Code

Headless & scalable Git forge. Its source of truth is a write-ahead log in object storage, not a disk.

Warning

Code is alpha software and is not proven in production. It is under active development, and the log format, HTTP and MCP interfaces, and configuration may change in breaking ways between releases, possibly without a migration path for data already in your object store.

Use it at your own responsibility. Keep independent backups of anything you care about, and treat a Code deployment as the only copy of nothing. As the license states, it comes with no warranty of any kind.

Code speaks plain Git smart HTTP, so git clone, git push and every tool built on them work unchanged. The difference is where a repository lives: the log in S3 is the repository, and a node's disk is only a warm cache.

🚀 Deploy one node

Each button creates a single-node Code service and asks for object-store credentials during setup.

CODE_AUTH_TOKENS takes token=account:read,write entries separated by semicolons. Vercel is fine for a quick evaluation; use one of the others, or Kubernetes, for long-lived Git traffic. The hosting guide covers object-store requirements and production options.

                 ┌──────────────────────────────────────┐
                 │        object storage (S3)           │
                 │                                      │
                 │  repos/<id>/index.pb    ← CAS'd      │
                 │  repos/<id>/wal/*.pb    ← immutable  │
                 │  repos/<id>/packs/*     ← immutable  │
                 └───────────▲──────────────▲───────────┘
                             │              │
        conditional GET      │              │  compare-and-swap
        (304 = serve now)    │              │  (decides push order)
                             │              │
        ┌────────────────────┴──┐   ┌───────┴───────────────┐
        │  node A               │   │  node B               │
        │  warm cache on NVMe   │   │  warm cache on NVMe   │
        │  git / MCP / admin    │   │  git / MCP / admin    │
        └───────────────────────┘   └───────────────────────┘
                     └──── distributed Erlang ────┘
                        membership + replication hints

💡 Why this shape

• Replicas are disposable. Everything a node holds can be rebuilt from the log. A crashed node, an evicted repository and a new pod all take the same path: materialize from the log.
• Placement is computed, not stored. Rendezvous hashing maps a repository and the live node set to a list of nodes. No routing table, no placement database.
• Any node can accept a push. One compare-and-swap on one object decides the order. No primary, no quorum, no consensus round.
• Reads are consistent without coordination. Before serving, a replica revalidates its view of the log with a conditional GET: 304 means serve now, 200 means catch up first.
• Replica count is a dial. A busy monorepo can name a hundred nodes; a repository an agent created a minute ago can name one.
• There is no control plane. Every node can answer every request, so Code runs as a plain Kubernetes Deployment behind a round-robin Service.

✨ Features

• Git smart HTTP: clone, fetch, push, protocol v2, shallow and partial clone.
• MCP server: read files, search, browse history and commit without cloning. See docs/mcp.md.
• OAuth 2.1 resource server: validates tokens, never issues them. Pods can authenticate with their projected service account token, and policy lives in object storage. See docs/kubernetes.md and docs/multi-tenancy.md.
• Recovery: restore a verified, retained state into a new repository. See docs/operations.md.
• Observability: Prometheus metrics and OpenTelemetry traces, including the signals worth autoscaling on.

🛠️ Quick start

mise install       # Erlang, Elixir, Rust, protoc and the rest of the toolchain
mise run setup     # deps and compile
mise run server    # a single node against a local filesystem object store

Then, in another shell:

curl -X POST localhost:4002/repositories \
  -H 'authorization: Bearer dev-admin-token' \
  -H 'content-type: application/json' \
  -d '{"repository":"acme/app"}'

git clone http://x-access-token:dev-token@localhost:4000/acme/app.git

To run the whole system in containers (RustFS for S3 plus two clustered nodes), use docker compose up --build -d.

📚 Documentation

Architecture	How the log works, and why there is no consensus protocol
Git verification	Compatibility coverage and a full-history Tuist migration rehearsal
Recovery	Restore exact retained states into a new repository and run a recovery drill
Operations	Configuration, metrics, failure modes, capacity
Kubernetes	Deploying, autoscaling, and authenticating pods
MCP	The agent-facing surface
Multi-tenancy	Authentication, authorization, and what isolation you actually get

🧑‍💻 Development

mise run test      # unit tests
mise run e2e       # end-to-end against RustFS and two nodes; needs Docker
mise run lint      # formatting and Credo
mise run typecheck # static type analysis
mise run proto     # regenerate the log schema after editing priv/proto

AGENTS.md covers the conventions the codebase depends on and the landmines that have already cost someone a day.

🌱 Prior art

The architecture follows the one Cursor described in Git at any scale, which in turn is a reaction to GitHub's Spokes. The substantial difference here is the runtime: on the BEAM, cluster membership, failure detection and reliable broadcast already exist, so Code uses distributed Erlang and :pg where the original design hand-rolls UDP gossip and a health table. What is left is the part that genuinely needs writing: the log, the compare-and-swap, and the convergence rule.

📄 License

Copyright 2026 Tuist GmbH. Mozilla Public License 2.0, see LICENSE.

MPL-2.0 is file-level copyleft: changes to Code's own source stay open, but it can be deployed alongside and linked from code under any license, including proprietary. Running a modified Code as a service does not oblige you to publish anything beyond the modified files themselves.