Merge branch 'danswer-ai:main' into main

backend/alembic/versions/0a2b51deb0b8_add_starter_prompts.py

@@ -0,0 +1,31 @@
+"""Add starter prompts
+
+Revision ID: 0a2b51deb0b8
+Revises: 5f4b8568a221
+Create Date: 2024-03-02 23:23:49.960309
+
+"""
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy.dialects import postgresql
+
+# revision identifiers, used by Alembic.
+revision = "0a2b51deb0b8"
+down_revision = "5f4b8568a221"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "persona",
+        sa.Column(
+            "starter_messages",
+            postgresql.JSONB(astext_type=sa.Text()),
+            nullable=True,
+        ),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("persona", "starter_messages")

backend/danswer/chat/load_yamls.py

@@ -89,6 +89,7 @@ def load_personas_from_yaml(
                 if persona.get("num_chunks") is not None
                 else default_chunks,
                 llm_relevance_filter=persona.get("llm_relevance_filter"),
+                starter_messages=persona.get("starter_messages"),
                 llm_filter_extraction=persona.get("llm_filter_extraction"),
                 llm_model_version_override=None,
                 recency_bias=RecencyBiasSetting(persona["recency_bias"]),

backend/danswer/configs/app_configs.py

@@ -19,7 +19,9 @@
 # User Facing Features Configs
 #####
 BLURB_SIZE = 128  # Number Encoder Tokens included in the chunk blurb
-GENERATIVE_MODEL_ACCESS_CHECK_FREQ = 86400  # 1 day
+GENERATIVE_MODEL_ACCESS_CHECK_FREQ = int(
+    os.environ.get("GENERATIVE_MODEL_ACCESS_CHECK_FREQ") or 86400
+)  # 1 day
 DISABLE_GENERATIVE_AI = os.environ.get("DISABLE_GENERATIVE_AI", "").lower() == "true"
 
 

backend/danswer/configs/chat_configs.py

@@ -37,6 +37,9 @@
 DISABLE_LLM_CHOOSE_SEARCH = (
     os.environ.get("DISABLE_LLM_CHOOSE_SEARCH", "").lower() == "true"
 )
+DISABLE_LLM_QUERY_REPHRASE = (
+    os.environ.get("DISABLE_LLM_QUERY_REPHRASE", "").lower() == "true"
+)
 # 1 edit per 20 characters, currently unused due to fuzzy match being too slow
 QUOTE_ALLOWED_ERROR_PERCENT = 0.05
 QA_TIMEOUT = int(os.environ.get("QA_TIMEOUT") or "60")  # 60 seconds

backend/danswer/db/chat.py

@@ -23,6 +23,7 @@
 from danswer.db.models import Prompt
 from danswer.db.models import SearchDoc
 from danswer.db.models import SearchDoc as DBSearchDoc
+from danswer.db.models import StarterMessage
 from danswer.search.models import RecencyBiasSetting
 from danswer.search.models import RetrievalDocs
 from danswer.search.models import SavedSearchDoc
@@ -465,6 +466,7 @@ def upsert_persona(
     prompts: list[Prompt] | None,
     document_sets: list[DBDocumentSet] | None,
     llm_model_version_override: str | None,
+    starter_messages: list[StarterMessage] | None,
     shared: bool,
     db_session: Session,
     persona_id: int | None = None,
@@ -490,6 +492,7 @@ def upsert_persona(
         persona.recency_bias = recency_bias
         persona.default_persona = default_persona
         persona.llm_model_version_override = llm_model_version_override
+        persona.starter_messages = starter_messages
         persona.deleted = False  # Un-delete if previously deleted
 
         # Do not delete any associations manually added unless
@@ -516,6 +519,7 @@ def upsert_persona(
             prompts=prompts or [],
             document_sets=document_sets or [],
             llm_model_version_override=llm_model_version_override,
+            starter_messages=starter_messages,
         )
         db_session.add(persona)
 

backend/danswer/db/models.py

@@ -716,6 +716,15 @@ class Prompt(Base):
     )
 
 
+class StarterMessage(TypedDict):
+    """NOTE: is a `TypedDict` so it can be used as a type hint for a JSONB column
+    in Postgres"""
+
+    name: str
+    description: str
+    message: str
+
+
 class Persona(Base):
     __tablename__ = "persona"
 
