Git Reimagined —
The AI-First Git Platform.
j0bGit is Git reimagined for autonomous AI coding agents (Claude Code, Cursor, Cline, OpenCode) and cloud-native engineering: S3 Write-Ahead Log durability, the live Lattice blast-radius & dependency graph, checkout-free atomic commits, and a native 85-tool MCP server.
git clone https://git.j0b.net/<repo>.git · MCP discovery: https://git.j0b.net/.well-known/mcp.json
Every repo, every dependency, every pull request —
one live SDLC graph.
Lattice is j0bGit's live relationship graph: files, Tree-sitter AST symbols, call graphs, cross-repo dependencies, pull requests, issues, CI runs, and SARIF security findings — connected the moment you push. Ask "what breaks if I change this?" and get an answer in milliseconds, with conflicting pull requests already flagged.
Cross-repo dependency maps
Manifests (Cargo.toml, package.json, pyproject.toml, go.mod, requirements.txt) become live edges. See your entire organization's dependency diamond scoped to your groups.
Blast radius & merge gating
Transactional impact analysis that runs at merge time, not in a nightly batch — risk level, impacted files and symbols, and which open PRs collide.
SARIF security & code subgraphs
Drop into any file to explore its symbols and calls, or traverse from a SARIF security finding directly to the open PRs and files it impacts — in the UI or over OGQL & MCP.
Stop starving AI coding agents
with flat text and token bloat.
On legacy Git hosts, agents burn millions of LLM tokens reading whole files and scraping CLI output. j0bGit speaks Model Context Protocol (MCP) natively — giving Claude Code, Cursor, Cline, and OpenCode surgical AST context and checkout-free write tools.
The autonomous AI work loop, built in
Point any MCP-compatible agent at /tree/mcp/sse (auto-discovered via /.well-known/mcp.json) with one bearer token:
- AST symbol slices & Tantivy search: fetch exact functions, structs, and references via
get_symbol_sliceinstead of reading 2,000-line files - Pre-edit blast radius & patch validation: run
get_blast_radiusandvalidate_patchbefore touching a line - Checkout-free atomic commits: commit multi-file changes directly to S3 via
atomic_commitwithout a local git clone - Agent inboxes & structured verdicts: claim issues, open PRs, trigger CI, read structured lint/test/security reports, and post review verdicts
{
"mcpServers": {
"jobgit": {
"url": "https://git.j0b.net/tree/mcp/sse",
"headers": {
"Authorization": "Bearer $JOBGIT_TOKEN"
}
}
}
}
# Agent runs the full SDLC loop over MCP:
✓ get_symbol_slice(repo="core", symbol="validate_token")
✓ get_blast_radius(repo="core", files=["src/auth.rs"])
✓ atomic_commit(repo="core", branch="fix/auth", files=[...])
✓ trigger_workflow_run(repo="core", workflow="ci")
✓ get_quality_report(pr_id="42")
Every primitive re-engineered for
AI agents and cloud-stateless scale.
Built from scratch in asynchronous Rust on Amazon S3 and DynamoDB — eliminating stateful disks, YAML indentation traps, and runaway CI bills.
AI-First with 85 MCP Tools
Every platform primitive — semantic code search, AST symbols, Lattice blast radius, PRs, code reviews, CI quality reports, issues, and audit logs — exposed over one authenticated MCP surface.
S3-First WAL Durability
Amazon S3 is the sole source of truth via a Write-Ahead Log (WAL), DynamoDB distributed push locks, and compare-and-swap manifests. Local checkouts are disposable NVMe caches.
Run CI Without Committing
Workflows are first-class database entities. Auto-discover .gitj0b/workflows/*.json on push, or register and run workflows via API/MCP against any ref with zero "fix CI" commits.
Durable Branch & AI Code Reviews
Review any protected branch, single commit, or path set — decoupled from pull requests. Combine human reviewers and Amazon Bedrock AI reviews with severity-tagged findings and verdicts.
PR Quality Gates & 2-Tier Namespaces
Pre-merge lint, test, coverage, and security reports attach to PRs as queryable data. Bounded 2-tier group namespaces keep paths O(1) and repo moves metadata-only.
FinOps Spend Caps & Zero-Trust Security
Hard organization spend caps prevent runaway AI agent loops from causing surprise cloud bills. Protected by push-rule secret scanning, Argon2id, SHA-256 tokens, and full audit trails.
Zero friction, from
git clone to multi-backend CI.
Push over HTTPS or SSH (ports 22 & 443), and fan out across serverless AWS Fargate, EC2, or your own Kubernetes & GPU runners.
Push like you always have
Token over HTTPS or your SSH key — both work everywhere, including CI pipelines and autonomous AI agents.
