diff --git a/lib/muninn/index_writer.ex b/lib/muninn/index_writer.ex
index 2e8dd6e..b6ea019 100644
--- a/lib/muninn/index_writer.ex
+++ b/lib/muninn/index_writer.ex
@@ -158,4 +158,46 @@ defmodule Muninn.IndexWriter do
       error -> error
     end
   end
+
+  @doc """
+  Deletes all documents matching a (field, value) term.
+
+  Tantivy marks the matching documents as deleted; the delete becomes
+  visible to searches after the next `commit/1`. Useful for incremental
+  refresh: delete docs whose key matches `value`, then re-add the new
+  versions in the same transaction.
+
+  Supports text, u64, i64, and bool fields. f64 is not supported (Tantivy
+  has no stable term encoding for floats).
+
+  ## Parameters
+
+    * `index` - The index to delete from
+    * `field_name` - Name of the indexed field to match against
+    * `value` - The value to match (string / integer / boolean)
+
+  ## Returns
+
+    * `:ok` - Delete queued (visible after `commit/1`)
+    * `{:error, reason}` - Failed (e.g. field not found, unsupported type)
+
+  ## Examples
+
+      :ok = Muninn.IndexWriter.add_document(index, %{"id" => "doc-1", "body" => "hello"})
+      :ok = Muninn.IndexWriter.commit(index)
+
+      :ok = Muninn.IndexWriter.delete_term(index, "id", "doc-1")
+      :ok = Muninn.IndexWriter.commit(index)
+      # "doc-1" no longer searchable
+
+  """
+  @spec delete_term(reference(), String.t(), String.t() | integer() | boolean()) ::
+          :ok | {:error, String.t()}
+  def delete_term(index, field_name, value) when is_binary(field_name) do
+    case Native.writer_delete_term(index, field_name, value) do
+      {:ok, _} -> :ok
+      :ok -> :ok
+      error -> error
+    end
+  end
 end
diff --git a/lib/muninn/native.ex b/lib/muninn/native.ex
index 56f61a6..e1ea682 100644
--- a/lib/muninn/native.ex
+++ b/lib/muninn/native.ex
@@ -81,6 +81,10 @@ defmodule Muninn.Native do
   @doc false
   def writer_rollback(_index), do: :erlang.nif_error(:nif_not_loaded)
 
+  @doc false
+  def writer_delete_term(_index, _field_name, _value),
+    do: :erlang.nif_error(:nif_not_loaded)
+
   ## Reader functions
 
   @doc false
diff --git a/native/muninn/src/lib.rs b/native/muninn/src/lib.rs
index f8f9f71..dd54f4f 100644
--- a/native/muninn/src/lib.rs
+++ b/native/muninn/src/lib.rs
@@ -64,6 +64,15 @@ fn writer_rollback(index: rustler::ResourceArc<index::IndexResource>) -> Result<
     writer::writer_rollback(index)
 }
 
+#[rustler::nif]
+fn writer_delete_term(
+    index: rustler::ResourceArc<index::IndexResource>,
+    field_name: String,
+    value: rustler::Term,
+) -> Result<(), String> {
+    writer::writer_delete_term(index, field_name, value)
+}
+
 #[rustler::nif]
 fn reader_new(
     index: rustler::ResourceArc<index::IndexResource>,
diff --git a/native/muninn/src/writer.rs b/native/muninn/src/writer.rs
index 8901438..0cdd42a 100644
--- a/native/muninn/src/writer.rs
+++ b/native/muninn/src/writer.rs
@@ -1,7 +1,7 @@
 use rustler::{Env, ResourceArc, Term};
 use std::collections::HashMap;
 use tantivy::schema::FieldType;
-use tantivy::TantivyDocument;
+use tantivy::{TantivyDocument, Term as TantivyTerm};
 
 use crate::index::IndexResource;
 
