[Blog](/blog/.md) # objgit is not a git forge objgit is not a git forge. It's a git server on Tigris object storage that you can script with hooks, explore over SSH, and grow with. Here's what's next. Author [![Xe Iaso](https://avatars.githubusercontent.com/u/529003?v=4)](https://xeiaso.net) [Xe Iaso](https://xeiaso.net) Senior Cloud Whisperer Published2026-10-07 Reading8 min Tags [Engineering](/blog/tags/engineering/.md)[Open Source](/blog/tags/open-source/.md)[git](/blog/tags/git/.md)+1 Contents * [The current state of the git](#the-current-state-of-the-git) * [The state of the objgit](#the-state-of-the-objgit) * [Scripting your Git server](#scripting-your-git-server) * [Events with webhooks](#events-with-webhooks) * [What’s coming up](#whats-coming-up) ![The objgit logo with the tagline 'a git server you can grow with', next to glowing git commit lines streaming cubes into a translucent bucket](/blog/assets/images/hero-image-0a6d2878f0d640d1ea499998ad503911.webp) Hey all, I’ve been working more on [objgit](https://github.com/tigrisdata/objgit) and I think it’s time to transition it from a rudderless experiment to something that we can actually rely on. Today I want to spell out the vision I have for objgit and what I think will make it or something built on top of it shine. TL;DR: objgit is not a git forge, it’s a git server you can grow with. Let me explain. ## The current state of the git[​](#the-current-state-of-the-git "Direct link to The current state of the git") Operating a Git server at scale is a nightmare. At a small scale you can just have a single VPS with Gitea, Forgejo, or [soft-serve](https://github.com/charmbracelet/soft-serve) and be fine. If you only have a few projects and want to keep it on your homelab, this will work and it will be good enough. The devil comes in with the details, and most of the problems begin when you expose that Git server to the public internet. The scrapers will discover your single person git forge and will gladly click on every link of every link of every link over and over until your server crashes and you need to install [a web application firewall](https://anubis.techaro.lol) or take other drastic measures to keep it online. Additionally, with most git forges your data is stored in a single folder on a single drive. You should be using RAID or some kind of distributed storage system to store your git data, however the process of actually doing that requires a lot of toil and maybe even a more than healthy dose of claudeops to just get to the point where the thing works. If this is your jam, this is pretty fun. Hell, I set up my own git server like that in my homelab until [Amazon’s scraper took it out](https://xeiaso.net/notes/2025/amazon-crawler/). ## The state of the objgit[​](#the-state-of-the-objgit "Direct link to The state of the objgit") What if things were set up to be robust by default? What if your git server just focused on being that: a git server? In the process you’d probably end up with objgit. In objgit, everything is stored in Tigris by default. Repository contents are temporarily cached on the disk for speed, but the default state is that things are in Tigris. When you push large objects, they automatically get added to LFS in Tigris. Every time you push you get a read-only snapshot of that tree saved to Tigris so that you can consume it later with other tools. Basically: we want you to have a Git *server*, and then you just deal with that using your git client. Other tools build on top of it. ### Scripting your Git server[​](#scripting-your-git-server "Direct link to Scripting your Git server") Historically the main extension point you have for git servers is CI. What if your main extension point was the git server itself? If you’re a certain kind of greybeard, this sounds incredibly obvious, however for a huge portion of developers they have never dealt with their own git server on an Ubuntu box. They don’t know you can “just” put a shell script in the repo because that detail is abstracted away on GitHub or with any git forge in a box project. Objgit lets you define post-receive hooks in your repo so that you can use the git server as the extension point like the good old days. Just put a shell script in `.objgit/hooks/receive-pack`: ``` #!/usr/bin/env bash # Variables visible in the shell script: # OBJGIT_REPO repository path, e.g. Xe/soteria # OBJGIT_SERVICE always "receive-pack" # OBJGIT_REF full ref name, e.g. refs/heads/main # OBJGIT_BRANCH short branch name, e.g. main # OBJGIT_OLD_SHA previous tip (all zeros when the branch was created) # OBJGIT_NEW_SHA new tip # OBJGIT_ADDED_FILES_JSON, OBJGIT_CHANGED_FILES_JSON, # OBJGIT_DELETED_FILES_JSON are JSON arrays of net file changes. # OBJGIT_CHANGES_FILE points to the full JSON object in /tmp. echo "push to ${OBJGIT_REPO} ${OBJGIT_REF}: ${OBJGIT_OLD_SHA} -> ${OBJGIT_NEW_SHA}" # Start Tekton tests only if you push to main if [ "${OBJGIT_BRANCH}" != "main" ]; then exit 0 fi kustomize build /src/.tekton | kube:apply && tekton:pipelinerun /src/.tekton/testrun.yaml ``` Et voila! You have CI with [Tekton](https://tekton.dev/). The `kustomize` command is the upstream `kustomize` compiled for WebAssembly and `kube:apply`/`tekton:pipelinerun` are custom pseudo-commands that do exactly what they say on the tin: apply Kubernetes manifests (for updating your pipeline definition) and create a new PipelineRun to make CI go. Additionally, why can’t you SSH into your git server to explore a