diff --git a/crates/service/web/storyteller_web/src/main.rs b/crates/service/web/storyteller_web/src/main.rs
index 9f28e84a45..387e47edb6 100644
--- a/crates/service/web/storyteller_web/src/main.rs
+++ b/crates/service/web/storyteller_web/src/main.rs
@@ -23,6 +23,7 @@ pub const RESERVED_SUBSTRINGS : &'static str = include_str!("../../../../../db/r
 pub mod billing;
 pub mod configs;
 pub mod http_server;
+pub mod memory_cache;
 pub mod model;
 pub mod routes;
 pub mod server_state;
diff --git a/crates/service/web/storyteller_web/src/memory_cache/mod.rs b/crates/service/web/storyteller_web/src/memory_cache/mod.rs
new file mode 100644
index 0000000000..93062b0a1d
--- /dev/null
+++ b/crates/service/web/storyteller_web/src/memory_cache/mod.rs
@@ -0,0 +1 @@
+pub mod model_token_to_info_cache;
\ No newline at end of file
diff --git a/crates/service/web/storyteller_web/src/memory_cache/model_token_to_info_cache.rs b/crates/service/web/storyteller_web/src/memory_cache/model_token_to_info_cache.rs
new file mode 100644
index 0000000000..4e4abb3d81
--- /dev/null
+++ b/crates/service/web/storyteller_web/src/memory_cache/model_token_to_info_cache.rs
@@ -0,0 +1,57 @@
+use anyhow::anyhow;
+use enums::by_table::generic_inference_jobs::inference_category::InferenceCategory;
+use enums::by_table::generic_inference_jobs::inference_model_type::InferenceModelType;
+use errors::AnyhowResult;
+use std::collections::HashMap;
+use std::sync::{Arc, LockResult, RwLock};
+
+#[derive(Clone)]
+pub struct ModelInfoLite {
+  pub inference_category: InferenceCategory,
+  pub model_type: InferenceModelType,
+}
+
+/// Model token to type will never change, so we can cache them indefinitely.
+/// This is for generic inference jobs only.
+pub struct ModelTokenToInfoCache {
+  database: Arc<RwLock<HashMap<String, ModelInfoLite>>>,
+}
+
+impl ModelTokenToInfoCache {
+  pub fn new() -> Self {
+    Self {
+      database: Arc::new(RwLock::new(HashMap::new())),
+    }
+  }
+
+  pub fn insert_one(&self, token: &str, info: ModelInfoLite) -> AnyhowResult<()> {
+    match self.database.write() {
+      Err(err) => return Err(anyhow!("lock err: {:?}", err)),
+      Ok(mut lock) => {
+        lock.insert(token.to_string(), info);
+      }
+    }
+    Ok(())
+  }
+
+  pub fn insert_many(&self, records: Vec<(&str, ModelInfoLite)>) -> AnyhowResult<()> {
+    match self.database.write() {
+      Err(err) => return Err(anyhow!("lock err: {:?}", err)),
+      Ok(mut lock) => {
+        for (token, info) in records.into_iter() {
+          lock.insert(token.to_string(), info);
+        }
+      }
+    }
+    Ok(())
+  }
+
+  pub fn get_info(&self, token: &str) -> AnyhowResult<Option<ModelInfoLite>> {
+    match self.database.read() {
+      Err(err) => return Err(anyhow!("lock err: {:?}", err)),
+      Ok(lock) => {
+        Ok(lock.get(token).map(|item| item.clone()))
+      }
+    }
+  }
+}