# atomr-accel-agents
Agentic / LLM GPU actor blueprints on
[atomr-accel-cuda](../atomr-accel-cuda): RAG, embedding cache, vector index,
shared-state coordination, and a LangGraph-style DAG executor.
## Add to your project
```toml
[dependencies]
atomr-accel = "0.0"
atomr-accel-agents = "0.0"
```
```rust
use atomr_accel_agents::prelude::*;
```
Depends only on `atomr-accel-cuda` (for `ManagedRef`, errors). No
patterns / train / realtime coupling.
## What's inside
| Retrieval-augmented generation | `RagPipeline` |
| LRU embedding cache | `EmbeddingCache` |
| CPU vector index (top-k cosine) | `CpuVectorIndex` |
| Shared-state write tokens | `SharedGpuStateCoordinator` |
| LangGraph DAG executor | `LangGraphGpuActor<S>` |
License: Apache-2.0.