Merge pull request #662 from PrefectHQ/fix-problems

need numpy for embeddings when http retrieval

cookbook/slackbot/Dockerfile.slackbot

@@ -13,7 +13,7 @@ RUN apt-get update && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
-RUN pip install ".[prefect]"
+RUN pip install ".[slackbot]"
 
 EXPOSE 4200
 

cookbook/slackbot/keywords.py

@@ -0,0 +1,68 @@
+from marvin import ai_fn
+from marvin.utilities.slack import post_slack_message
+from prefect import task
+from prefect.blocks.system import JSON, Secret, String
+from prefect.exceptions import ObjectNotFound
+
+"""
+Define a map between keywords and the relationships we want to check for
+in a given message related to that keyword.
+"""
+
+keywords = (
+    ("429", "rate limit"),
+    ("SSO", "Single Sign On", "RBAC", "Roles", "Role Based Access Controls"),
+)
+
+relationships = (
+    "The user is getting rate limited",
+    "The user is asking about a paid feature",
+)
+
+
+async def get_reduced_kw_relationship_map() -> dict:
+    try:
+        json_map = (await JSON.load("keyword-relationship-map")).value
+    except (ObjectNotFound, ValueError):
+        json_map = {"keywords": keywords, "relationships": relationships}
+        await JSON(value=json_map).save("keyword-relationship-map")
+
+    return {
+        keyword: relationship
+        for keyword_tuple, relationship in zip(
+            json_map["keywords"], json_map["relationships"]
+        )
+        for keyword in keyword_tuple
+    }
+
+
+@ai_fn
+def activation_score(message: str, keyword: str, target_relationship: str) -> float:
+    """Return a score between 0 and 1 indicating whether the target relationship exists
+    between the message and the keyword"""
+
+
+@task
+async def handle_keywords(message: str, channel_name: str, asking_user: str, link: str):
+    keyword_relationships = await get_reduced_kw_relationship_map()
+    keywords = [
+        keyword for keyword in keyword_relationships.keys() if keyword in message
+    ]
+    for keyword in keywords:
+        target_relationship = keyword_relationships.get(keyword)
+        if not target_relationship:
+            continue
+        score = activation_score(message, keyword, target_relationship)
+        if score > 0.5:
+            await post_slack_message(
+                message=(
+                    f"A user ({asking_user}) just asked a question in"
+                    f" {channel_name} that contains the keyword `{keyword}`, and I'm"
+                    f" {score*100:.0f}% sure that their message indicates the"
+                    f" following:\n\n**{target_relationship!r}**.\n\n[Go to"
+                    f" message]({link})"
+                ),
+                channel_id=(await String.load("ask-marvin-tests-channel-id")).value,
+                auth_token=(await Secret.load("slack-api-token")).get(),
+            )
+            return

cookbook/slackbot/start.py

@@ -4,12 +4,18 @@
 import uvicorn
 from cachetools import TTLCache
 from fastapi import FastAPI, HTTPException, Request
+from keywords import handle_keywords
 from marvin import Assistant
 from marvin.beta.assistants import Thread
 from marvin.tools.github import search_github_issues
 from marvin.tools.retrieval import multi_query_chroma
 from marvin.utilities.logging import get_logger
-from marvin.utilities.slack import SlackPayload, post_slack_message
+from marvin.utilities.slack import (
+    SlackPayload,
+    get_channel_name,
+    get_workspace_info,
+    post_slack_message,
+)
 from prefect import flow, task
 from prefect.states import Completed
 
@@ -31,6 +37,16 @@ async def handle_message(payload: SlackPayload):
         assistant_thread = CACHE.get(thread, Thread())
         CACHE[thread] = assistant_thread
 
+        await handle_keywords.submit(
+            message=cleaned_message,
+            channel_name=await get_channel_name(event.channel),
+            asking_user=event.user,
+            link=(  # to user's message
+                f"{(await get_workspace_info()).get('url')}archives/"
+                f"{event.channel}/p{event.ts.replace('.', '')}"
+            ),
+        )
+
         with Assistant(
             name="Marvin (from Hitchhiker's Guide to the Galaxy)",
             tools=[task(multi_query_chroma), task(search_github_issues)],

pyproject.toml

@@ -27,6 +27,7 @@ dependencies = [
 
 [project.optional-dependencies]
 generator = ["datamodel-code-generator>=0.20.0"]
+prefect = ["prefect>=2.14.9"]
 dev = [
     "marvin[tests]",
     "black[jupyter]",
@@ -43,16 +44,14 @@ dev = [
     "ruff",
 ]
 tests = [
-    "marvin[openai,anthropic]",
     "pytest-asyncio~=0.20",
     "pytest-env>=0.8,<2.0",
     "pytest-rerunfailures>=10,<13",
     "pytest-sugar~=0.9",
     "pytest~=7.3.1",
     "pytest-timeout",
 ]
-
-prefect = ["prefect>=2.14.5"]
+slackbot = ["marvin[prefect]", "numpy"]
 
 [project.urls]
 Code = "https://github.com/prefecthq/marvin"

src/marvin/tools/github.py

@@ -1,3 +1,4 @@
+import os
 from datetime import datetime
 from typing import List, Optional
 
@@ -8,6 +9,13 @@
 from marvin.utilities.strings import slice_tokens
 
 
+def get_token() -> str:
+    try:
+        return marvin.settings.github_token
+    except AttributeError:
+        return os.environ.get("MARVIN_GITHUB_TOKEN", "")
+
+
 class GitHubUser(BaseModel):
     """GitHub user."""
 
@@ -60,8 +68,7 @@ async def search_github_issues(
     """
     headers = {"Accept": "application/vnd.github.v3+json"}
 
-    if token := marvin.settings.github_token:
-        headers["Authorization"] = f"Bearer {token}"
+    headers["Authorization"] = f"Bearer {get_token()}"
 
     async with httpx.AsyncClient() as client:
         response = await client.get(

src/marvin/tools/retrieval.py

@@ -34,7 +34,9 @@ async def create_openai_embeddings(texts: list[str]) -> list[list[float]]:
 
     return (
         (
-            await AsyncOpenAI().embeddings.create(
+            await AsyncOpenAI(
+                api_key=marvin.settings.openai.api_key.get_secret_value()
+            ).embeddings.create(
                 input=[text.replace("\n", " ") for text in texts],
                 model="text-embedding-ada-002",
             )