@@ -744,6 +753,9 @@ class Persona(Base):
     llm_model_version_override: Mapped[str | None] = mapped_column(
         String, nullable=True
     )
+    starter_messages: Mapped[list[StarterMessage] | None] = mapped_column(
+        postgresql.JSONB(), nullable=True
+    )
     # Default personas are configured via backend during deployment
     # Treated specially (cannot be user edited etc.)
     default_persona: Mapped[bool] = mapped_column(Boolean, default=False)

backend/danswer/db/slack_bot_config.py

@@ -59,6 +59,7 @@ def create_slack_bot_persona(
         prompts=None,
         document_sets=document_sets,
         llm_model_version_override=None,
+        starter_messages=None,
         shared=True,
         default_persona=False,
         db_session=db_session,

backend/danswer/llm/chat_llm.py

@@ -134,7 +134,11 @@ def __init__(
             max_tokens=max_output_tokens,
             temperature=temperature,
             request_timeout=timeout,
-            model_kwargs=DefaultMultiLLM.DEFAULT_MODEL_PARAMS,
+            # LiteLLM and some model providers don't handle these params well
+            # only turning it on for OpenAI
+            model_kwargs=DefaultMultiLLM.DEFAULT_MODEL_PARAMS
+            if model_provider == "openai"
+            else {},
             verbose=should_be_verbose(),
             max_retries=0,  # retries are handled outside of langchain
         )

backend/danswer/secondary_llm_flows/query_expansion.py

@@ -2,6 +2,7 @@
 from typing import cast
 
 from danswer.chat.chat_utils import combine_message_chain
+from danswer.configs.chat_configs import DISABLE_LLM_QUERY_REPHRASE
 from danswer.configs.model_configs import GEN_AI_HISTORY_CUTOFF
 from danswer.db.models import ChatMessage
 from danswer.llm.exceptions import GenAIDisabledException
@@ -95,6 +96,10 @@ def history_based_query_rephrase(
 ) -> str:
     user_query = cast(str, query_message.message)
 
+    # Globally disabled, just use the exact user query
+    if DISABLE_LLM_QUERY_REPHRASE:
+        return user_query
+
     if not user_query:
         raise ValueError("Can't rephrase/search an empty query")
 

backend/danswer/server/features/persona/api.py

@@ -66,6 +66,7 @@ def create_update_persona(
             prompts=prompts,
             document_sets=document_sets,
             llm_model_version_override=create_persona_request.llm_model_version_override,
+            starter_messages=create_persona_request.starter_messages,
             shared=create_persona_request.shared,
             db_session=db_session,
         )

backend/danswer/server/features/persona/models.py

@@ -1,6 +1,7 @@
 from pydantic import BaseModel
 
 from danswer.db.models import Persona
+from danswer.db.models import StarterMessage
 from danswer.search.models import RecencyBiasSetting
 from danswer.server.features.document_set.models import DocumentSet
 from danswer.server.features.prompt.models import PromptSnapshot
@@ -17,6 +18,7 @@ class CreatePersonaRequest(BaseModel):
     prompt_ids: list[int]
     document_set_ids: list[int]
     llm_model_version_override: str | None = None
+    starter_messages: list[StarterMessage] | None = None
 
 
 class PersonaSnapshot(BaseModel):
@@ -30,6 +32,7 @@ class PersonaSnapshot(BaseModel):
     llm_relevance_filter: bool
     llm_filter_extraction: bool
     llm_model_version_override: str | None
+    starter_messages: list[StarterMessage] | None
     default_persona: bool
     prompts: list[PromptSnapshot]
     document_sets: list[DocumentSet]
@@ -50,6 +53,7 @@ def from_model(cls, persona: Persona) -> "PersonaSnapshot":
             llm_relevance_filter=persona.llm_relevance_filter,
             llm_filter_extraction=persona.llm_filter_extraction,
             llm_model_version_override=persona.llm_model_version_override,
+            starter_messages=persona.starter_messages,
             default_persona=persona.default_persona,
             prompts=[PromptSnapshot.from_model(prompt) for prompt in persona.prompts],
             document_sets=[

backend/danswer/server/manage/administrative.py

@@ -10,6 +10,7 @@
 from sqlalchemy.orm import Session
 
 from danswer.auth.users import current_admin_user
+from danswer.configs.app_configs import GENERATIVE_MODEL_ACCESS_CHECK_FREQ
 from danswer.configs.constants import GEN_AI_API_KEY_STORAGE_KEY
 from danswer.configs.constants import GEN_AI_DETECTED_MODEL
 from danswer.configs.model_configs import GEN_AI_MODEL_PROVIDER
@@ -170,8 +171,7 @@ def validate_existing_genai_api_key(
         last_check = datetime.fromtimestamp(
             cast(float, kv_store.load(GEN_AI_KEY_CHECK_TIME)), tz=timezone.utc
         )
-        # GENERATIVE_MODEL_ACCESS_CHECK_FREQ
-        check_freq_sec = timedelta(seconds=1)
+        check_freq_sec = timedelta(seconds=GENERATIVE_MODEL_ACCESS_CHECK_FREQ)
         if curr_time - last_check < check_freq_sec:
             return
     except ConfigNotFoundError:

