AI should always search for n results (n determing by env setting) and pick the best 1 or 2

Author: mbertrand

ai_chat/agents.py

@@ -142,31 +142,33 @@ class SearchAgentService(BaseChatAgentService):
     TASK_NAME = "SEARCH_TASK"
 
     INSTRUCTIONS = f"""You are a helpful MIT learning resource recommendation
-    assistant. Use the provided tool to search the MIT catalog and provide relevant
-    recommendations based on the user's message.  If the user asks subsequent questions
-    about those results, answer using the information provided in that response.
+assistant. Use the provided tool to search the MIT catalog for results then
+recommend the best 1 or 2 based on the user's message.  Make the title of the
+resource a clickable link to the resource's URL. If the user asks subsequent
+questions about those results, answer using the information provided in that
+response.
 
-    Always run the tool to answer questions only based on the function search results.
-    VERY IMPORTANT: NEVER USE ANY INFORMATION OUTSIDE OF THE FUNCTION RESULTS TO
-    ANSWER QUESTIONS.
+Always run the tool to answer questions only based on the function search results.
+VERY IMPORTANT: NEVER USE ANY INFORMATION OUTSIDE OF THE FUNCTION RESULTS TO
+ANSWER QUESTIONS.
 
-    If a user asks for resources "offered by" or "from" an institution,
-    you should include the offered_by parameter based on the following
-    dictionary: {OfferedBy.as_dict()}  DO NOT USE THE offered_by FILTER OTHERWISE.
+If a user asks for resources "offered by" or "from" an institution,
+you should include the offered_by parameter based on the following
+dictionary: {OfferedBy.as_dict()}  DO NOT USE THE offered_by FILTER OTHERWISE.
 
-    If the user mentions courses, programs, videos, or podcasts in particular, filter
-    the search by resource_category.  DO NOT USE THE resource_type FILTER OTHERWISE.
+If the user mentions courses, programs, videos, or podcasts in particular, filter
+the search by resource_category.  DO NOT USE THE resource_type FILTER OTHERWISE.
 
-    If the user asks what other courses are taught by a particular instructor,
-    search the catalog for courses taught by that instructor.
+If the user asks what other courses are taught by a particular instructor,
+search the catalog for courses taught by that instructor.
 
-    If the user asks for introductory, intermediate, or advanced courses,
-    use the level filter. DO NOT USE THE level FILTER OTHERWISE.
+If the user asks for introductory, intermediate, or advanced courses,
+use the level filter. DO NOT USE THE level FILTER OTHERWISE.
 
-    Do not use the certificate filer unless the user specifically asks for
-    resources that do or do not offer certificates.
+Do not use the certificate filer unless the user specifically asks for
+resources that do or do not offer certificates.
 
-    Always explain your reasoning for recommending specific resources.
+Always explain your reasoning for recommending specific resources.
     """
 
     class SearchToolSchema(pydantic.BaseModel):
@@ -178,7 +180,6 @@ class SearchToolSchema(pydantic.BaseModel):
         free: Optional[bool]
         certificate: Optional[bool]
         offered_by: Optional[enum_zip("offered_by", OfferedBy.names())]
-        limit: Optional[int]
 
     def __init__(  # noqa: PLR0913
         self,
@@ -213,14 +214,13 @@ def search_courses(self, q: str, **kwargs) -> str:
         return simplified results as a JSON string
         """
 
-        params = {"q": q}
+        params = {"q": q, "limit": settings.AI_MIT_SEARCH_LIMIT}
 
         valid_params = {
             "level": kwargs.get("level"),
             "resource_type": kwargs.get("resource_type"),
             "free": kwargs.get("free"),
             "offered_by": kwargs.get("offered_by"),
-            "limit": kwargs.get("limit"),
             "certificate": kwargs.get("certificate"),
         }
         params.update({k: v for k, v in valid_params.items() if v is not None})

ai_chat/utils.py

@@ -6,6 +6,6 @@
 log = logging.getLogger(__name__)
 
