomi supplies what is worth remembering from earlier work. Memory is private to each application unless you choose to share it.

    6. 06

      A local model reasons

      A GGUF model runs through llama.cpp on your hardware. A cloud provider is used only if you add one yourself.

    7. 07

      The result is verified

      Before anything is applied, ORCHA checks the result. In Anvira, file changes come back as a diff you can review and undo.

    AICL — the message layer between componentsAnvira · Anvira NotesApplicationAnvira RuntimeLocal API · permissionsORCHAPlans · routes · verifiesNomiMemoryToolsFiles · terminal · searchLocal modelGGUF via llama.cpp
    See the full architecture

    From the lab

    Our first products

    Meet Anvira and Anvira Notes.

    A local-first agent workspace and a notes-and-study app, both running on one shared local runtime — on your machine, with models you choose.

    New workspace Anvira Guide Approval

    Connect the findings in my interview notes with the model evaluation. What should we change?

    Anvira Guide

    The strongest pattern isn't model capability — it's continuity. Learners lose momentum whenever context must be rebuilt.

    Message Anvira…

    • Anvira Notes

      Version 1.2.20 · Windows 10 & 11 · 138 MB

      Notion-style notes and a study coach in one app: flashcards, quizzes and graded mock exams made from your own notes, Google Docs and Notion sync, and study groups.

      Download freeLearn more

    • Anvira

      Version 1.0.1 · Windows 10 & 11 · 131 MB

      A local AI chat and coding agent: point it at a project folder and it plans, writes and verifies the work — every change a diff you can undo.

      Download freeLearn more

    Local-first

    Intelligence should belong to the people using it.

    Local-first by default. Cloud when you choose it.

    ModelsWorkspaceMemoryKnowledgeAgents

    External connections only exist when you add them.

    Intelligence, built to belong.

    Building infrastructure for modular, persistent, executable intelligence.

    Try Anvira Developer docs