deployment/docker_compose/docker-compose.dev.yml

@@ -42,6 +42,7 @@ services:
       - DISABLE_LLM_FILTER_EXTRACTION=${DISABLE_LLM_FILTER_EXTRACTION:-}
       - DISABLE_LLM_CHUNK_FILTER=${DISABLE_LLM_CHUNK_FILTER:-}
       - DISABLE_LLM_CHOOSE_SEARCH=${DISABLE_LLM_CHOOSE_SEARCH:-}
+      - DISABLE_LLM_QUERY_REPHRASE=${DISABLE_LLM_QUERY_REPHRASE:-}
       - DISABLE_GENERATIVE_AI=${DISABLE_GENERATIVE_AI:-}
       # Query Options
       - DOC_TIME_DECAY=${DOC_TIME_DECAY:-}  # Recency Bias for search results, decay at 1 / (1 + DOC_TIME_DECAY * x years)
@@ -76,6 +77,8 @@ services:
       - model_cache_torch:/root/.cache/torch/
       - model_cache_nltk:/root/nltk_data/
       - model_cache_huggingface:/root/.cache/huggingface/
+    extra_hosts:
+      - "host.docker.internal:host-gateway"
     logging:
       driver: json-file
       options:
@@ -106,7 +109,9 @@ services:
       - DISABLE_LLM_FILTER_EXTRACTION=${DISABLE_LLM_FILTER_EXTRACTION:-}
       - DISABLE_LLM_CHUNK_FILTER=${DISABLE_LLM_CHUNK_FILTER:-}
       - DISABLE_LLM_CHOOSE_SEARCH=${DISABLE_LLM_CHOOSE_SEARCH:-}
+      - DISABLE_LLM_QUERY_REPHRASE=${DISABLE_LLM_QUERY_REPHRASE:-}
       - DISABLE_GENERATIVE_AI=${DISABLE_GENERATIVE_AI:-}
+      - GENERATIVE_MODEL_ACCESS_CHECK_FREQ=${GENERATIVE_MODEL_ACCESS_CHECK_FREQ:-}
       # Query Options
       - DOC_TIME_DECAY=${DOC_TIME_DECAY:-}  # Recency Bias for search results, decay at 1 / (1 + DOC_TIME_DECAY * x years)
       - HYBRID_ALPHA=${HYBRID_ALPHA:-}  # Hybrid Search Alpha (0 for entirely keyword, 1 for entirely vector)
@@ -164,6 +169,8 @@ services:
       - model_cache_torch:/root/.cache/torch/
       - model_cache_nltk:/root/nltk_data/
       - model_cache_huggingface:/root/.cache/huggingface/
+    extra_hosts:
+      - "host.docker.internal:host-gateway"
     logging:
       driver: json-file
       options:
@@ -176,6 +183,7 @@ services:
       dockerfile: Dockerfile
       args:
         - NEXT_PUBLIC_DISABLE_STREAMING=${NEXT_PUBLIC_DISABLE_STREAMING:-false}
+        - NEXT_PUBLIC_NEW_CHAT_DIRECTS_TO_SAME_PERSONA=${NEXT_PUBLIC_NEW_CHAT_DIRECTS_TO_SAME_PERSONA:-false}
     depends_on:
       - api_server
     restart: always

deployment/docker_compose/docker-compose.prod-no-letsencrypt.yml

@@ -25,6 +25,8 @@ services:
       - model_cache_torch:/root/.cache/torch/
       - model_cache_nltk:/root/nltk_data/
       - model_cache_huggingface:/root/.cache/huggingface/
+    extra_hosts:
+      - "host.docker.internal:host-gateway"
     logging:
       driver: json-file
       options:
@@ -52,6 +54,8 @@ services:
       - model_cache_torch:/root/.cache/torch/
       - model_cache_nltk:/root/nltk_data/
       - model_cache_huggingface:/root/.cache/huggingface/
+    extra_hosts:
+      - "host.docker.internal:host-gateway"
     logging:
       driver: json-file
       options:
@@ -64,6 +68,7 @@ services:
       dockerfile: Dockerfile
       args:
         - NEXT_PUBLIC_DISABLE_STREAMING=${NEXT_PUBLIC_DISABLE_STREAMING:-false}
+        - NEXT_PUBLIC_NEW_CHAT_DIRECTS_TO_SAME_PERSONA=${NEXT_PUBLIC_NEW_CHAT_DIRECTS_TO_SAME_PERSONA:-false}
     depends_on:
       - api_server
     restart: always