@@ -101,6 +101,87 @@ pub fn writer_add_document(
     Ok(())
 }
 
+/// Deletes all documents containing the given term (field+value pair).
+///
+/// Tantivy marks matching documents as deleted; the delete becomes visible
+/// to readers after the next `commit`. Supports text, u64, i64, and bool
+/// fields. f64 fields are not supported because Tantivy does not provide a
+/// stable term encoding for floats.
+pub fn writer_delete_term(
+    index_res: ResourceArc<IndexResource>,
+    field_name: String,
+    value: Term,
+) -> Result<(), String> {
+    let index = index_res
+        .index
+        .lock()
+        .map_err(|_| "Failed to acquire index lock".to_string())?;
+
+    let schema = index.schema();
+    let field = schema
+        .get_field(&field_name)
+        .map_err(|_| format!("Field '{}' not found in schema", field_name))?;
+    let field_entry = schema.get_field_entry(field);
+
+    let term = match field_entry.field_type() {
+        FieldType::Str(_) => {
+            let s = value
+                .decode::<String>()
+                .map_err(|_| "Expected string value for text field".to_string())?;
+            TantivyTerm::from_field_text(field, &s)
+        }
+        FieldType::U64(_) => {
+            let n = value
+                .decode::<u64>()
+                .or_else(|_| value.decode::<i64>().map(|i| i as u64))
+                .map_err(|_| "Expected integer value for u64 field".to_string())?;
+            TantivyTerm::from_field_u64(field, n)
+        }
+        FieldType::I64(_) => {
+            let n = value
+                .decode::<i64>()
+                .or_else(|_| value.decode::<u64>().map(|u| u as i64))
+                .map_err(|_| "Expected integer value for i64 field".to_string())?;
+            TantivyTerm::from_field_i64(field, n)
+        }
+        FieldType::Bool(_) => {
+            let b = value
+                .decode::<bool>()
+                .map_err(|_| "Expected boolean value for bool field".to_string())?;
+            TantivyTerm::from_field_bool(field, b)
+        }
+        _ => {
+            return Err(format!(
+                "Field '{}' has unsupported type for delete_term (text, u64, i64, bool only)",
+                field_name
+            ));
+        }
+    };
+
+    drop(index);
+
+    let mut writer_lock = index_res
+        .writer
+        .lock()
+        .map_err(|_| "Failed to acquire writer lock".to_string())?;
+
+    if writer_lock.is_none() {
+        let index = index_res
+            .index
+            .lock()
+            .map_err(|_| "Failed to acquire index lock".to_string())?;
+        let new_writer = index
+            .writer(50_000_000)
+            .map_err(|e| format!("Failed to create writer: {}", e))?;
+        *writer_lock = Some(new_writer);
+    }
+
+    let writer = writer_lock.as_mut().unwrap();
+    writer.delete_term(term);
+
+    Ok(())
+}
+
 /// Commits all pending changes to the index
 pub fn writer_commit(index_res: ResourceArc<IndexResource>) -> Result<(), String> {
     let mut writer_lock = index_res
diff --git a/test/muninn/index_writer_test.exs b/test/muninn/index_writer_test.exs
index 581c591..bc2c596 100644
--- a/test/muninn/index_writer_test.exs
+++ b/test/muninn/index_writer_test.exs
@@ -265,6 +265,151 @@ defmodule Muninn.IndexWriterTest do
     end
   end
 
