Everything to try Eldric.
The platform for your server, the app for your Mac, and models you can test on your own data. All files are served from repo.eldric.ai.
Install with one command.
Fedora 43 and Fedora 44, on x86_64 and aarch64; the GPU package is built for Fedora 43 on x86_64. Fedora 42 and RHEL-family systems are not supported at the moment, and the script stops on them with a reason. The script adds the package repository, installs Eldric and starts it; updates then come through dnf.
Then follow Get started: the first user created in the chat becomes the administrator.
Packages directly
CPU package, 5.0.177: eldric-aios-5.0.177-1.fc43.x86_64.rpm (22.2 MB).
GPU package for NVIDIA hosts, 5.0.177: eldric-aios-cuda-5.0.177-1.fc43.x86_64.rpm (821.3 MB).
aarch64 (for example Raspberry Pi 4/5): eldric-aios-5.0.177-1.fc43.aarch64.rpm (21.1 MB), eldric-aios-5.0.177-1.fc44.aarch64.rpm (20.8 MB).
Checksums for every file are listed below.
Verify and inspect
Signing key: RPM-GPG-KEY-eldric.
The install scripts: install-eldric.sh (full install), setup-dnf-repo.sh (repository only). Read before you run.
Eldric for your Mac.
The native app and command-line client, signed with an Apple Developer ID and notarised. It updates itself once installed.
Eldric Nexus F1, a time-series forecaster.
A zero-shot forecaster, pretrained by us from scratch: give it a history of values and it forecasts the next steps, with nine quantiles from q10 to q90, without training on your data first. 82.5 million parameters; it runs on a CPU or an Apple Silicon Mac, and uses an NVIDIA GPU only if you ask it to. Python 3.11 to 3.13.
Try it in three commands
The second command fetches the weights once and verifies their checksum. The README has the Python API and a worked example.
Nexus F1
Package (Python wheel) · 70.2 KB
Weights · 314.8 MB
README — install, CLI, Python API, licence · 6.9 KB
Runnable example · 3.1 KB
Nexus F1 — second checkpoint
Package (Python wheel) · 70.7 KB
Weights · 314.9 MB
Licence: the code is Apache-2.0. The trained weights are made available for benchmarking, evaluation and research; for production or commercial use, write to license@eldric.ai.
For retrieval on your own hardware.
Eldric uses an embedding model to make your documents searchable. These model files are available for download.
bge-m3 (GGUF, F16)
1.1 GB
bge-m3 is by BAAI, under the MIT licence.
Every file, with its checksum.
Compare a download with sha256sum <file> (Linux) or shasum -a 256 <file> (macOS). This list was generated 2026-10-08 15:50 UTC from the files repo.eldric.ai serves.
Linux packages (Fedora 43 and 44; x86_64 and aarch64)
| eldric-aios-5.0.177-1.fc43.x86_64.rpm eldric-aios, any x86_64 server (Fedora 43 build; Fedora 44 installs its own build via install.sh) | 22.2 MB | a87c3e1e59b3bbc96e8fc65460d4562a2b0f98f0b9bf398102a7d63ab1e64e24 |
| eldric-aios-cuda-5.0.177-1.fc43.x86_64.rpm eldric-aios-cuda, CUDA 12.8 (Fedora 43) | 821.3 MB | 9391272f736f10412c185abcde7048e1e57fbea805b5d19a55674b0ef014bf1b |
| eldric-aios-5.0.177-1.fc43.aarch64.rpm eldric-aios for aarch64, e.g. Raspberry Pi 4/5 (Fedora 43); install.sh picks the right one | 21.1 MB | dba7913080ed4d0aadbd6ebcf009e525bf7e2f0fc2c32f329345f4ce2b4020eb |
| eldric-aios-5.0.177-1.fc44.aarch64.rpm eldric-aios for aarch64, e.g. Raspberry Pi 4/5 (Fedora 44); install.sh picks the right one | 20.8 MB | c0670cc23a458796d1dc502c9a07ea09a8341b30330d0962a0cebb433671b649 |
