PreviewEarly access is rolling out in batches

More than a conversation.
A lifetime that adds up.

OpenMiRA understands your photos, videos, voice recordings and documents on your own device, and keeps building on them.
Your data never leaves the device.

How teams and enterprises use itHow individuals and families use it
Runs locally · Works offline · No cloud account · x86 / ARM / RISC-V
Any MCP-capable agent can call it directly →
Built on mature open-source inference engines llama.cppONNX RuntimeMLX
Start here

Which one are you?

Teams & enterprise · Work that ships
Everything your team knows,
turned into work it can deliver.
Not another knowledge base: answers have sources, people review the results, finished work returns to the library.
Explore Teams & enterprise →
Personal & Family · Lifelong memory
A person's life, a family's life,
kept on your own device.
Photos, videos, voices, documents and everyday life on one timeline.
Explore Personal & Family →
The loop

From files to published work,
one loop, on your own device.

Library, knowledge base, creation and publishing share one body of local data. Finished work returns to the library, and how it performs shapes the next piece.

01
Library
Documents, images, audio and video in one local index. Your files stay where they are.
→ Structured understanding
02
Knowledge base
Long-term memory of people, events and organizations, with a source on every conclusion.
→ Grounding and consistency
03
Creation
Generate images, video, reports and copy on-device, grounded in your own material.
→ Spec-ready output
04
Publishing
Review, schedule and publish to each platform’s specs, then bring the results back.
↩ Results flow back
Filed back · Performance flows back
ThroughoutPeople reviewOutbound ledgerRuns locally
Examples

All made on-device.

Examples, generated on-device by OpenMiRA.

Video generation
Video generation
Video generation
Video generation
Retouching · time of day and weather
Retouching · time of day and weather
Image generation
Image generation
Retouching · sky swap
Retouching · sky swap
Line art · colorized
Line art · colorized
01 · What it always does

It remembers. It cites.
It delivers. You decide.

These four capabilities apply to individuals, families and teams alike, and they’re what sets it apart from a typical AI assistant.

01
Remembers
Not a single conversation, but context that builds up over time. Every round of organizing, Q&A and creation is kept, so the longer you use it, the better it fits your material.
Time Machine · People · Team memory
02
Cites sources
Answers only from your files, and every conclusion traces back to the original: if it can be checked, it can be trusted.
Sources [1] [2] [3]
03
Delivers
More than answers: it plans tasks, runs long ones in the background, and delivers reports, sheets, albums and films.
Reports · Sheets · Albums · Clips
04
You decide
People decide who is who, which records merge, and whether results are accepted; the AI only suggests.
Confirm · Accept · Rework
02 · Formats

Photos, videos, audio, documents.
A lifetime of files, one index.

Text is searchable right away; images, audio and video are understood step by step in the background. Original files stay where they are; it only builds an index.

Docspdf docx doc xlsx xls pptx ppt md txt
Imagesjpg jpeg png heic webp
Audiomp3 wav m4a flac
Videomp4 mov mkv
03 · Developers

A complete agent platform
with open interfaces.

OpenMiRA ships with its own agents and apps; the same capabilities are open to external agents and apps over MCP and HTTP API, with local or cloud models. Search, cited Q&A, long-running tasks and generation share one interface, with SSE for streaming results. Listens on localhost only by default.

Agents & developersAPI status
Terminal
# The token is in the token file in the data directory
TOKEN=$(cat <data-dir>/token)

# Search: natural-language query
curl http://127.0.0.1:8090/api/v1/search \
  -H "Authorization: Bearer $TOKEN" \
  -d '{"query": "payment terms in the supplier contract", "top_k": 5}'

# Ask: streamed, with sources
curl -N http://127.0.0.1:8090/api/v1/chat/stream \
  -H "Authorization: Bearer $TOKEN" \
  -d '{"messages": [{"role": "user", "content": "What was decided at the meeting last week?"}], "search": true}'
POST /api/v1/search
Search
POST /api/v1/chat/stream
Ask · streaming
POST /api/v1/generate
Images & video
GET /hub/index.json
Scenario catalog
04 · Platforms

Runs on the hardware you already have.

The same capabilities, on different chip architectures. Underneath are open-source inference engines such as llama.cpp, ONNX Runtime and MLX.

Supported platforms →
x86
Desktops, workstations and small servers. Faster with a dedicated GPU.
ARM
Macs with Apple silicon, and ARM-based edge devices.
RISC-V
For the next generation of local AI chips.
05 · Privacy

Your files
stay on your device.

Data & privacy →
Works offline
Models run on your device. Offline, search and Q&A keep working as usual.
Cloud models are off by default
Whether to use cloud models is up to you; when they’re off, all processing stays on the machine.
Files stay where they are
It reads only the folders you point to, and doesn’t move files or change formats. If you stop using it, just uninstall; your original files are untouched.

Still early.
Let's build it well, together.

People who join now, users and developers alike, will help decide where it goes.

Early accessWrite a scenario
One email, sent only when a spot opens
PreviewEarly access is rolling out in batches

More than a conversation.
A lifetime that adds up.

Understands your photos, videos, voice recordings and documents on your own device, and keeps building on them. Your data never leaves the device.

How teams and enterprises use it How individuals and families use it
Local · Offline · x86 / ARM / RISC-V
Any MCP-capable agent can call it directly →
Start here

Which one are you?

Teams & enterprise · Work that ships
Everything your team knows, turned into work it can deliver.
Not another knowledge base: answers have sources, people review the results, finished work returns to the library.
Learn more →
Personal & Family · Lifelong memory
A person's life and a family's life, kept on your own device.
Photos, videos, voices, documents and everyday life on one timeline.
Learn more →
The loop

From files to published work,
one loop.

One body of local data. Finished work returns to the library, and how it performs shapes what you make next.

01
Library
Documents, images, audio and video in one local index. Your files stay where they are.
→ Structured understanding
02
Knowledge base
Long-term memory of people, events and organizations, with a source on every conclusion.
→ Grounding and consistency
03
Creation
Generate images, video, reports and copy on-device, grounded in your own material.
→ Spec-ready output
04
Publishing
Review, schedule and publish to each platform’s specs, then bring the results back.
↩ Results flow back
ThroughoutPeople reviewOutbound ledgerRuns locally
01 · What it always does

It remembers. It cites.
It delivers. You decide.

Remembers
Not a single conversation, but context that builds up over time.
Cites sources
Answers only from your files; every conclusion traces back to the original.
Delivers
Runs in the background and delivers reports, sheets, albums and films.
You decide
People decide who is who and whether results are accepted.
02 · Developers

A complete agent platform with open interfaces.

Ships with its own agents and apps; open to external callers over MCP and HTTP API.

Terminal
curl 127.0.0.1:8090/api/v1/search \
  -H "Authorization: Bearer $TOKEN" \
  -d '{"query": "payment terms"}'

Still early.
Let's build it well, together.

Early access