diff --git a/assets/settings/default.json b/assets/settings/default.json
index 059608ae04..e6842ae4fa 100644
--- a/assets/settings/default.json
+++ b/assets/settings/default.json
@@ -2427,7 +2427,7 @@
     "ollama": {
       "api_url": "http://localhost:11434",
     },
-    "llama_cpp": {
+    "llama.cpp": {
       "api_url": "http://localhost:8080",
     },
     "openai": {
diff --git a/crates/language_models/src/provider/llama_cpp.rs b/crates/language_models/src/provider/llama_cpp.rs
index 935ec74f9a..921a2ae63e 100644
--- a/crates/language_models/src/provider/llama_cpp.rs
+++ b/crates/language_models/src/provider/llama_cpp.rs
@@ -36,7 +36,7 @@ use crate::AllLanguageModelSettings;
 const LLAMA_CPP_DOWNLOAD_URL: &str = "https://llama.app";
 const LLAMA_CPP_MODELS_URL: &str = "https://huggingface.co/models?library=gguf&sort=trending";
 
-const PROVIDER_ID: LanguageModelProviderId = LanguageModelProviderId::new("llama_cpp");
+const PROVIDER_ID: LanguageModelProviderId = LanguageModelProviderId::new("llama.cpp");
 const PROVIDER_NAME: LanguageModelProviderName = LanguageModelProviderName::new("llama.cpp");
 
 const API_KEY_ENV_VAR_NAME: &str = "LLAMACPP_API_KEY";
@@ -448,7 +448,7 @@ fn display_name_for(id: &str) -> String {
 }
 
 fn telemetry_id_for(id: &str) -> String {
-    format!("llama_cpp/{}", display_name_for(id))
+    format!("{PROVIDER_ID}/{}", display_name_for(id))
 }
 
 impl LlamaCppLanguageModelProvider {
@@ -1656,13 +1656,13 @@ mod tests {
     fn telemetry_id_strips_local_model_paths() {
         assert_eq!(
             telemetry_id_for("/Users/alice/models/Qwen2.5-Coder-7B-Instruct-Q4_K_M.gguf"),
-            "llama_cpp/Qwen2.5-Coder-7B-Instruct-Q4_K_M"
+            "llama.cpp/Qwen2.5-Coder-7B-Instruct-Q4_K_M"
         );
         assert_eq!(
             telemetry_id_for(r"C:\Users\alice\models\Qwen2.5-Coder-7B-Instruct-Q4_K_M.gguf"),
-            "llama_cpp/Qwen2.5-Coder-7B-Instruct-Q4_K_M"
+            "llama.cpp/Qwen2.5-Coder-7B-Instruct-Q4_K_M"
         );
-        assert_eq!(telemetry_id_for("my-alias"), "llama_cpp/my-alias");
+        assert_eq!(telemetry_id_for("my-alias"), "llama.cpp/my-alias");
     }
 
     #[test]
diff --git a/crates/settings_content/src/language_model.rs b/crates/settings_content/src/language_model.rs
index ab11097728..e58f3dcc02 100644
--- a/crates/settings_content/src/language_model.rs
+++ b/crates/settings_content/src/language_model.rs
@@ -15,6 +15,7 @@ pub struct AllLanguageModelSettingsContent {
     pub bedrock: Option<AmazonBedrockSettingsContent>,
     pub deepseek: Option<DeepseekSettingsContent>,
     pub google: Option<GoogleSettingsContent>,
+    #[serde(rename = "llama.cpp")]
     pub llama_cpp: Option<LlamaCppSettingsContent>,
     pub lmstudio: Option<LmStudioSettingsContent>,
     pub mistral: Option<MistralSettingsContent>,
diff --git a/docs/src/ai/use-a-local-model.md b/docs/src/ai/use-a-local-model.md
index 89984d10f3..5ffc60172c 100644
--- a/docs/src/ai/use-a-local-model.md
+++ b/docs/src/ai/use-a-local-model.md
@@ -39,7 +39,7 @@ Zed automatically discovers the served models with their context length and tool
 ```json [settings]
 {
   "language_models": {
-    "llama_cpp": {
+    "llama.cpp": {
       "api_url": "http://localhost:8080",
       "auto_discover": false,
       "available_models": [
@@ -63,7 +63,7 @@ Zed uses the context length the server reports (`/props`). Override it for all m
 ```json [settings]
 {
   "language_models": {
-    "llama_cpp": {
+    "llama.cpp": {
       "context_window": 8192
     }
   }