Merge pull request #705 from Kiln-AI/main

Update docs for v0.22.1

Author: scosman

app/desktop/studio_server/tool_api.py

@@ -12,12 +12,12 @@
     ToolServerType,
 )
 from kiln_ai.datamodel.tool_id import (
-    KILN_TASK_TOOL_ID_PREFIX,
     MCP_LOCAL_TOOL_ID_PREFIX,
     MCP_REMOTE_TOOL_ID_PREFIX,
-    RAG_TOOL_ID_PREFIX,
     KilnBuiltInToolId,
     ToolId,
+    build_kiln_task_tool_id,
+    build_rag_tool_id,
 )
 from kiln_ai.tools.kiln_task_tool import KilnTaskTool
 from kiln_ai.tools.mcp_session_manager import MCPSessionManager
@@ -227,7 +227,7 @@ async def get_available_tools(
         if rag_configs:
             tools = [
                 ToolApiDescription(
-                    id=f"{RAG_TOOL_ID_PREFIX}{rag_config.id}",
+                    id=build_rag_tool_id(rag_config.id),
                     name=rag_config.tool_name,
                     description=f"{rag_config.name}: {rag_config.tool_description}",
                 )
@@ -259,7 +259,7 @@ async def get_available_tools(
                     if not server.properties.get("is_archived", False):
                         task_tools.append(
                             ToolApiDescription(
-                                id=f"{KILN_TASK_TOOL_ID_PREFIX}{server.id}",
+                                id=build_kiln_task_tool_id(server.id),
                                 name=server.properties.get("name") or "",
                                 description=server.properties.get("description") or "",
                             )

app/web_ui/src/routes/(app)/prompts/[project_id]/[task_id]/create/+page.svelte

@@ -8,6 +8,7 @@
   import { createKilnError, KilnError } from "$lib/utils/error_handlers"
   import { goto } from "$app/navigation"
   import posthog from "posthog-js"
+  import { onMount } from "svelte"
 
   $: project_id = $page.params.project_id
   $: task_id = $page.params.task_id
@@ -21,6 +22,12 @@
     "Think step by step, explaining your reasoning."
   let create_error: KilnError | null = null
   let create_loading = false
+  let warn_before_unload = false
+  let mounted = false
+
+  onMount(() => {
+    mounted = true
+  })
 
   async function create_prompt() {
     try {
@@ -64,6 +71,20 @@
       create_loading = false
     }
   }
+
+  // Warn before unload if there's any user input
+  $: prompt_name,
+    prompt_description,
+    prompt,
+    is_chain_of_thought,
+    chain_of_thought_instructions,
+    user_input_detected()
+
+  function user_input_detected() {
+    if (mounted) {
+      warn_before_unload = true
+    }
+  }
 </script>
 
 <div class="max-w-[1400px]">
@@ -83,6 +104,7 @@
         on:submit={create_prompt}
         bind:error={create_error}
         bind:submitting={create_loading}
+        {warn_before_unload}
       >
         <FormElement
           label="Prompt Name"

conftest.py

@@ -207,7 +207,11 @@ def mock_attachment_factory(mock_file_factory):
 
     def create_attachment(
         mime_type: MockFileFactoryMimeType,
+        text: str | None = None,
     ) -> KilnAttachmentModel:
+        if text is not None:
+            return KilnAttachmentModel.from_data(text, mime_type)
+
         path = mock_file_factory(mime_type)
         return KilnAttachmentModel.from_file(path)
 

libs/core/README.md

@@ -43,6 +43,10 @@ The library has a [comprehensive set of docs](https://kiln-ai.github.io/Kiln/kil
   - [Building and Running a Kiln Task from Code](#building-and-running-a-kiln-task-from-code)
   - [Tagging Task Runs Programmatically](#tagging-task-runs-programmatically)
   - [Adding Custom Model or AI Provider from Code](#adding-custom-model-or-ai-provider-from-code)
+- [Taking Kiln RAG to production](#taking-kiln-rag-to-production)
+  - [Load a LlamaIndex Vector Store](#load-a-llamaindex-vector-store)
+  - [Example: LanceDB Cloud](#example-lancedb-cloud)
+  - [Deploy RAG without LlamaIndex](#deploy-rag-without-llamaindex)
 - [Full API Reference](#full-api-reference)
 