 
-def enum_zip(label: str, names: tuple[str]) -> type[Enum]:
+def enum_zip(label: str, names: tuple[str or int]) -> type[Enum]:
     """Create a new enum from a tuple"""
     return Enum(label, dict(zip(names, names)))

app.json

@@ -35,6 +35,46 @@
       "description": "The API used for the AI model",
       "required": false
     },
+    "AI_PROXY_CLASS": {
+      "description": "Proxy class for AI functionality",
+      "required": false
+    },
+    "AI_PROXY_URL": {
+      "description": "URL for AI proxy",
+      "required": false
+    },
+    "AI_PROXY_AUTH_TOKEN": {
+      "description": "Auth token for AI proxy",
+      "required": false
+    },
+    "AI_MAX_PARALLEL_REQUESTS": {
+      "description": "Max parallel requests/user for AI functionality",
+      "required": false
+    },
+    "AI_TPM_LIMIT": {
+      "description": "Tokens/minute limit per user for AI functionality",
+      "required": false
+    },
+    "AI_RPM_LIMIT": {
+      "description": "Requests/minute limit per user for AI functionality",
+      "required": false
+    },
+    "AI_BUDGET_DURATION": {
+      "description": "Length of time before a user's budget usage resets",
+      "required": false
+    },
+    "AI_MAX_BUDGET": {
+      "description": "Max budget per user for AI functionality",
+      "required": false
+    },
+    "AI_ANON_LIMIT_MULTIPLIER": {
+      "description": "Multiplier for per-user limit/budget shared by anonymous users",
+      "required": false
+    },
+    "AI_MIT_SEARCH_LIMIT": {
+      "description": "Limit parameter value for AI search agent",
+      "required": false
+    },
     "ALLOWED_HOSTS": {
       "description": "",
       "required": false

frontends/api/src/generated/v0/api.ts

@@ -528,12 +528,6 @@ export interface ChatRequestRequest {
    * @memberof ChatRequestRequest
    */
   message: string
-  /**
-   *
-   * @type {string}
-   * @memberof ChatRequestRequest
-   */
-  ai?: string
   /**
    *
    * @type {string}

frontends/api/src/generated/v0/common.ts

@@ -194,7 +194,7 @@ export const createRequestFunction = function (
     const axiosRequestArgs = {
       ...axiosArgs.options,
       url:
-        (axios.defaults.baseURL ? "" : (configuration?.basePath ?? basePath)) +
+        (axios.defaults.baseURL ? "" : configuration?.basePath ?? basePath) +
         axiosArgs.url,
     }
     return axios.request<T, R>(axiosRequestArgs)

frontends/api/src/generated/v1/common.ts

@@ -194,7 +194,7 @@ export const createRequestFunction = function (
     const axiosRequestArgs = {
       ...axiosArgs.options,
       url:
-        (axios.defaults.baseURL ? "" : (configuration?.basePath ?? basePath)) +
+        (axios.defaults.baseURL ? "" : configuration?.basePath ?? basePath) +
         axiosArgs.url,
     }
     return axios.request<T, R>(axiosRequestArgs)

main/settings.py

@@ -828,3 +828,4 @@ def get_all_config_keys():
 AI_BUDGET_DURATION = get_string(name="AI_BUDGET_DURATION", default="60m")
 AI_MAX_BUDGET = get_float(name="AI_MAX_BUDGET", default=0.05)
 AI_ANON_LIMIT_MULTIPLIER = get_float(name="AI_ANON_LIMIT_MULTIPLIER", default=10.0)
+AI_MIT_SEARCH_LIMIT = get_int(name="AI_MIT_SEARCH_LIMIT", default=10)

openapi/specs/v0.yaml

@@ -1278,10 +1278,6 @@ components:
         message:
           type: string
           minLength: 1
-        ai:
-          type: string
-          minLength: 1
-          default: openai
         model:
           type: string
           minLength: 1