Skip to content

Releases: CodexManvik/Aethel

Aethel v2.0.0-beta.1

Aethel v2.0.0-beta.1 Pre-release
Pre-release

Choose a tag to compare

@CodexManvik CodexManvik released this 21 Oct 13:23

A beta. This is the first release of Aethel v2, a companion and computer-use agent that lives on your Windows PC. It's complete enough to use, but the evaluation work and first-run onboarding are still to come, so expect rough edges.

Install

  1. Download Aethel_2.0.0-beta.1_x64-setup.exe below.
  2. Optionally, check it against SHA256SUMS.txt:
    Get-FileHash .\Aethel_2.0.0-beta.1_x64-setup.exe -Algorithm SHA256
    It should print 33a42f79ecdfbbeb14795024fe30b4a234af654242c43429c611ed3dec44f13c.
  3. Run it. It installs to Program Files for every user of the PC and asks for admin rights once.
    • The installer isn't code-signed yet, so Windows SmartScreen may say it "protected your PC". Choose More info → Run anyway.
  4. Start Aethel from the Start menu.
    • The first launch sets up Aethel's own Python and packages in C:\Users\<you>\.aethel. That takes about 3 minutes and about 580 MB, needs the internet, and happens once. Every package is checked against a pinned hash.
    • Later launches take a few seconds.
  5. Give it a model. Either:
    • Settings → Providers: a key for Groq, Gemini or OpenRouter, or Add endpoint for your own server (LM Studio, Ollama, vLLM, another PC); or
    • Settings → Local model: download llama-server there and choose a .gguf, to run everything on this PC.

Your conversations, memory, personas, settings and the runtime live in C:\Users\<you>\.aethel. Uninstalling leaves them, so delete that folder to remove everything.

What's in it

  • Chat, and personas you create:
    • a drafted personality, a face picked from four portraits, a character sheet;
    • a voice blend, a style learned from a chat export, a tone engine;
    • photos that keep its face.
  • Calls: hands-free voice with barge-in.
  • Tasks on your PC:
    • files, shell, desktop control of any app, Office, and a background browser;
    • approvals, undo, and a replay timeline.
  • It learns: finished tasks become skills, and repeated ones compile into macros that call the LLM only when something changed.
  • Memory you can see and edit, and web search with citations.
  • Models anywhere:
    • hosted providers, your own OpenAI-compatible endpoints, or a local .gguf;
    • images from Gemini, OpenRouter, any OpenAI-compatible images server, ComfyUI with your own workflow, or local Z-Image.
  • Private mode keeps everything on this computer.

In-app extras, each pinned to a SHA-256:

  • Kokoro voices (about 350 MB);
  • llama.cpp's llama-server (GPU or CPU);
  • the face check (about 250 MB);
  • local image support (about 3 GB).

The full list is in CHANGELOG.md.

Known limits

  • Windows 10/11 x64 only.
  • The installer is unsigned, so expect the SmartScreen warning.
  • The background browser and Desktop Commander need Node.js. Without it they say so, and everything else works.
  • Some thresholds are marked "not measured" until the evaluation release.

Licence

  • Aethel is MIT licensed (LICENSE).
  • Models and tools it downloads keep their own licences: see docs/third_party.md.
  • In particular, the optional face check's ArcFace model derives from weights published for non-commercial research.