 ## Installation
@@ -310,6 +314,78 @@ custom_model_ids.append(new_model)
 Config.shared().custom_models = custom_model_ids
 ```
 
+## Taking Kiln RAG to production
+
+When you're ready to deploy your RAG system, you can export your processed documents to any vector store supported by LlamaIndex. This allows you to use your Kiln-configured chunking and embedding settings in production.
+
+### Load a LlamaIndex Vector Store
+
+Kiln provides a `VectorStoreLoader` that yields your processed document chunks as LlamaIndex `TextNode` objects. These nodes contain the same metadata, chunking and embedding data as your Kiln Search Tool configuration.
+
+```py
+from kiln_ai.datamodel import Project
+from kiln_ai.datamodel.rag import RagConfig
+from kiln_ai.adapters.vector_store_loaders import VectorStoreLoader
+
+# Load your project and RAG configuration
+project = Project.load_from_file("path/to/your/project.kiln")
+rag_config = RagConfig.from_id_and_parent_path("rag-config-id", project.path)
+
+# Create the loader
+loader = VectorStoreLoader(project=project, rag_config=rag_config)
+
+# Export chunks to any LlamaIndex vector store
+async for batch in loader.iter_llama_index_nodes(batch_size=10):
+    # Insert into your chosen vector store
+    # Examples: LanceDB, Pinecone, Chroma, Qdrant, etc.
+    pass
+```
+
+**Supported Vector Stores:** LlamaIndex supports 20+ vector stores including LanceDB, Pinecone, Weaviate, Chroma, Qdrant, and more. See the [full list](https://developers.llamaindex.ai/python/framework/module_guides/storing/vector_stores/).
+
+### Example: LanceDB Cloud
+
+Internally Kiln uses LanceDB. By using LanceDB cloud you'll get the same indexing behaviour as in app.
+
+Here's a complete example using LanceDB Cloud:
+
+```py
+from kiln_ai.datamodel import Project
+from kiln_ai.datamodel.rag import RagConfig
+from kiln_ai.datamodel.vector_store import VectorStoreConfig
+from kiln_ai.adapters.vector_store_loaders import VectorStoreLoader
+from kiln_ai.adapters.vector_store.lancedb_adapter import lancedb_construct_from_config
+
+# Load configurations
+project = Project.load_from_file("path/to/your/project.kiln")
+rag_config = RagConfig.from_id_and_parent_path("rag-config-id", project.path)
+vector_store_config = VectorStoreConfig.from_id_and_parent_path(
+    rag_config.vector_store_config_id, project.path,
+)
+
+# Create LanceDB vector store
+lancedb_store = lancedb_construct_from_config(
+    vector_store_config=vector_store_config,
+    uri="db://my-project",
+    api_key="sk_...",
+    region="us-east-1",
+    table_name="my-documents",  # Created automatically
+)
+
+# Export and insert your documents
+loader = VectorStoreLoader(project=project, rag_config=rag_config)
+async for batch in loader.iter_llama_index_nodes(batch_size=100):
+    await lancedb_store.async_add(batch)
+
+print("Documents successfully exported to LanceDB!")
+```
+
+After export, query your data using [LlamaIndex](https://developers.llamaindex.ai/python/framework-api-reference/storage/vector_store/lancedb/) or the [LanceDB client](https://lancedb.github.io/lancedb/).
+
+### Deploy RAG without LlamaIndex
+
+While Kiln is designed for deploying to LlamaIndex, you don't need to use it. The `iter_llama_index_nodes` returns a `TextNode` object which includes all the data you need to build a RAG index in any stack: embedding, text, document name, chunk ID, etc.
+
 ## Full API Reference
 
 The library can do a lot more than the examples we've shown here.

libs/core/kiln_ai/adapters/model_adapters/litellm_adapter.py

@@ -31,7 +31,11 @@
 )
 from kiln_ai.adapters.model_adapters.litellm_config import LiteLlmConfig
 from kiln_ai.datamodel.json_schema import validate_schema_with_value_error
-from kiln_ai.tools.base_tool import KilnToolInterface, ToolCallContext
+from kiln_ai.tools.base_tool import (
+    KilnToolInterface,
+    ToolCallContext,
+    ToolCallDefinition,
+)
 from kiln_ai.tools.kiln_task_tool import KilnTaskToolResult
 from kiln_ai.utils.exhaustive_error import raise_exhaustive_enum_error
 from kiln_ai.utils.litellm import get_litellm_provider_info
