olenji
A photorealistic, fluffy orange cat reaches from behind an open laptop, its head at the upper right, gently biting the screen’s corner. The laptop screen faces you.

A soft spot for your words.

Transcribe. Translate. Make yourself at home.

A little about Olenji

Some words
are worth keeping.

A conversation. A half-finished thought. The recording you keep meaning to come back to. Olenji gives your words a place to land.

Import an audio or video file, follow live speech, correct the transcript and speakers, and translate with your agent or a configured provider. Read in your browser, work from the command line, or give your agent the same session through MCP.

At home on your machine.

The app, sessions and downloaded models stay in your local workspace. Local inference works offline once its model and runtime are installed. Export TXT, Markdown, SRT, VTT, HTML, JSON or JSONL when you’re ready.

What can leave my computer?

Translation can use your existing agent or an endpoint you configure. Remote providers receive the data you send them. A session’s local-only setting blocks remote recognition and translation inside Olenji; an external agent has its own data policy.

This website has no accounts, uploads, transcription service or visitor trackers. Install Olenji to work with your recordings. Read the privacy model ↗

Will it work on my machine?

The development build uses CPython 3.11–3.14 and FFmpeg. Portable packaging supports Linux, macOS and Windows with backend-specific limits. The Ubuntu checks cover source, installed packages, browsers, real speech and synthetic native audio; dependency simulations do not certify every OS, device or model.

Run olenji doctor to inspect your machine. Native audio needs OS permissions and a working backend. Accuracy depends on the model, language, audio and hardware; speaker labels are anonymous acoustic clusters, not verified identities.

How does it work with my agent?

Run olenji mcp serve to connect over stdio. Your agent can read committed text, write revision-aware translations, open a bilingual view and export subtitles. The CLI, browser and MCP tools share the same session.

A durable local service owns file jobs and native capture. Those can continue when a viewer closes. A paired browser microphone belongs to its tab and stops when that tab closes. Read the agent workflow ↗

Explore the handbook Get the development build
Make yourself at home
1.0.0 · Development build

Meet Olenji.

A local workspace for your conversations.
A slightly less local cat.

This early release is for people comfortable with a terminal. These are reviewed direct downloads; this version is not published to PyPI or npm registries.

Install with uv

Requires CPython 3.11–3.14 and FFmpeg / ffprobe on your PATH. CPython 3.12 is the recommended starting point. Need uv or FFmpeg? Start with the platform setup guide ↗.

01 Download these two files into one folder

Python wheel Frozen dependencies

02 In that folder, run

uv tool install --python 3.12 --with-requirements runtime-requirements.txt ./olenji-1.0.0-py3-none-any.whl
olenji doctor
olenji models install faster-whisper.tiny
olenji ui open

Model downloads are separate. Native capture and model availability depend on your OS permissions, backend and hardware. The handbook covers installation, troubleshooting and everyday use.

The complete toolkit.

Complete development bundleSource, packages, tests and documentationZIP ↓ Python source packagetar.gz ↓ npm launcher packagetgz ↓ SHA-256 checksumsVerify ↓ Artifact manifestJSON ↓
Verify the Python wheel

Compare your file’s SHA-256 with this reviewed value and the downloaded checksums.

863382819d095dec51167963d97e36ded766ed28d2989c99f37703197b7f1c22

Application code: Apache 2.0. Models and optional backends have their own licenses. The development bundle includes scripts/verify_ubuntu.py and docs/LOCAL_TESTING.md for comprehensive checks from Ubuntu.