# cgn-core
[](https://crates.io/crates/cgn-core)
[](https://docs.rs/cgn-core)
[](https://github.com/antonellof/cognitora-inference/blob/main/LICENSE)
Shared library used by every Cognitora binary: configuration loading, the
workspace `Error` enum, BLAKE3 prefix hashing, and the concurrent prefix
index. Telemetry, TLS, and Kubernetes helpers live in dedicated `cgn-*`
crates so this one stays free of heavy transitive dependencies.
Cognitora is the open-source datacenter-scale LLM inference stack.
`cgn-core` is its smallest internal crate: cross-cutting types and a few
constants that the routing, agent, kvcached, metrics, ctl, and operator
binaries all depend on.
## Use
```toml
[dependencies]
cgn-core = "0.1"
```
```rust
use cgn_core::{config::Config, Error, Result, hash::prefix_chain};
fn load() -> Result<Config> {
Config::load(cgn_core::DEFAULT_CONFIG_PATH)
}
```
## Modules
| `config` | Layered TOML loading with env-var overrides (built on `config-rs`). |
| `error` | The workspace-wide `Error` / `Result` and process exit-code mapping. |
| `hash` | Sequence-chained BLAKE3 prefix digests used by KV-aware routing. |
| `prefix` | Lock-free radix-trie index for the digests above. |
| `etcd_keys`| Canonical etcd key prefixes for nodes, models, routing policy, … |
## License
Apache-2.0. See [LICENSE](https://github.com/antonellof/cognitora-inference/blob/main/LICENSE).
Part of [Cognitora](https://github.com/antonellof/cognitora-inference).