@@ -560,7 +564,7 @@ async def cached_available_tools(self) -> list[KilnToolInterface]:
             self._cached_available_tools = await self.available_tools()
         return self._cached_available_tools
 
-    async def litellm_tools(self) -> list[Dict]:
+    async def litellm_tools(self) -> list[ToolCallDefinition]:
         available_tools = await self.cached_available_tools()
 
         # LiteLLM takes the standard OpenAI-compatible tool call format

libs/core/kiln_ai/adapters/vector_store/lancedb_adapter.py

@@ -5,12 +5,7 @@
 from typing import Any, Dict, List, Literal, Optional, Set, TypedDict
 
 from llama_index.core import StorageContext, VectorStoreIndex
-from llama_index.core.schema import (
-    BaseNode,
-    NodeRelationship,
-    RelatedNodeInfo,
-    TextNode,
-)
+from llama_index.core.schema import BaseNode, TextNode
 from llama_index.core.vector_stores.types import (
     VectorStoreQuery as LlamaIndexVectorStoreQuery,
 )
@@ -24,15 +19,19 @@
     SearchResult,
     VectorStoreQuery,
 )
+from kiln_ai.adapters.vector_store.lancedb_helpers import (
+    convert_to_llama_index_node,
+    deterministic_chunk_id,
+    lancedb_construct_from_config,
+    store_type_to_lancedb_query_type,
+)
 from kiln_ai.datamodel.rag import RagConfig
 from kiln_ai.datamodel.vector_store import (
     VectorStoreConfig,
-    VectorStoreType,
     raise_exhaustive_enum_error,
 )
 from kiln_ai.utils.config import Config
 from kiln_ai.utils.env import temporary_env
-from kiln_ai.utils.uuid import string_to_uuid
 
 logger = logging.getLogger(__name__)
 
@@ -48,6 +47,7 @@ def __init__(
         self,
         rag_config: RagConfig,
         vector_store_config: VectorStoreConfig,
+        lancedb_vector_store: LanceDBVectorStore | None = None,
     ):
         super().__init__(rag_config, vector_store_config)
         self.config_properties = self.vector_store_config.lancedb_properties
@@ -56,17 +56,15 @@ def __init__(
         if vector_store_config.lancedb_properties.nprobes is not None:
             kwargs["nprobes"] = vector_store_config.lancedb_properties.nprobes
 
-        self.lancedb_vector_store = LanceDBVectorStore(
-            mode="create",
-            uri=LanceDBAdapter.lancedb_path_for_config(rag_config),
-            query_type=self.query_type,
-            overfetch_factor=vector_store_config.lancedb_properties.overfetch_factor,
-            vector_column_name=vector_store_config.lancedb_properties.vector_column_name,
-            text_key=vector_store_config.lancedb_properties.text_key,
-            doc_id_key=vector_store_config.lancedb_properties.doc_id_key,
-            **kwargs,
+        # allow overriding the vector store with a custom one, useful for user loading into an arbitrary
+        # deployment
+        self.lancedb_vector_store = (
+            lancedb_vector_store
+            or lancedb_construct_from_config(
+                vector_store_config,
+                uri=LanceDBAdapter.lancedb_path_for_config(rag_config),
+            )
         )
-
         self._index = None
 
     @property
@@ -149,7 +147,7 @@ async def add_chunks_with_embeddings(
 
             chunk_count_for_document = len(chunks)
             deterministic_chunk_ids = [
-                self.compute_deterministic_chunk_id(document_id, chunk_idx)
+                deterministic_chunk_id(document_id, chunk_idx)
                 for chunk_idx in range(chunk_count_for_document)
             ]
 
@@ -176,42 +174,12 @@ async def add_chunks_with_embeddings(
                 zip(chunks_text, embeddings)
             ):
                 node_batch.append(
-                    TextNode(
-                        id_=deterministic_chunk_ids[chunk_idx],
+                    convert_to_llama_index_node(
+                        document_id=document_id,
+                        chunk_idx=chunk_idx,
+                        node_id=deterministic_chunk_id(document_id, chunk_idx),
                         text=chunk_text,
-                        embedding=embedding.vector,
-                        metadata={