+  describe "delete_term/3" do
+    alias Muninn.{IndexReader, Searcher}
+    alias Muninn.Query.Term, as: TermQuery
+
+    test "deletes documents matching a text term", %{test_path: test_path} do
+      schema =
+        Schema.new()
+        |> Schema.add_text_field("id", stored: true, indexed: true)
+        |> Schema.add_text_field("body", stored: true, indexed: true)
+
+      {:ok, index} = Index.create(test_path, schema)
+
+      :ok =
+        IndexWriter.add_documents(index, [
+          %{"id" => "doc1", "body" => "hello"},
+          %{"id" => "doc2", "body" => "world"}
+        ])
+
+      :ok = IndexWriter.commit(index)
+
+      {:ok, reader} = IndexReader.new(index)
+      {:ok, searcher} = Searcher.new(reader)
+
+      assert {:ok, %{"total_hits" => 1}} =
+               Searcher.search(searcher, %TermQuery{field: "id", value: "doc1"}, limit: 10)
+
+      :ok = IndexWriter.delete_term(index, "id", "doc1")
+      :ok = IndexWriter.commit(index)
+
+      {:ok, reader} = IndexReader.new(index)
+      {:ok, searcher} = Searcher.new(reader)
+
+      assert {:ok, %{"total_hits" => 0}} =
+               Searcher.search(searcher, %TermQuery{field: "id", value: "doc1"}, limit: 10)
+
+      assert {:ok, %{"total_hits" => 1}} =
+               Searcher.search(searcher, %TermQuery{field: "id", value: "doc2"}, limit: 10)
+    end
+
+    test "deletes all documents matching the term", %{test_path: test_path} do
+      schema =
+        Schema.new()
+        |> Schema.add_text_field("category", stored: true, indexed: true)
+        |> Schema.add_text_field("title", stored: true, indexed: true)
+
+      {:ok, index} = Index.create(test_path, schema)
+
+      :ok =
+        IndexWriter.add_documents(index, [
+          %{"category" => "draft", "title" => "A"},
+          %{"category" => "draft", "title" => "B"},
+          %{"category" => "published", "title" => "C"}
+        ])
+
+      :ok = IndexWriter.commit(index)
+      :ok = IndexWriter.delete_term(index, "category", "draft")
+      :ok = IndexWriter.commit(index)
+
+      {:ok, reader} = IndexReader.new(index)
+      {:ok, searcher} = Searcher.new(reader)
+
+      assert {:ok, %{"total_hits" => 0}} =
+               Searcher.search(
+                 searcher,
+                 %TermQuery{field: "category", value: "draft"},
+                 limit: 10
+               )
+
+      assert {:ok, %{"total_hits" => 1}} =
+               Searcher.search(
+                 searcher,
+                 %TermQuery{field: "category", value: "published"},
+                 limit: 10
+               )
+    end
+
+    test "delete then re-add same term (incremental refresh pattern)", %{test_path: test_path} do
+      schema =
+        Schema.new()
+        |> Schema.add_text_field("id", stored: true, indexed: true)
+        |> Schema.add_text_field("body", stored: true, indexed: true)
+
+      {:ok, index} = Index.create(test_path, schema)
+
+      :ok = IndexWriter.add_document(index, %{"id" => "doc1", "body" => "old content"})
+      :ok = IndexWriter.commit(index)
+
+      :ok = IndexWriter.delete_term(index, "id", "doc1")
+      :ok = IndexWriter.add_document(index, %{"id" => "doc1", "body" => "new content"})
+      :ok = IndexWriter.commit(index)
+
+      {:ok, reader} = IndexReader.new(index)
+      {:ok, searcher} = Searcher.new(reader)
+
+      assert {:ok, %{"total_hits" => 1, "hits" => [hit]}} =
+               Searcher.search(searcher, %TermQuery{field: "id", value: "doc1"}, limit: 10)
+
+      body = hit["doc"]["body"]
+      assert body == ["new content"] or body == "new content"
+    end
+
+    test "delete on uncommitted writer is queued and applied at commit",
+         %{test_path: test_path} do
+      schema =
+        Schema.new()
+        |> Schema.add_text_field("id", stored: true, indexed: true)
+
+      {:ok, index} = Index.create(test_path, schema)
+
+      :ok = IndexWriter.add_document(index, %{"id" => "ghost"})
+      :ok = IndexWriter.delete_term(index, "id", "ghost")
+      :ok = IndexWriter.commit(index)
+
+      {:ok, reader} = IndexReader.new(index)
+      {:ok, searcher} = Searcher.new(reader)
+
+      assert {:ok, %{"total_hits" => 0}} =
+               Searcher.search(searcher, %TermQuery{field: "id", value: "ghost"}, limit: 10)
+    end
+
+    test "delete by u64 term", %{test_path: test_path} do
+      schema =
+        Schema.new()
+        |> Schema.add_u64_field("id", stored: true, indexed: true)
+        |> Schema.add_text_field("body", stored: true)
+
+      {:ok, index} = Index.create(test_path, schema)
+
+      :ok = IndexWriter.add_document(index, %{"id" => 42, "body" => "answer"})
+      :ok = IndexWriter.add_document(index, %{"id" => 43, "body" => "next"})
+      :ok = IndexWriter.commit(index)
+
+      assert :ok = IndexWriter.delete_term(index, "id", 42)
+      :ok = IndexWriter.commit(index)
+    end
+
+    test "returns error for unknown field", %{test_path: test_path} do
+      schema = Schema.new() |> Schema.add_text_field("id", stored: true, indexed: true)
+      {:ok, index} = Index.create(test_path, schema)
+
+      assert {:error, reason} = IndexWriter.delete_term(index, "nope", "x")
+      assert reason =~ "not found"
+    end
+  end
+
   describe "real-world scenarios" do
     test "e-commerce product indexing", %{test_path: test_path} do
       schema =