Backwards compat

backend/app/agent.py

@@ -268,7 +268,7 @@ def __init__(
     )
     .configurable_alternatives(
         ConfigurableField(id="type", name="Bot Type"),
-        default_key="assistant",
+        default_key="agent",
         prefix_keys=True,
         chatbot=chatbot,
         chat_retrieval=chat_retrieval,

backend/app/checkpoint.py

@@ -71,8 +71,11 @@ def get(self, config: RunnableConfig) -> Checkpoint | None:
             value.pop("__pregel_version")
             value.pop("__pregel_ts")
             checkpoint = empty_checkpoint()
-            checkpoint["channel_values"] = value
-            for key in value:
+            if value.get("messages"):
+                checkpoint["channel_values"] = {"__root__": value["messages"][1]}
+            else:
+                checkpoint["channel_values"] = {}
+            for key in checkpoint["channel_values"]:
                 checkpoint["channel_versions"][key] = 1
             return checkpoint
         else:

frontend/src/components/Config.tsx

@@ -321,8 +321,8 @@ export function Config(props: {
   }, [props.config, props.configDefaults]);
   useEffect(() => {
     if (dropzone.acceptedFiles.length > 0) {
-      if (typeValue === "assistant") {
-        const toolsKey = "type==assistant/tools";
+      if (typeValue === "agent") {
+        const toolsKey = "type==agent/tools";
         setValues((values) => ({
           configurable: {
             ...values?.configurable,

frontend/src/components/ConfigList.tsx

@@ -36,8 +36,8 @@ function ConfigItem(props: {
             {
               TYPES[
                 (props.config.config.configurable?.type ??
-                  "assistant") as keyof typeof TYPES
-              ].title
+                  "agent") as keyof typeof TYPES
+              ]?.title
             }
           </span>
         </div>

frontend/src/constants.ts

@@ -1,6 +1,6 @@
 export const TYPES = {
-  assistant: {
-    id: "assistant",
+  agent: {
+    id: "agent",
     title: "Assistant",
     description:
       "These GPTs can use an arbitrary number of tools, and you can give them arbitrary instructions. The LLM itself is responsible for deciding which tools to call and how many times to call them. This makes them super powerful and flexible, but they can be unreliable at times! As such, only a subset of the most performant models work with these.",