deployment/docker_compose/docker-compose.prod.yml

@@ -25,6 +25,8 @@ services:
       - model_cache_torch:/root/.cache/torch/
       - model_cache_nltk:/root/nltk_data/
       - model_cache_huggingface:/root/.cache/huggingface/
+    extra_hosts:
+      - "host.docker.internal:host-gateway"
     logging:
       driver: json-file
       options:
@@ -52,6 +54,8 @@ services:
       - model_cache_torch:/root/.cache/torch/
       - model_cache_nltk:/root/nltk_data/
       - model_cache_huggingface:/root/.cache/huggingface/
+    extra_hosts:
+      - "host.docker.internal:host-gateway"
     logging:
       driver: json-file
       options:
@@ -64,6 +68,7 @@ services:
       dockerfile: Dockerfile
       args:
         - NEXT_PUBLIC_DISABLE_STREAMING=${NEXT_PUBLIC_DISABLE_STREAMING:-false}
+        - NEXT_PUBLIC_NEW_CHAT_DIRECTS_TO_SAME_PERSONA=${NEXT_PUBLIC_NEW_CHAT_DIRECTS_TO_SAME_PERSONA:-false}
     depends_on:
       - api_server
     restart: always

deployment/kubernetes/env-configmap.yaml

@@ -26,6 +26,7 @@ data:
   DISABLE_LLM_FILTER_EXTRACTION: ""
   DISABLE_LLM_CHUNK_FILTER: ""
   DISABLE_LLM_CHOOSE_SEARCH: ""
+  DISABLE_LLM_QUERY_REPHRASE: ""
   # Query Options
   DOC_TIME_DECAY: ""
   HYBRID_ALPHA: ""

web/Dockerfile

@@ -35,6 +35,9 @@ ENV NEXT_TELEMETRY_DISABLED 1
 ARG NEXT_PUBLIC_DISABLE_STREAMING
 ENV NEXT_PUBLIC_DISABLE_STREAMING=${NEXT_PUBLIC_DISABLE_STREAMING}
 
+ARG NEXT_PUBLIC_NEW_CHAT_DIRECTS_TO_SAME_PERSONA
+ENV NEXT_PUBLIC_NEW_CHAT_DIRECTS_TO_SAME_PERSONA=${NEXT_PUBLIC_NEW_CHAT_DIRECTS_TO_SAME_PERSONA}
+
 RUN npm run build
 
 # Step 3. Production image, copy all the files and run next
@@ -70,6 +73,9 @@ COPY --from=builder --chown=nextjs:nodejs /app/.next/static ./.next/static
 ARG NEXT_PUBLIC_DISABLE_STREAMING
 ENV NEXT_PUBLIC_DISABLE_STREAMING=${NEXT_PUBLIC_DISABLE_STREAMING}
 
+ARG NEXT_PUBLIC_NEW_CHAT_DIRECTS_TO_SAME_PERSONA
+ENV NEXT_PUBLIC_NEW_CHAT_DIRECTS_TO_SAME_PERSONA=${NEXT_PUBLIC_NEW_CHAT_DIRECTS_TO_SAME_PERSONA}
+
 # Note: Don't expose ports here, Compose will handle that for us if necessary. 
 # If you want to run this without compose, specify the ports to 
 # expose via cli

web/src/app/admin/personas/HidableSection.tsx

@@ -0,0 +1,50 @@
+import { useState } from "react";
+import { FiChevronDown, FiChevronRight } from "react-icons/fi";
+
+export function SectionHeader({
+  children,
+  includeMargin = true,
+}: {
+  children: string | JSX.Element;
+  includeMargin?: boolean;
+}) {
+  return (
+    <div
+      className={"font-bold text-xl my-auto" + (includeMargin ? " mb-4" : "")}
+    >
+      {children}
+    </div>
+  );
+}
+
+export function HidableSection({
+  children,
+  sectionTitle,
+  defaultHidden = false,
+}: {
+  children: string | JSX.Element;
+  sectionTitle: string | JSX.Element;
+  defaultHidden?: boolean;
+}) {
+  const [isHidden, setIsHidden] = useState(defaultHidden);
+
+  return (
+    <div>
+      <div
+        className="flex hover:bg-hover-light rounded cursor-pointer p-2"
+        onClick={() => setIsHidden(!isHidden)}
+      >
+        <SectionHeader includeMargin={false}>{sectionTitle}</SectionHeader>
+        <div className="my-auto ml-auto p-1">
+          {isHidden ? (
+            <FiChevronRight size={24} />
+          ) : (
+            <FiChevronDown size={24} />
+          )}
+        </div>
+      </div>
+
+      {!isHidden && <div className="mx-2 mt-2">{children}</div>}
+    </div>
+  );
+}