repository yourself? ``` λ ssh -t git@github.com sh Xe/x master PTY allocation request failed on channel 0 ``` Imagine a world where you can just do that: ``` λ ssh -t git@objgit.objgit.svc.alrest.xeserv.us sh Xe/x master objgit hook shell: Xe/x refs/heads/master @ 0a1d4562cab28281b6d2c288d6d1184afd0bbf42 /src is read-only and /tmp is writable. Type exit or press Ctrl-D to leave. $ head -n3 go.mod module within.website/x go 1.25.7 $ uname -av wasi localhost 0.0.0 0.0.0 wasm32 WASI ``` This looks like it’d be catastrophically unsafe, but it’s all sandboxed with [userspace shells](https://www.tigrisdata.com/blog/agent-sandbox-go/) and the same coreutils implementation that Ubuntu uses. Again, everything is compiled to WebAssembly and runs in the scope of your repository. Sure right now the UX is kinda bad, it doesn’t support tab-completion, some commands are missing (such as `grep`), however right now it works just enough that it’s good enough to be the basis to make something better. ### Events with webhooks[​](#events-with-webhooks "Direct link to Events with webhooks") Webhooks are the message queue of the people. As such, objgit supports them using a format [similar to GitHub’s webhooks](https://github.com/tigrisdata/objgit/blob/main/proto/tigrisdata/objgit/events/push/v1/push.example.json). Right now the UX for configuring webhooks is kinda crummy and needs a lot of love, but here’s how you set them up today: ``` $ ssh -t git@objgit.objgit.svc.alrest.xeserv.us objgit-webhook-set Xe/x --url https://domain.tld/path/with/segments ``` Webhooks can be validated using [standard HMAC validation](https://github.com/tigrisdata/objgit/blob/main/docs/usage/webhooks.md#headers-and-verification). You will get details about every commit pushed so you can do whatever it is you want to do, such as integrating with your Jenkins install that speaks GitHub webhooks. ## What’s coming up[​](#whats-coming-up "Direct link to What’s coming up") Here’s what I’m cooking up with objgit to make it the best git server it can possibly be: * I plan to finish out POSIX compatibility with `grep`, `find`, and other core utilities. * I want to add System One / Jev API commands so that you can make sure that docs changes don’t cause a full CI run. * Maybe I should figure out an easy way to get [WASIX](https://www.wasix.org/) support so that WebAssembly binaries can make outgoing HTTP requests. * I plan to continue working on optimization so that you can push git repos of any size. Currently I can barely push the Linux kernel git repo. I plan to make this better so that the size of your git repo is a moot point. * I want to make it easy for you to add your own commands to objgit hook contexts. Maybe this could be done with some kind of meta-repo that has a `bin` folder full of WebAssembly binaries in LFS. * I want to use the foundation of objgit to build a scraper-resistant web UI for git. Most of the expensive calls (tree generation, archive generation, tree browsing, git blame, etc) can be cached but usually aren’t because it’s assumed that disk space isn’t infinite. Storing everything in Tigris makes storage effectively infinite. This would only serve cached information to users over the web and everything else will just give you an error page telling you what commands you need to run to get that information yourself. * I want to build a robust authentication and authorization system built on the lessons I’ve learned writing my own IAM server and implementing [SigV4 authentication](https://www.tigrisdata.com/blog/sigv4/) myself. Ideally, IAM should let you control how people are allowed to push, pull, and create repos, branches, and more. This would also hook into the API when that’s ready. * I want to make objgit’s primitives easy to embed into other projects so that you can swap out the storage backend with Tigris. * I want to make an API for controlling various aspects of the objgit service. * I want to build administrative dashboards so you can account for who is using the git server the most. * Maybe it would be cool to build a static website server on top of the primitives of objgit. This could be an interesting experiment! * I’d also like to explore agentic integrations so that your agent can create a git repository to help it work on a task. Above all though, I want to make objgit the git server of my dreams. I’m crossing a lot of things off of my git server design bucket list as I’ve learned how git servers fail in production helping keep core open source infrastructure online. I can only hope it will be useful for you too. Please give it a try and let me know what you think! Want a git server you can grow with? objgit stores every repository in Tigris, so your git data is globally distributed and durable by default. [Try objgit→](https://github.com/tigrisdata/objgit) ## Share [X](https://twitter.com/intent/tweet?text=Check%20out%20this%20post%20on%20the%20%40tigrisdata%20blog%3A%20objgit%20is%20not%20a%20git%20forge%0A%0A\&url=https%3A%2F%2Fwww.tigrisdata.com%2Fblog%2Fobjgit-vision)[Hacker News](https://news.ycombinator.com/submitlink?u=https%3A%2F%2Fwww.tigrisdata.com%2Fblog%2Fobjgit-vision\&t=objgit%20is%20not%20a%20git%20forge)[LinkedIn](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.tigrisdata.com%2Fblog%2Fobjgit-vision)[Email](mailto:?subject=objgit%20is%20not%20a%20git%20forge\&body=Hey!%0A%0ACheck%20out%20this%20article%20on%20the%20Tigris%20blog%3A%20https%3A%2F%2Fwww.tigrisdata.com%2Fblog%2Fobjgit-vision)Copy link