# `ml-cellar` with Custom Transfer Agent
> [!WARNING]
> The feature of `ml-cellar` with Custom Transfer Agent is not implemented for now.
> I’d appreciate it if you could wait for the implementation.
## Purpose
### Limitation in GitHub Git LFS
According to [the GitHub documentation](https://docs.github.com/en/billing/concepts/product-billing/git-lfs
), Git LFS on GitHub has a storage quota limit of 10 GB for Free and Pro accounts, as well as bandwidth limits.
It also has limitations such as not being able to choose the server region.
Using a Custom Transfer Agent, you can operate these on AWS S3 and are free from these limitations.
(You can address the storage limitation by paying GitHub for additional capacity.)
### Compare fee between GitHub Git LFS and AWS S3
(Information on 2025/12/01)
GitHub includes monthly download bandwidth and storage for Git LFS, depending on your plan:
- Free / Pro / Free for organizations: 10 GiB/month bandwidth + 10 GiB storage
- Team / Enterprise Cloud: 250 GiB/month bandwidth + 250 GiB storage
Notes:
- Bandwidth resets monthly, but storage does not reset (it accumulates over time).
- Downloads by GitHub Actions and collaborators also count against the repo owner’s LFS bandwidth.
If you exceed the included quota and have billing enabled:
- Storage: $0.07 per GiB-month
- Download bandwidth: $0.0875 per GiB
Maximum LFS object size depends on the plan in GitHub-hosted LFS:
- Free / Pro: 2 GB
- Team: 4 GB
- Enterprise Cloud: 5 GB
By using custom transfer backend, you can replace to Amazon S3.
- Storage: about $0.023 per GB-month (first 50 TB tier)
- Data transfer: first 100 GB/month free, then roughly $0.09 per GB
### Example use case
Assumptions:
- Stored: 200 GiB
- Downloaded per month: 1 TiB (≈ 1024 GiB)
- 1. GitHub Free + AWS S3
- GitHub: $0
- S3 storage: ~200 GB × $0.023 ≈ $4.60 / month (region may vary)
- S3 egress: (1024 − 100) GB × $0.09 = $83.16 / month
- Total ≈ $87.76 / month
- 2. GitHub Enterprise Cloud + GitHub-hosted LFS
- GitHub Enterprise Cloud: $21 × N users / month
- Overage bandwidth: (1024 − 250) GiB × $0.0875 = $67.73 / month
- Total ≈ ($21 × N) + $67.73 / month
## Get started
### Install
- Install dependency
```sh
sudo apt install git-lfs awscli
pip install aws2-wrap
cargo install lfs-dal
```
### Setup
- Edit `.mlcellar.toml` and set `custom_transfer_agent = true`.
```toml
[ml_cellar]
use_custom_transfer_agent = true
[aws]
profile = "{your_profile}"
```
- Make `.lfsdalconfig` if you use AWS S3.
```txt
[lfs-dal]
scheme = s3
bucket = your_bucket
region = your_region
```
- Run setup command
```sh
ml-cellar setup
```