-                            # metadata is populated by some internal llama_index logic
-                            # that uses for example the source_node relationship
-                            "kiln_doc_id": document_id,
-                            "kiln_chunk_idx": chunk_idx,
-                            #
-                            # llama_index lancedb vector store automatically sets these metadata:
-                            # "doc_id": "UUID node_id of the Source Node relationship",
-                            # "document_id": "UUID node_id of the Source Node relationship",
-                            # "ref_doc_id": "UUID node_id of the Source Node relationship"
-                            #
-                            # llama_index file loaders set these metadata, which would be useful to also support:
-                            # "creation_date": "2025-09-03",
-                            # "file_name": "file.pdf",
-                            # "file_path": "/absolute/path/to/the/file.pdf",
-                            # "file_size": 395154,
-                            # "file_type": "application\/pdf",
-                            # "last_modified_date": "2025-09-03",
-                            # "page_label": "1",
-                        },
-                        relationships={
-                            # when using the llama_index loaders, llama_index groups Nodes under Documents
-                            # and relationships point to the Document (which is also a Node), which confusingly
-                            # enough does not map to an actual file (for a PDF, a Document is a page of the PDF)
-                            # the Document structure is not something that is persisted, so it is fine here
-                            # if we have a relationship to a node_id that does not exist in the db
-                            NodeRelationship.SOURCE: RelatedNodeInfo(
-                                node_id=document_id,
-                                node_type="1",
-                                metadata={},
-                            ),
-                        },
+                        vector=embedding.vector,
                     )
                 )
 
@@ -330,10 +298,6 @@ async def search(self, query: VectorStoreQuery) -> List[SearchResult]:
                 return []
             raise
 
-    def compute_deterministic_chunk_id(self, document_id: str, chunk_idx: int) -> str:
-        # the id_ of the Node must be a UUID string, otherwise llama_index / LanceDB fails downstream
-        return str(string_to_uuid(f"{document_id}::{chunk_idx}"))
-
     async def count_records(self) -> int:
         try:
             table = self.lancedb_vector_store.table
@@ -346,15 +310,7 @@ async def count_records(self) -> int:
 
     @property
     def query_type(self) -> Literal["fts", "hybrid", "vector"]:
-        match self.vector_store_config.store_type:
-            case VectorStoreType.LANCE_DB_FTS:
-                return "fts"
-            case VectorStoreType.LANCE_DB_HYBRID:
-                return "hybrid"
-            case VectorStoreType.LANCE_DB_VECTOR:
-                return "vector"
-            case _:
-                raise_exhaustive_enum_error(self.vector_store_config.store_type)
+        return store_type_to_lancedb_query_type(self.vector_store_config.store_type)
 
     @staticmethod
     def lancedb_path_for_config(rag_config: RagConfig) -> str:
@@ -380,9 +336,7 @@ async def delete_nodes_not_in_set(self, document_ids: Set[str]) -> None:
                 kiln_doc_id = row["metadata"]["kiln_doc_id"]
                 if kiln_doc_id not in document_ids:
                     kiln_chunk_idx = row["metadata"]["kiln_chunk_idx"]
-                    record_id = self.compute_deterministic_chunk_id(
-                        kiln_doc_id, kiln_chunk_idx
-                    )
+                    record_id = deterministic_chunk_id(kiln_doc_id, kiln_chunk_idx)
                     rows_to_delete.append(record_id)
 
             if rows_to_delete: