pixelrag-langchain 0.2.1__py3-none-any.whl

pixelrag_langchain/__init__.py

@@ -0,0 +1,32 @@
+"""LangChain integration for PixelRAG (screenshot-native visual search).
+
+Unofficial community package. Not affiliated with or endorsed by
+StarTrail-org, the maintainers of the upstream PixelRAG project
+(https://github.com/StarTrail-org/PixelRAG).
+"""
+
+from pixelrag_langchain.client import (
+    PixelRAGAPIError,
+    PixelRAGClient,
+    PixelRAGConnectionError,
+    PixelRAGError,
+    PixelRAGTile,
+    PixelRAGTimeoutError,
+)
+from pixelrag_langchain.config import PixelRAGConfig
+from pixelrag_langchain.retriever import PixelRAGRetriever
+from pixelrag_langchain.tool import PixelRAGSearchTool
+
+__all__ = [
+    "PixelRAGConfig",
+    "PixelRAGClient",
+    "PixelRAGTile",
+    "PixelRAGError",
+    "PixelRAGConnectionError",
+    "PixelRAGTimeoutError",
+    "PixelRAGAPIError",
+    "PixelRAGRetriever",
+    "PixelRAGSearchTool",
+]
+
+__version__ = "0.2.0"

pixelrag_langchain/client.py

@@ -0,0 +1,306 @@
+"""Thin HTTP client for a pixelrag-serve `/search` endpoint.
+
+Request shape follows the upstream README's documented example:
+
+    POST /search
+    {"queries": [{"text": "What is the capital of France?"}], "n_docs": 5}
+
+The authoritative schema is published by each server at `/openapi.json`.
+PixelRAG is still pre-1.0, so parsing remains deliberately defensive: additive
+fields are retained in `raw`, and malformed result containers degrade to an
+empty result rather than crashing an agent mid-run.
+"""
+
+from __future__ import annotations
+
+import logging
+from dataclasses import dataclass, field
+from typing import Any
+from urllib.parse import quote
+
+import httpx
+
+from pixelrag_langchain.config import PixelRAGConfig
+
+logger = logging.getLogger(__name__)
+
+
+class PixelRAGError(Exception):
+    """Base class for all errors raised by this client."""
+
+
+class PixelRAGConnectionError(PixelRAGError):
+    """Could not reach the pixelrag-serve instance at all."""
+
+
+class PixelRAGTimeoutError(PixelRAGError):
+    """Request exceeded the configured timeout."""
+
+
+class PixelRAGAPIError(PixelRAGError):
+    """Server reachable but returned a non-2xx response."""
+
+    def __init__(self, status_code: int, body: str):
+        self.status_code = status_code
+        self.body = body
+        super().__init__(f"pixelrag-serve returned HTTP {status_code}: {body[:500]}")
+
+
+@dataclass
+class PixelRAGTile:
+    """One retrieved screenshot tile, parsed from the API response.
+
+    Field mapping verified against a live api.pixelrag.ai response (July
+    2026), where each hit looks like:
+
+        {"score": 0.588, "vector_id": 18869670, "article_id": 5663927,
+         "tile_index": 0, "chunk_index": 7, "y_offset": 7168,
+         "tile_height": 1024,
+         "path": "shard_683/.../5663927.png.tiles/chunk_0000_07.png",
+         "url": "Outline_of_France",        # NB: article TITLE, not a URL
+         "article_pages": "0:0-7,...", "image_base64": null}
+
+    Quirks preserved here on purpose:
+      - The server's "url" field holds the article title -> mapped to
+        `source`, NOT to `image_url`.
+      - "path" is a relative tile path within the server's tile store, not a
+        directly fetchable URL -> exposed as `tile_path`.
+      - Image bytes come back only when the server includes "image_base64".
+    Anything unmapped stays available in `raw`.
+    """
+
+    score: float
+    tile_id: str | None = None
+    source: str | None = None          # article title (server's "url") or doc/article id
+    article_id: str | None = None
+    tile_index: int | None = None
+    chunk_index: int | None = None
+    tile_path: str | None = None       # server-relative and informational only
+    image_url: str | None = None       # only if server returns a real fetchable URL
+    image_base64: str | None = None    # inline image bytes, when provided
+    caption: str | None = None         # accompanying text/OCR, when provided
+    raw: dict[str, Any] = field(default_factory=dict)  # untouched original record
+
+    @property
+    def has_tile_coordinates(self) -> bool:
+        """Whether this hit can be addressed through the `/tile` endpoint."""
+        return (
+            self.article_id is not None
+            and self.tile_index is not None
+            and self.chunk_index is not None
+        )
+
+    @classmethod
+    def from_api_record(cls, record: dict[str, Any]) -> "PixelRAGTile":
+        score = record.get("score", record.get("similarity", 0.0))
+        try:
+            score = float(score)
+        except (TypeError, ValueError):
+            score = 0.0
+
+        article_id = record.get("article_id") or record.get("doc_id")
+        tile_id = record.get("tile_id") or record.get("vector_id") or record.get("id")
+
+        def as_int(value: Any) -> int | None:
+            try:
+                return int(value)
+            except (TypeError, ValueError):
+                return None
+
+        # "url" is the article title in the live schema. Only treat it as a
+        # fetchable image URL if it actually looks like one.
+        url_field = record.get("url")
+        looks_like_real_url = isinstance(url_field, str) and url_field.startswith(
+            ("http://", "https://")
+        )
+
+        return cls(
+            score=score,
+            tile_id=str(tile_id) if tile_id is not None else None,
+            source=record.get("source")
+            or (url_field if not looks_like_real_url else None)
+            or (str(article_id) if article_id is not None else None),
+            article_id=str(article_id) if article_id is not None else None,
+            tile_index=as_int(record.get("tile_index")),
+            chunk_index=as_int(record.get("chunk_index")),
+            tile_path=record.get("path"),
+            image_url=record.get("image_url")
+            or (url_field if looks_like_real_url else None),
+            image_base64=record.get("image_base64") or record.get("image"),
+            caption=record.get("caption") or record.get("text"),
+            raw=record,
+        )
+
+
+class PixelRAGClient:
+    """Synchronous client for a running pixelrag-serve instance.
+
+    Example:
+        >>> client = PixelRAGClient(PixelRAGConfig(base_url="http://localhost:30001"))
+        >>> tiles = client.search("capital of France", n_docs=3)
+    """
+
+    def __init__(self, config: PixelRAGConfig | None = None):
+        self.config = config or PixelRAGConfig()
+        headers = {}
+        if self.config.api_key:
+            headers["Authorization"] = f"Bearer {self.config.api_key}"
+        self._http = httpx.Client(timeout=self.config.timeout_s, headers=headers)
+
+    def search(
+        self,
+        query: str,
+        n_docs: int | None = None,
+        *,
+        include_images: bool = False,
+    ) -> list[PixelRAGTile]:
+        """Run one text query against the index and return ranked tiles.
+
+        Set `include_images=True` to request inline base64 images. The default
+        keeps search responses small; use `tile_url()` or `fetch_tile()` to
+        retrieve only the images needed downstream.
+
+        Raises:
+            PixelRAGConnectionError: server unreachable (wrong URL, not running).
+            PixelRAGTimeoutError: request exceeded `config.timeout_s`.
+            PixelRAGAPIError: server responded with an error status.
+        """
+        if not query or not query.strip():
+            return []
+
+        requested_n_docs = self.config.n_docs if n_docs is None else n_docs
+        if requested_n_docs < 1:
+            raise ValueError("n_docs must be at least 1")
+
+        payload: dict[str, Any] = {
+            "queries": [{"text": query}],
+            "n_docs": requested_n_docs,
+        }
+        if include_images:
+            payload["include_images"] = True
+
+        response = self._request("POST", self.config.search_url, json=payload)
+
+        try:
+            data = response.json()
+        except ValueError as exc:
+            raise PixelRAGAPIError(response.status_code, response.text) from exc
+
+        return self._parse_results(data)
+
+    def tile_url(self, tile: PixelRAGTile) -> str | None:
+        """Return the server URL for a tile, or `None` without coordinates."""
+        if not tile.has_tile_coordinates:
+            return None
+        return (
+            f"{self.config.base_url.rstrip('/')}/tile/"
+            f"{quote(tile.article_id, safe='')}/{tile.tile_index}/{tile.chunk_index}"
+        )
+
+    def fetch_tile(self, tile: PixelRAGTile) -> bytes:
+        """Download image bytes for one retrieved tile."""
+        url = self.tile_url(tile)
+        if url is None:
+            raise ValueError(
+                "Tile is missing article_id, tile_index, or chunk_index; "
+                "cannot construct its image URL."
+            )
+        return self._request("GET", url).content
+
+    def status(self) -> dict[str, Any]:
+        """Return server/index diagnostics from the public `/status` endpoint."""
+        response = self._request("GET", self.config.status_url)
+        try:
+            data = response.json()
+        except ValueError as exc:
+            raise PixelRAGAPIError(response.status_code, response.text) from exc
+        if not isinstance(data, dict):
+            raise PixelRAGAPIError(
+                response.status_code, "Expected a JSON object from /status"
+            )
+        return data
+
+    def health(self) -> bool:
+        """Best-effort check that the server is up. Never raises."""
+        try:
+            self._request("GET", self.config.health_url)
+            return True
+        except PixelRAGError:
+            return False
+
+    def _request(self, method: str, url: str, **kwargs: Any) -> httpx.Response:
+        """Send one request and normalize all transport/API failures."""
+        try:
+            response = self._http.request(method, url, **kwargs)
+        except httpx.TimeoutException as exc:
+            raise PixelRAGTimeoutError(
+                f"Request to {url} timed out after {self.config.timeout_s}s."
+            ) from exc
+        except httpx.RequestError as exc:
+            raise PixelRAGConnectionError(
+                f"Could not reach pixelrag-serve at {self.config.base_url}: {exc}"
+            ) from exc
+
+        if response.status_code >= 400:
+            raise PixelRAGAPIError(response.status_code, response.text)
+        return response
+
+    def close(self) -> None:
+        self._http.close()
+
+    def __enter__(self) -> "PixelRAGClient":
+        return self
+
+    def __exit__(self, *exc_info: object) -> None:
+        self.close()
+
+    @staticmethod
+    def _parse_results(data: Any) -> list[PixelRAGTile]:
+        """Handle a couple of plausible top-level shapes without raising.
+
+        Known/expected shapes we defend against:
+          {"results": [[{...}, {...}]]}          # per-query list of lists
+          {"results": [{...}, {...}]}             # flat list (single query)
+          {"data": {"results": [...]}}             # nested wrapper
+        Anything else logs a warning and returns an empty list rather than
+        raising, so a single malformed response doesn't take down an agent.
+        """
+        if not isinstance(data, dict):
+            logger.warning(
+                "Unrecognized pixelrag-serve response type: %s", type(data).__name__
+            )
+            return []
+
+        results = data.get("results")
+        if results is None and isinstance(data.get("data"), dict):
+            results = data["data"].get("results")
+
+        if results is None:
+            logger.warning(
+                "Unrecognized pixelrag-serve response shape (no 'results' key); "
+                "keys were: %s", list(data.keys()),
+            )
+            return []
+
+        if not isinstance(results, list):
+            logger.warning("pixelrag-serve 'results' field was not a list")
+            return []
+
+        # Real observed schema (api.pixelrag.ai, July 2026): results is a list
+        # of per-query objects, each wrapping its records in a "hits" list:
+        #   {"results": [{"hits": [{...}, {...}]}]}
+        if results and isinstance(results[0], dict) and "hits" in results[0]:
+            results = results[0].get("hits") or []
+        # Older/alternate shape: list-of-lists grouped by query.
+        elif results and isinstance(results[0], list):
+            results = results[0]
+
+        if not isinstance(results, list):
+            logger.warning("pixelrag-serve 'hits' field was not a list")
+            return []
+
+        tiles: list[PixelRAGTile] = []
+        for record in results:
+            if isinstance(record, dict):
+                tiles.append(PixelRAGTile.from_api_record(record))
+        return tiles