Install with dnf (recommended)
| install-eldric.sh adds the repository, installs and starts eldric-aios: curl -fsSL https://repo.eldric.ai/install-eldric.sh | sudo bash | 2.3 KB | 111e8674f0bba609655ea27846ded8d63a846b9dee984934f6d24e26752a4935 |
| setup-dnf-repo.sh adds the signed dnf repository only | 10.1 KB | f57b10cc2ffe485c8baffddf3e81251899d0ef6e2c996c44363b5eaeba49dbd6 |
| RPM-GPG-KEY-eldric the package signing key | 1.7 KB | 2d06b576fdd06ec059073e3c5436b0b86cced8cce78310ff99b8525472501bb6 |
macOS
| Eldric-5.0.167-macos.pkg desktop app, signed and notarised | 30.8 MB | dcbc2f3ba72a923f8e0037f29eea7e2be8f489677da3376e1966f038e6472b65 |
Models
| bge-m3.gguf BAAI/bge-m3, MIT licence; used for tool preselection and search | 1.1 GB | daec91ffb5dd0c27411bd71f29932917c49cf529a641d0168496c3a501e3062c |
| bge-m3-q8_0.gguf BAAI/bge-m3, MIT licence (same licence file); the name Eldric's default embedder looks for | 605.2 MB | aa473d51f451a22f0fcf39ba3330c14bed38a385712b1113440f69df4047a173 |
| bge-m3-LICENSE.txt licence text for bge-m3.gguf | 1.2 KB | 3500e24cb1ff4ada083f10aa05f8d9e6d7cacf402074e1f2652d21767aee0a74 |
| nomic-embed-text-v1.5.Q4_K_M.gguf smaller embedding model | 80.2 MB | d4e388894e09cf3816e8b0896d81d265b55e7a9fff9ab03fe8bf4ef5e11295ac |
| eldric-nexus-f1-model.ckpt time-series forecasting checkpoint | 314.8 MB | 33a34a333e987d3050cd854117ad763ee33da089fd9d8ed33850e4d5e68f9e05 |
| eldric-nexus-f1-dyst-model.ckpt variant for dynamical systems | 314.9 MB | 034d052e4b457cc6775eec04e8ecb325fbebfa7502116e85b13ea0b75ee3b5f5 |
| eldric_nexus_f1-1.0.0-py3-none-any.whl pip install for testing the f1 forecaster | 70.2 KB | 372bbf88e329831c6d353f65b3f82dd193d120fc6c9c2f60325ad189fa06e75d |
| eldric_nexus_f1_dyst-1.0.0-py3-none-any.whl pip install for testing the f1-dyst forecaster | 70.7 KB | ba3f756399fcfb74b85927ca4ab2a0f2549c94cab521effcdd3fa71231a1ffc3 |
| example_forecast.py minimal example for the f1 packages | 3.1 KB | bf419a9261689197eaf525607827c36557eb5a6c24a45b80f4a4127945ba08ba |
| eldric-nexus-f1-README.md usage and licence of the f1 models | 6.9 KB | 75400321a6e7975d905be7e232e71688db9868497c033429c25a8419631e7ed5 |
World-model demos (CPU)
| README.md what the demos are and how to run them (CPU, no GPU, no Python at runtime) | 4.1 KB | 72d4c967ce65bd3b8ea833cea40a6d9764bd7f68bc9060cc68ef2adeeb56dc06 |
| viewer.html browser viewer for any .nsl/.ensl world model | 32.6 KB | ba33a332dc199e557ec8b888db78f836afa2e8acc586520cbf64fcda6088994c |
| eikonal_crust_v1.ensl world-model operator (eikonal / wavefront); grids, manifests and parity vectors are in /models/worldmodel/ | 44.2 KB | 0d198c4cfb6f093fcdfea05ff79f0286f8bdc84a99f9ace0e13aff2899dbfedd |
| eikonal_phi_crust_v1.ensl world-model operator (eikonal / wavefront); grids, manifests and parity vectors are in /models/worldmodel/ | 74.8 KB | 54ef438505ab41b0a8dbbd4ffa5a02902b0a77b3489f34d330741d9fb9493e51 |
| eikonal_production_v1.ensl world-model operator (eikonal / wavefront); grids, manifests and parity vectors are in /models/worldmodel/ | 40.7 KB | c7921bc7a2928638da710de0d116c598c62f08d0228c85faaa72094a868c1579 |
| eikonal_wavefront_v1.ensl world-model operator (eikonal / wavefront); grids, manifests and parity vectors are in /models/worldmodel/ | 40.7 KB | d5240329cde97386e8b08c7f1d389b0dca00cfa7dbf5c76b86df73e54a4e5c46 |
