pageindex 0.2.10.dev1__tar.gz → 0.2.10.dev2__tar.gz

{pageindex-0.2.10.dev1 → pageindex-0.2.10.dev2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pageindex
-Version: 0.2.10.dev1
+Version: 0.2.10.dev2
 Summary: Python SDK for PageIndex — reasoning-based, vectorless document retrieval, cloud and local
 License: MIT
 License-File: LICENSE

{pageindex-0.2.10.dev1 → pageindex-0.2.10.dev2}/pageindex/client.py

@@ -4,7 +4,7 @@ from __future__ import annotations
 import os
 import time
 import warnings
-from typing import Any, Callable, Iterator, Optional, Union
+from typing import Any, Callable, Iterator, Optional, Union, cast
 
 from .errors import PageIndexAPIError
 
@@ -345,7 +345,44 @@ class PageIndexClient:
             "favor of chat completions; use chat_completions instead."
         ).get_retrieval(retrieval_id=retrieval_id)
 
-    # ---------- CHAT COMPLETIONS ----------
+    # ---------- CHAT ----------
+
+    def chat(
+        self,
+        messages: Union[str, list[dict[str, str]]],
+        doc_id: Optional[Union[str, list[str]]] = None,
+        stream: bool = False,
+        model: Optional[str] = None,
+    ) -> Union[str, Iterator[str]]:
+        """
+        Ask a question about your documents, get the answer.
+
+        Thin sugar over ``chat_completions()`` in both modes — same
+        engine, same wire, minus the envelope. Multi-turn: keep your own
+        role/content list of the visible conversation (append each answer
+        as an assistant message) and pass it back. For usage accounting,
+        streaming metadata, or the tool-use process, use the protocol
+        surfaces: ``chat_completions()``, ``responses()``, ``messages()``.
+
+        Args:
+            messages: A question string, or role/content conversation
+                history.
+            doc_id: Document ID or list of IDs to scope the conversation.
+                Keep it identical across a conversation's calls.
+            stream: Yield the answer as text chunks as it is produced.
+            model: Local only — backend model name (defaults to
+                ``retrieve_model``).
+
+        Returns:
+            - stream=False: the answer string
+            - stream=True: iterator of text chunks
+        """
+        result = self.chat_completions(messages, stream=stream,
+                                       doc_id=doc_id, model=model)
+        if stream:
+            return cast(Iterator[str], result)
+        envelope = cast(dict[str, Any], result)
+        return envelope["choices"][0]["message"]["content"] or ""
 
     def chat_completions(
         self,
@@ -368,7 +405,9 @@ class PageIndexClient:
         backend, so any OpenAI-compatible server works; a ``/`` in the
         model name means LiteLLM provider routing, so prefix ``openai/``
         when the backend itself serves slashed ids, e.g.
-        ``openai/Qwen/...`` on vLLM). The non-stream
+        ``openai/Qwen/...`` on vLLM; LiteLLM-routed Claude models —
+        Anthropic direct, Bedrock, Vertex — get the managed prompt
+        prefix cache-marked automatically). The non-stream
         response carries the final answer only; streaming yields the
         agent's visible text as it is produced, including narration before
         tool calls. ``finish_reason`` reports loop completion ("stop") —
@@ -380,7 +419,10 @@ class PageIndexClient:
             messages: Conversation messages with 'role' and 'content' keys,
                 or a bare query string (it becomes a single user message).
                 Local also accepts system/developer messages — their content
-                is appended to the managed system prompt.
+                is appended to the managed system prompt. Local takes text
+                history only: tool-role turns are rejected (the cloud
+                endpoint forwards them verbatim), and message fields beyond
+                role/content are dropped.
             stream: Enable streaming responses.
             doc_id: Document ID or list of IDs to scope the conversation.
                 Keep it identical across a conversation's calls — the

{pageindex-0.2.10.dev1 → pageindex-0.2.10.dev2}/pageindex/local_chat.py

@@ -206,7 +206,8 @@ def _require_openai_agents(method: str) -> None:
     except ImportError as exc:
         raise PageIndexAPIError(
             f"{method} in local mode requires the OpenAI Agents SDK — "
-            "pip install openai-agents (or pip install 'pageindex[openai]')."
+            "pip install openai-agents (or pip install 'pageindex[openai]'). "
+            "messages() runs on the anthropic extra instead."
         ) from exc
 
 
@@ -271,6 +272,29 @@ def _reported_model(model_name: str) -> str:
     return model_name.removeprefix("litellm/").removeprefix("openai/")
 
 
+def _cache_extra_args(model_name: str) -> Optional[dict]:
+    """Claude's prompt caching is opt-in per request: on Claude models
+    routed through LiteLLM (Anthropic direct, Bedrock, Vertex — each
+    channel live-verified), mark the managed system prefix via LiteLLM's
+    injection param so the loop's later turns and a conversation's next
+    calls read it instead of repaying full price. Provider resolution is
+    LiteLLM's own, so this predicate can never disagree with where the
+    request actually routes."""
+    if "/" not in model_name or model_name.startswith("openai/"):
+        return None
+    try:
+        from litellm import get_llm_provider
+        model, provider, _, _ = get_llm_provider(
+            model=model_name.removeprefix("litellm/"))
+    except Exception:
+        return None
+    if provider == "anthropic" or (provider in ("bedrock", "vertex_ai")
+                                   and "claude" in model.lower()):
+        return {"cache_control_injection_points": [
+            {"location": "message", "role": "system"}]}
+    return None
+
+
 def _openai_agent(client, protocol: str, model_name: str, instructions: str,
                   temperature, top_p, doc_ids=None):
     from agents import Agent, ModelSettings
@@ -280,7 +304,8 @@ def _openai_agent(client, protocol: str, model_name: str, instructions: str,
         instructions=instructions,
         tools=build_openai_tools(client, doc_ids=doc_ids),
         model=_openai_model(protocol, model_name),
-        model_settings=ModelSettings(temperature=temperature, top_p=top_p),
+        model_settings=ModelSettings(temperature=temperature, top_p=top_p,
+                                     extra_args=_cache_extra_args(model_name)),
     )
 
 

{pageindex-0.2.10.dev1 → pageindex-0.2.10.dev2}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "pageindex"
-version = "0.2.10.dev1"
+version = "0.2.10.dev2"
 description = "Python SDK for PageIndex — reasoning-based, vectorless document retrieval, cloud and local"
 readme = "README.md"
 license = "MIT"