diff --git a/README.md b/README.md
index 28dfd5b..2e0c25e 100644
--- a/README.md
+++ b/README.md
@@ -11,38 +11,16 @@ pip install semble
 ## Python API
 
 ```python
-from semble import SearchMode, SembleIndex
+from semble import SembleIndex
 
 index = SembleIndex.from_path("./my-project")
 
-# Hybrid search (semantic + BM25, default)
 results = index.search("how does authentication work?", top_k=5)
-for r in results:
-    print(r.chunk.location, f"score={r.score:.3f}")
-    print(r.chunk.content[:200])
-
-# Keyword-only
-results = index.search("JWT token", mode=SearchMode.BM25)
-```
-
-## Search modes
-
-| Mode | Description |
-|------|-------------|
-| `hybrid` | Semantic + BM25, normalized and combined (default) |
-| `semantic` | Embedding similarity only |
-| `bm25` | Keyword search only |
-
-## Disk embedding cache
-
-Embeddings are cached to `~/.cache/semble` by default so re-indexing unchanged files is instant. When using a custom encoder, pass `model_name` to enable caching:
-
-```python
-index = SembleIndex.from_path("./my-project", model=my_model, model_name="my-org/my-model")
+for result in results:
+    print(result.chunk.location, f"score={result.score:.3f}")
+    print(result.chunk.content[:200])
 ```
 
-Only embeddings are cached; BM25 and the ANNS index are always rebuilt fresh.
-
 ## MCP server
 
 Semble can run as an MCP server so agents (Claude Code, Cursor, etc.) can search your codebase directly.
@@ -63,5 +41,5 @@ This indexes the directory at startup and exposes two tools:
 
 | Tool | Description |
 |------|-------------|
-| `search` | Search with a natural-language or code query. Supports `hybrid` (default), `semantic`, and `bm25` modes. |
+| `search` | Search your codebase with a natural-language or code query. |
 | `find_related` | Given a file path and line number, return chunks semantically similar to the code at that location. |