< Back

Data & Privacy

What taraengine does with your data: nothing leaves your machine, except the two downloads you ask for.

Your prompts stay on your machine. taraengine runs the models locally, on your own hardware. Your prompts, your files and the models' answers never leave your computer.

Privacy-First Architecture

taraengine sends no usage data, makes no update checks and files no crash reports. It talks to two hosts, and only when you run a command that needs them: huggingface.co on taraengine pull, and github.com on taraengine runtime install and runtime update. Both redirect downloads to their own content delivery hosts. Nothing else leaves the machine. A Hugging Face token is sent to Hugging Face only, and it is never logged.

What stays local (never sent anywhere)
  • Your prompts and the models' answers - Every request to /v1 is served on your machine, and taraengine does not log prompts or answers
  • Your files and images - A model reads what you send it, on your hardware, and nothing is kept
  • Your models - The files you pull, their presets and the state the engine keeps about them
  • Your logs - The engine's and the server's logs are in your data directory, readable by you only

Local Storage

taraengine keeps everything in one data directory: ~/Library/Application Support/taraengine on macOS, and $XDG_DATA_HOME/taraengine (default ~/.local/share/taraengine) on Linux.

Path Purpose
runtime/<build>/ llama-server, its libraries, upstream's LICENSE and the manifest the engine checks at start
models/ GGUF files by owner and repository, links to local files, the index
presets.ini The llama-server presets, generated on every change
state.json Faulty-model marks, remembered context windows, the last model loaded
child.key, child.pid, lock The supervised server's generated key (mode 0600), its process id, the daemon's lock
logs/ taraengine.log and llama-server.log, rotated at 20 MB; no prompts, no answers

Open Source

taraengine is fully open source under the MIT License. You can inspect, modify, and distribute the code freely. It is built on llama.cpp by ggml-org (MIT): taraengine runtime install downloads an unmodified upstream build, and upstream's LICENSE stays next to it in the data directory.

Benefits of open source
  • Full transparency - See exactly what the code does
  • No hidden telemetry - Verify there is no data collection
  • Verified downloads - Every release carries checksums, a keyless signature and provenance
  • Self-hosting - Run everything on your own infrastructure
Source code

Data Flow Diagram

Everything stays on your machine. The two dotted arrows are the only connections the engine ever makes, and only when you run the command.

Data Flow Architecture
YOUR MACHINE (100% Local - Nothing Sent Anywhere) ┌────────────────────────────────────────────────────────────────┐ │ │ │ ┌───────────────────┐ ┌──────────────────────────┐ │ │ │ your client │ ───> │ taraengine │ │ │ │ taracode, curl, │ <─── │ 127.0.0.1:8272 │ │ │ │ an SDK │ └──────────┬───────────────┘ │ │ └───────────────────┘ │ loopback port │ │ v │ │ ┌──────────────────────────┐ │ │ │ llama-server │ │ │ │ your models on disk │ │ │ └──────────────────────────┘ │ │ │ │ PROMPTS, ANSWERS, FILES: THEY STAY HERE │ └────────────────────────────────────────────────────────────────┘ ╎ only on taraengine pull ╎ only on runtime install v v huggingface.co (model files) github.com (the pinned llama.cpp build)

Your Rights

You have full control over your data.

Access Your Data

Everything is in the data directory on your machine: plain GGUF files, JSON state and text logs. You have full control.

Delete Your Data

  • A model: taraengine rm <model> deletes it and its sibling files
  • Everything: stop the daemon and delete the data directory
  • No account needed: taraengine is fully open source with no account requirement

Export Your Data

The models are standard GGUF files and the state is JSON; copy them wherever you like.

Contact Us

For privacy-related questions or data requests: