One 47 MB binary. Raft consensus. AI ops. GPU scheduling. WebAssembly. Self-healing. Auto-TLS. Multi-node clusters up to 20 nodes. Zero external dependencies.
cargo install mallorca && sudo setcap 'cap_net_bind_service=+ep' $(which orca) Agent, control plane, CLI, and proxy in one 47 MB executable. scp it to any Linux server and run.
Raft consensus via openraft. Embedded redb storage. No etcd. orca join and you're in the cluster.
orca ask "why is the API slow?" โ diagnose issues in plain English. Works with any OpenAI-compatible LLM.
ACME/Let's Encrypt built in. Certificates provisioned and renewed automatically for every domain.
Run Wasm modules via wasmtime alongside containers. Cold start in ~5 ms, ~2 MB per instance, hundreds on one node.
Declare NVIDIA or AMD GPU requirements. The scheduler filters nodes by VRAM availability automatically.
Watchdog restarts crashed containers in ~30s. Agents reconnect automatically after network partitions.
Every service fits in ~10 lines of TOML. Put it in git. Push to deploy via the infra webhook.