pixelrag_langchain/config.py

@@ -0,0 +1,67 @@
+"""Connection configuration for a pixelrag-serve-compatible search API.
+
+The constructor default points at a local `pixelrag serve` instance
+(`pip install 'pixelrag[serve]'`) rather than any third-party host — a
+library shouldn't silently route your queries somewhere you didn't choose.
+
+Upstream PixelRAG (github.com/StarTrail-org/PixelRAG) also runs an officially
+documented public production endpoint for its Wikipedia index, with no API key
+required. Use `PixelRAGConfig.hosted()` to point at it — that's one explicit
+line, not something this package assumes for you.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+
+DEFAULT_BASE_URL = "http://localhost:30001"
+# Documented in upstream's README as a first-class, intentional feature (not
+# an internal/staging URL) — see the "Live, hosted endpoint" callout at
+# https://github.com/StarTrail-org/PixelRAG#readme. It only serves their
+# pre-built Wikipedia index, not your own documents.
+PIXELRAG_HOSTED_BASE_URL = "https://api.pixelrag.ai"
+DEFAULT_TIMEOUT_S = 30.0
+DEFAULT_N_DOCS = 5
+
+
+@dataclass(frozen=True)
+class PixelRAGConfig:
+    """Settings for talking to a pixelrag-serve search API.
+
+    Attributes:
+        base_url: Root URL of a running pixelrag-serve instance. Defaults to
+            a local instance on the default port from the upstream README.
+            This is intentionally NOT defaulted to any third-party hosted
+            endpoint — point it there explicitly if you have one.
+        timeout_s: Per-request timeout in seconds.
+        n_docs: Default number of results to request per query.
+        api_key: Optional bearer token, for deployments that sit behind auth
+            (the local pixelrag-serve quickstart does not require one).
+    """
+
+    base_url: str = DEFAULT_BASE_URL
+    timeout_s: float = DEFAULT_TIMEOUT_S
+    n_docs: int = DEFAULT_N_DOCS
+    api_key: str | None = None
+
+    @classmethod
+    def hosted(cls, **overrides: object) -> "PixelRAGConfig":
+        """Point at upstream's public production endpoint (api.pixelrag.ai).
+
+        No API key needed. Searches their Wikipedia index only — not your own
+        documents. Self-host when you need private data or guaranteed capacity.
+        """
+        return cls(base_url=PIXELRAG_HOSTED_BASE_URL, **overrides)  # type: ignore[arg-type]
+
+    @property
+    def search_url(self) -> str:
+        return f"{self.base_url.rstrip('/')}/search"
+
+    @property
+    def health_url(self) -> str:
+        return f"{self.base_url.rstrip('/')}/health"
+
+    @property
+    def status_url(self) -> str:
+        return f"{self.base_url.rstrip('/')}/status"