- Scoped API access tokens and Ed25519/RSA SSH keys
- SSH over port 22 and 443 so corporate firewalls never block pushes
- Built-in pre-receive secret scanning and 2-tier group RBAC
# add a remote and push to S3-backed storage git remote add origin https://git.j0b.net/aurora.git git push -u origin main remote: ✓ WAL pack persisted to S3 remote: ✓ workflow "Test, Lint & Security" queued on Fargate To https://git.j0b.net/aurora.git abc1234..def5678 main -> main
{
"name": "Test, Lint & Security",
"on": ["push", "workflow_dispatch"],
"jobs": {
"test": {
"runs-on": "fargate",
"steps": [
{ "name": "Checkout", "action": "checkout" },
{ "name": "Test", "run": "pytest --cov" }
]
}
}
}
Many workflow types on the same codebase
Drop schema-validated JSON workflows in .gitj0b/workflows/ or register workflows dynamically in the database via API and MCP.
- Mix compute per job: ephemeral AWS Fargate, EC2, rootless
crane, or self-hostedj0b-runner(with Kubernetes Helm chart) - Structured lint, test, coverage, and security reports surfaced on PRs
- Self-hosted runners isolate jobs per group with ephemeral read-only repo grants
j0bGit vs GitHub vs GitLab —
architectural differences, not marketing fog.
Why AI-first engineering teams are moving off legacy disk-bound monoliths.
| Architectural Axis | GitHub / GitLab (Gitaly & Orbit) | j0bGit (Git Reimagined) |
|---|---|---|
| AI Agent Operability | ✗REST API afterthought; agents burn tokens reading raw files | ✓Native 85-tool MCP server: AST symbol slices, semantic diffs, blast radius, atomic commits |
| Git Storage Model | ✗Stateful disks & replication (Gitaly / Spokes limits) | ✓Amazon S3 source of truth via Write-Ahead Log (WAL) + CAS manifests; local repos are disposable caches |
| Code & SDLC Graph | ✗Batch CDC indexes lagging minutes to hours; non-transactional | ✓Live Lattice graph fed by the write path; gates merges with real-time blast radius & PR collision checks |
| Code Reviews | ✗Locked to pull requests; cannot review arbitrary branches or paths | ✓Standalone durable reviews on any branch, commit, or path set (human + Amazon Bedrock AI) |
| Workflow Definition | ✗Repo YAML file only; requires committing config to test CI | ✓First-class DB entity + JSON repo files; run workflows against any commit without committing config |
| CI Compute & Runners | ✗Fixed runner pools or long-lived VMs to patch and babysit | ✓Per-job ephemeral AWS Fargate, EC2, crane, or self-hosted GPU / Kubernetes runners mixed in one workflow |
| Pre-Merge Quality Signal | ✗Pass/fail status ticks; findings buried in console log soup | ✓Structured lint, test, coverage, and SARIF security reports attached to the PR and queryable via MCP |
| Namespaces & FinOps | ✗Flat orgs or slow recursive subgroups; post-mortem billing | ✓Bounded 2-tier O(1) namespaces with instant repo moves; real-time FinOps metering & hard spend caps |
Why teams choose Git Reimagined.
Answers to common questions about j0bGit, AI-first Git workflows, Lattice, and S3-native durability.
What does "Git Reimagined" mean?
Traditional Git platforms tie repositories to stateful disks and treat AI agents as second-class REST scrapers. j0bGit (Git Reimagined) separates Git compute from state using an S3 Write-Ahead Log, indexes every AST symbol and relationship into the live Lattice graph, and exposes the entire SDLC over a native MCP server.
How is j0bGit an AI-first Git platform?
AI agents (Claude Code, Cursor, Cline, OpenCode) connect to git.j0b.net via the Model Context Protocol (MCP) to search code with Tantivy, fetch Tree-sitter AST slices, compute blast radius before editing, commit atomically without a local clone, open PRs, trigger CI, and read structured quality reports.
How does S3 storage compare to Gitaly or disk-backed Git?
Every push writes packfiles to an S3 Write-Ahead Log with DynamoDB distributed locks and compare-and-swap manifests before acknowledging the client. You get 99.999999999% (11 nines) S3 durability, stateless horizontal scaling across Rust microservices, and instant metadata-only repository moves.
How do FinOps spend caps protect against runaway AI agents?
Autonomous agents can trigger hundreds of CI runs an hour. j0bGit meters every ephemeral Fargate job and S3 storage sample in real time and enforces configurable organization spend caps (hard-stop or alert) so an agent loop never produces a surprise bill.
Serious Rust & AWS plumbing,
quietly doing its job.
j0bGit is built and operated by j0b LLC as a cloud-native Rust microservice platform — here's what's holding it up.
Guides that get you moving.
Focused platform guides written for developers and AI engineers who want to ship code immediately.
Getting Started with j0bGit →
Create an account, configure tokens or SSH keys, create your first S3-backed repository, push code, and connect your AI agent over MCP.
Actions & Multi-Workflow CI Guide →
JSON workflow schema (.gitj0b/workflows/*.json), push and MCP triggers, live run logs, structured PR quality reports, and self-hosted runners.
Ship your next repo on j0bGit today.
Experience Git Reimagined — create an account, push a repo, and watch your workflows and Lattice graph come alive.
Create your account →