origin
agent 01agent 02agent N
Git evolved for the agent era
The fastest Git
on the planet.
Stargit is where humans and software agents build together—at machine speed, with trust and control built into every change.
Two keystrokes.One command to clone, delegate, verify, and ship.
stargit · acme/checkout
live
Demo chapter 1: Get the repository. The measured cold path—not a warm-cache trick.
❯sg
What do you want to do?
First contactGet the repository
complete250 files · 190.8 MiB materializedcontent identical
git
6.709ssg
3.341s2.01× fastermeasured cold clone
Cold clone · same content6.709sGit
Stargit native path3.341s2.01× faster
Measured first-time clone on the same approximately 200 MB corpus. No warm cache. Hardware and protocol details belong with every published result.
A network that gets smarter
Software should remember, not restart.
Today, every new agent starts from zero. Stargit gives the network a memory: bring verified code close once, reuse it everywhere, and move only what is genuinely new.
origin
verified generation
CoW 01CoW 02CoW N
How it works underneath · measured todayClone once. After that, ask only what changed.
OriginVerified generation
fetch oncefirst fetchcontent addressed
Machine memoryGolden cache
Measuredtip 9f2c…a17eworkspace 02~264 ms0 pack bytes
workspace 03~264 ms0 pack bytes
workspace Nlocal CoWsame generation
L1 · TodayShipped
Your machine remembers
Same remote tip means no repository payload is downloaded again. A small freshness check selects the verified local generation and creates the workspace with copy-on-write.
Clone #2–#8: ~264 ms · 0 pack bytesL2 · Planetary stepNext
Your company remembers
Put the content-addressed generation beside the fleet. One origin fetch feeds every runner and developer in the cell; local machines receive pages over the nearest path.
Target: one origin fetch per cellL3 · GitLakeDirection
The network knows what you have
When the remote moves, reconcile what the client already holds and transfer only missing immutable objects or extents—not another repository-shaped archive.
Target: bytes proportional to novelty6.8×measured steady-state time advantage
0pack bytes after the first verified generation
Measured on one machine with a 51.7 MiB, 7,115-file real-source repository across eight sequential workspaces. Cache hits still perform a tiny remote-tip freshness check. L2 and L3 are explicitly roadmap targets.
A new foundation for making software
Git tracked code. Stargit coordinates the work.
Software is no longer made by humans alone. Stargit gives people and agents one shared system to build faster, remember what happened, and decide what is safe to release.
01
Move at machine speed
A first-time clone completed 2.01× faster than Git on the same measured corpus.
02
Let agents build safely
Every agent gets an isolated place to work while your source and policies stay under control.
03
Know what is safe to ship
Every change arrives with its tests, history, evidence, and approval—not just a promise.
From intent to trusted software
Ask. Build. Prove. Ship.
Ask
A person describes the outcome.
Agents build
Work happens safely in parallel.
Stargit proves
Every result carries its evidence.
You ship
Only trusted work reaches production.
Open adoption. Compounding enterprise value.
The command is free. Trust is the product.
Developers adopt a fast, open sg workflow. Companies pay for the shared network that governs every agent, preserves proof, protects private code, and controls what ships.
The Stargit NetworkPolicy enforced
One place to control every agentConsistent rules across tools, models, teams, and runners.
Proof that never disappearsEvery important change keeps the story of how it was made.
Runs where your code livesPrivate infrastructure, dedicated network cells, and enterprise support.
Start with sg.
Then watch every agent change become visible, reviewable, and safe to merge.