pixelrag_langchain/retriever.py

@@ -0,0 +1,90 @@
+"""LangChain-compatible retriever backed by a pixelrag-serve instance.
+
+Note on Documents: LangChain's `Document` is a text-container. Screenshot
+tiles are images, so we don't lie about that — `page_content` holds whatever
+caption/OCR text the server gave us (often none), and the actual image
+reference (URL or base64) lives in `metadata`. If your downstream chain needs
+the pixels, read `metadata["image_url"]` (the fetchable coordinate URL) or
+`metadata["image_base64"]` and pass it to a vision-capable model.
+"""
+
+from __future__ import annotations
+
+from typing import Any
+
+from langchain_core.callbacks import CallbackManagerForRetrieverRun
+from langchain_core.documents import Document
+from langchain_core.retrievers import BaseRetriever
+
+from pixelrag_langchain.client import PixelRAGClient, PixelRAGError
+from pixelrag_langchain.config import PixelRAGConfig
+
+
+class PixelRAGRetriever(BaseRetriever):
+    """Retrieve screenshot tiles from a pixelrag-serve index as Documents.
+
+    Example:
+        >>> retriever = PixelRAGRetriever(config=PixelRAGConfig(base_url="http://localhost:30001"))
+        >>> docs = retriever.invoke("What is the capital of France?")
+    """
+
+    config: PixelRAGConfig = PixelRAGConfig()
+    n_docs: int | None = None
+    include_images: bool = False
+    """Request base64 images inline. Off by default to keep searches small."""
+    raise_on_error: bool = False
+    """If False (default), connection/timeout errors return [] instead of
+    raising, so a flaky visual index doesn't crash the whole chain. Set True
+    if you'd rather fail loudly and handle it yourself."""
+
+    _client: PixelRAGClient | None = None
+
+    model_config = {"arbitrary_types_allowed": True}
+
+    def _get_client(self) -> PixelRAGClient:
+        if self._client is None:
+            self._client = PixelRAGClient(self.config)
+        return self._client
+
+    def _get_relevant_documents(
+        self, query: str, *, run_manager: CallbackManagerForRetrieverRun
+    ) -> list[Document]:
+        client = self._get_client()
+        try:
+            tiles = client.search(
+                query,
+                n_docs=self.n_docs,
+                include_images=self.include_images,
+            )
+        except PixelRAGError as exc:
+            run_manager.on_retriever_error(exc)
+            if self.raise_on_error:
+                raise
+            return []
+
+        documents: list[Document] = []
+        for tile in tiles:
+            tile_url = client.tile_url(tile)
+            metadata: dict[str, Any] = {
+                "score": tile.score,
+                "tile_id": tile.tile_id,
+                "source": tile.source,
+                "article_id": tile.article_id,
+                "tile_index": tile.tile_index,
+                "chunk_index": tile.chunk_index,
+                "tile_path": tile.tile_path,
+                "tile_url": tile_url,
+                "image_url": tile.image_url or tile_url,
+                "image_base64": tile.image_base64,
+                "modality": "image",
+            }
+            documents.append(
+                Document(page_content=tile.caption or "", metadata=metadata)
+            )
+        return documents
+
+    def close(self) -> None:
+        """Release the lazily-created HTTP client, if any."""
+        if self._client is not None:
+            self._client.close()
+            self._client = None

