Astraios FAQ
Direct answers, checked against the code and the project facts on .
What is Astraios?
Astraios (Astraios Astrophotography) is a free, open-source desktop application for calibrating, registering, stacking and processing deep-sky, planetary and solar astrophotography on Windows, Linux and macOS. It was called Cosmica until June 2026. It is not related to BigCode's Astraios models or the EU ASTRAIOS project.
Is Astraios free and open source?
Yes. It is licensed under the GNU GPL, version 3 or (at your option) any later version (SPDX GPL-3.0-or-later). There is no paid tier, licence key or account.
Is it a PixInsight alternative?
For calibration through final processing of deep-sky data, yes: it covers the same workflow with a guided wizard and no licence fee, with built-in deconvolution and AI denoise. It is alpha software and does not match PixInsight's maturity or ecosystem. See the comparison page for a sourced, dated table.
How does it differ from Siril?
Both are free and GPL. Siril is mature, stable and CPU-based, with GPU use only inside external tools it can call. Astraios runs its own engine on CUDA or Apple Metal, bundles AI denoise and deconvolution, and offers a guided workflow, at alpha maturity. On the same subs with matched settings both produce the same star peaks and sharpness.
Which operations use the GPU?
Registration (star detection, matching, warping), pixel rejection and integration, drizzle, wavelet, chroma and TGV denoise, AI denoise and super-resolution, Richardson-Lucy, spatially varying and multi-frame deconvolution, wavelet sharpening, convolutions, CLAHE, background extraction, stretches and curves. Master calibration combination, non-local-means denoise, plate solving and catalog lookups run on the CPU by design.
Does it work without NVIDIA hardware?
Yes. Device selection is automatic: CUDA, then Apple Metal, then CPU. Every tool works on a CPU-only machine with the same output, only slower. AMD and Intel GPUs are not accelerated.
Does it support Apple Silicon?
Yes. The macOS installer creates Astraios.app, and Apple Silicon Macs run the GPU stages through PyTorch's Metal (MPS) backend. Intel Macs run on the CPU. Both are validated on real Apple hardware in CI. macOS 11 or later is required.
Does it upload images or collect telemetry?
No telemetry, analytics or crash reports are collected. Processing and AI inference run locally. The only feature that uploads image data is the astrometry.net plate-solving backend, and only when you select it and enter an API key; ASTAP and the local Gaia solver work offline. Other network use is on demand: the startup update check (can be turned off), model and catalog downloads, and catalog queries to SIMBAD and VizieR.
Which formats are supported?
Read: FITS, XISF, TIFF, PNG, JPEG (OSC frames are debayered) and SER video for planetary work. Write: FITS, XISF, TIFF 8/16-bit, PNG 8/16-bit, JPEG. Camera raw files (CR2, NEF, ARW, DNG) are not supported; convert them to FITS or TIFF first.
Do the installers require Python or manual CUDA setup?
No. The Windows installer, the Linux and macOS scripts and the AppImage bring their own Python environment, detect an NVIDIA driver and download the matching PyTorch build (CUDA 12.8 or CPU). Expect a download of about 2 GB for the CUDA build.
Is Astraios production ready?
No. It is alpha software under active development. Workflows are complete end to end but bugs are expected; tagged releases are more stable than the main branch.
What is CI-tested versus user-verified?
CI validates the installers on every packaging change on hosted Windows, Linux, macOS Apple Silicon and macOS Intel runners, all without an NVIDIA GPU, and runs the full test suite on a CPU-only Linux runner. The CUDA path is verified only on the maintainer's machine (NVIDIA RTX 5060 Laptop GPU under Bazzite Linux). Reports from other real machines are collected through the compatibility report form on GitHub.
Something missing? Ask in GitHub Discussions.