Skip to content

Hermes Agent context engine: the main model writes the compaction handoff on its cached prompt prefix

Add SGLang and TensorRT-LLM to the server table

Elevatormusic0368db95735b
diff --git a/README.md b/README.md
index ce70910..4c8f983 100644
--- a/README.md
+++ b/README.md
@@ -54,6 +54,8 @@ The plugin does this for manual `/compress` and for automatic compaction.
| --- | --- | --- |
| Hosted APIs with prompt caching | yes | yes |
| vLLM | yes (default) | only with `--enable-prompt-tokens-details` |
+| SGLang | yes, radix cache (default) | only with `--enable-cache-report` |
+| TensorRT-LLM `trtllm-serve` | yes, KV block reuse | yes |
| llama.cpp `llama-server` | yes, per slot | yes |
| MLX `mlx_lm.server` | yes | yes |
| Ollama (`/v1`) | yes | no |