| elastic3d_nslot.nsl world-model operator (seismic); grids, manifests and parity vectors are in /models/worldmodel/ | 141.7 KB | 8f70c5b95a40df54c2d7faa8da8b6a47aef24ce39ffef7f61cfb45b974029d01 |
| elastic3d_wavefront_nslot.nsl world-model operator (eikonal / wavefront); grids, manifests and parity vectors are in /models/worldmodel/ | 50.7 KB | 198d790af88b72a1ea93d6a7492ea59aebac95ca03252608574f7dbc2be80407 |
| elastic_seismic_nslot.nsl world-model operator (seismic); grids, manifests and parity vectors are in /models/worldmodel/ | 65.2 KB | 65279a157e580f314bd9558f01b5c75cb897da2e9fb878fadd982d3005b0141f |
| elastic_wavefront_nslot.nsl world-model operator (eikonal / wavefront); grids, manifests and parity vectors are in /models/worldmodel/ | 17.1 KB | 36c76e825bb630153ac77db56ae5e411df384311f69ee62bcc1e8eee2f4fbecb |
| forecast_mackeyglass.nsl world-model operator (dynamics); grids, manifests and parity vectors are in /models/worldmodel/ | 12.8 KB | ff569f2a8822df4152b6d60a017165e4d3a612de352bfbc2c259d3446a36bee9 |
| fullcrust_wavefront_nslot.nsl world-model operator (eikonal / wavefront); grids, manifests and parity vectors are in /models/worldmodel/ | 17.1 KB | 0d0bbd07d778a19020e694baae7908b6892bbdd81f00fc1e8e6f770b252061f6 |
| marmousi_elastic_nslot.nsl world-model operator (seismic); grids, manifests and parity vectors are in /models/worldmodel/ | 65.2 KB | 894db91ac6c9324aa2b2086551b14bec21b170617eb4d47522908c4378743882 |
| marmousi_wavefront_nslot.nsl world-model operator (eikonal / wavefront); grids, manifests and parity vectors are in /models/worldmodel/ | 17.1 KB | 7d169ef27b632571882e674e409ea6b1f538f888cca03405d5057605c330099a |
| nbody_gravity_nslot.nsl world-model operator (dynamics); grids, manifests and parity vectors are in /models/worldmodel/ | 1.0 KB | 2f972e046225c48cca7a8e1455bbbf213d129aa7942a8d9ea3cfcd97bdb20049 |
| policy_mz07_nslot.nsl world-model operator (dynamics); grids, manifests and parity vectors are in /models/worldmodel/ | 46.0 KB | c4b9fdf067b0f8d0b13cb9a057cb16ab1a8a88fd2b1ed820b728add3bbfa3f00 |
| scalar_wavefront_nslot.nsl world-model operator (eikonal / wavefront); grids, manifests and parity vectors are in /models/worldmodel/ | 17.1 KB | c27322c25ac2ffe98c2a551d03b7cd2f56d5350847c1bf83c37e78762e1bec01 |
| scalar_wavefront_stable_v1.ensl world-model operator (eikonal / wavefront); grids, manifests and parity vectors are in /models/worldmodel/ | 21.5 KB | 224bb1f0188041f10a3b48b6b7aa73cabb51c8279f60a2ffada16dd66ed78c48 |
| seismic_marmousi_stable.nsl world-model operator (seismic); grids, manifests and parity vectors are in /models/worldmodel/ | 23.9 KB | 79028dd80d7440e83d1a190a1c6b89042c5efe1d260428d4d4373d7f63c753cf |
| seismic_student_v1.ensl world-model operator (seismic); grids, manifests and parity vectors are in /models/worldmodel/ | 23.9 KB | 39ef93cb58e12f2c64d03e5e1c1ded394f1b2ddff6654c7a9e13b88c1d71ff7e |
| seismic_wave_nslot.nsl world-model operator (seismic); grids, manifests and parity vectors are in /models/worldmodel/ | 26.2 KB | cb72226b0fe8158eca62ff4ba681ab5b54f9b151eaad465d48351f16122ad64b |
| threebody_nslot.nsl world-model operator (dynamics); grids, manifests and parity vectors are in /models/worldmodel/ | 19.7 KB | 9e16c3014c706c5c8c80fa0372948f1e19ca1e1231cb9f4b8f5940058853bc56 |
Testing for your organisation?
We help you evaluate Eldric on your own data and hardware.