pixelrag_langchain/tool.py

@@ -0,0 +1,119 @@
+"""LangChain Tool wrapping PixelRAG visual search for use in agents.
+
+Two entry points, because agents have two different needs:
+
+  - `PixelRAGSearchTool`  — a standard `BaseTool`. `_run()` returns a plain
+    text summary (source, score, any caption) so it drops straight into a
+    normal text-only tool-calling agent loop.
+
+  - `PixelRAGSearchTool.search_tiles()` — returns the raw `PixelRAGTile`
+    objects, image data included. Use this in a LangGraph node (or anywhere
+    you control message construction) when you want to hand the retrieved
+    screenshots to a vision-capable model as actual image content, which is
+    the point of PixelRAG — reading pixels, not just a text summary of them.
+"""
+
+from __future__ import annotations
+
+from typing import Type
+
+from langchain_core.callbacks import CallbackManagerForToolRun
+from langchain_core.tools import BaseTool
+from pydantic import BaseModel, Field
+
+from pixelrag_langchain.client import PixelRAGClient, PixelRAGError, PixelRAGTile
+from pixelrag_langchain.config import PixelRAGConfig
+
+
+class PixelRAGSearchInput(BaseModel):
+    query: str = Field(description="Natural-language search query.")
+    n_docs: int = Field(default=5, ge=1, le=50, description="Number of tiles to retrieve.")
+
+
+class PixelRAGSearchTool(BaseTool):
+    """Search a PixelRAG visual index and return the top matching tiles.
+
+    Requires a running pixelrag-serve instance (self-hosted by default —
+    see the upstream README: https://github.com/StarTrail-org/PixelRAG).
+    Point `config.base_url` at your own server; nothing is assumed for you.
+    """
+
+    name: str = "pixelrag_visual_search"
+    description: str = (
+        "Search an index of document screenshots (web pages, PDFs, images) "
+        "and return the most visually/semantically relevant tiles, with "
+        "similarity scores and source references. Prefer this over a plain "
+        "text search when the answer likely lives in a table, chart, "
+        "infographic, or other layout that plain text extraction would "
+        "mangle."
+    )
+    args_schema: Type[BaseModel] = PixelRAGSearchInput
+
+    config: PixelRAGConfig = PixelRAGConfig()
+    _client: PixelRAGClient | None = None
+
+    model_config = {"arbitrary_types_allowed": True}
+
+    def _get_client(self) -> PixelRAGClient:
+        if self._client is None:
+            self._client = PixelRAGClient(self.config)
+        return self._client
+
+    def search_tiles(
+        self,
+        query: str,
+        n_docs: int = 5,
+        *,
+        include_images: bool = False,
+    ) -> list[PixelRAGTile]:
+        """Return structured hits for use in LangGraph or custom pipelines."""
+        return self._get_client().search(
+            query, n_docs=n_docs, include_images=include_images
+        )
+
+    def tile_image_url(self, tile: PixelRAGTile) -> str | None:
+        """Return the fetchable coordinate-based image URL for a tile."""
+        return self._get_client().tile_url(tile)
+
+    def fetch_tile_bytes(self, tile: PixelRAGTile) -> bytes:
+        """Download image bytes for a retrieved tile."""
+        return self._get_client().fetch_tile(tile)
+
+    def _run(
+        self,
+        query: str,
+        n_docs: int = 5,
+        *,
+        run_manager: CallbackManagerForToolRun | None = None,
+    ) -> str:
+        try:
+            tiles = self.search_tiles(query, n_docs=n_docs)
+        except PixelRAGError as exc:
+            return f"pixelrag_visual_search failed: {exc}"
+
+        if not tiles:
+            return "No results found."
+
+        lines = [f"Top {len(tiles)} visual matches for: {query!r}"]
+        for i, tile in enumerate(tiles, start=1):
+            bits = [f"{i}. score={tile.score:.3f}"]
+            if tile.source:
+                bits.append(f"source={tile.source}")
+            if tile.tile_id:
+                bits.append(f"tile_id={tile.tile_id}")
+            if tile.caption:
+                bits.append(f'caption="{tile.caption[:200]}"')
+            image_url = tile.image_url or self.tile_image_url(tile)
+            if image_url:
+                bits.append(f"image={image_url}")
+            elif tile.image_base64:
+                bits.append("(inline image attached)")
+            lines.append("  " + " | ".join(bits))
+
+        return "\n".join(lines)
+
+    def close(self) -> None:
+        """Release the lazily-created HTTP client, if any."""
+        if self._client is not None:
+            self._client.close()
+            self._client = None

pixelrag_langchain-0.2.1.dist-info/METADATA

@@ -0,0 +1,181 @@
+Metadata-Version: 2.5
+Name: pixelrag-langchain
+Version: 0.2.1
+Summary: LangChain Tool + Retriever for PixelRAG (screenshot-native visual search) — unofficial community integration
+Project-URL: Homepage, https://github.com/navneet-singh2907/pixelrag-langchain
+Project-URL: Repository, https://github.com/navneet-singh2907/pixelrag-langchain
+Project-URL: Upstream PixelRAG, https://github.com/StarTrail-org/PixelRAG
+Project-URL: PixelRAG paper, https://papers.cool/arxiv/2606.28344
+Author: Navneet Singh
+License: Apache-2.0
+License-File: LICENSE
+Keywords: agents,langchain,langgraph,pixelrag,rag,visual-search,vlm
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Software Development :: Libraries :: Python Modules
+Requires-Python: >=3.10
+Requires-Dist: httpx<0.28.0,>=0.27.0
+Requires-Dist: langchain-core<0.4.0,>=0.3.0
+Requires-Dist: pydantic<3.0.0,>=2.7.0
+Provides-Extra: dev
+Requires-Dist: pytest-httpx<0.31.0,>=0.30.0; extra == 'dev'
+Requires-Dist: pytest<9.0.0,>=8.0.0; extra == 'dev'
+Requires-Dist: respx<0.22.0,>=0.21.0; extra == 'dev'
+Provides-Extra: render
+Requires-Dist: playwright<2.0.0,>=1.45.0; extra == 'render'
+Description-Content-Type: text/markdown
+
+[![DeepSource](https://app.deepsource.com/gh/navneet-singh2907/pixelrag-langchain.svg/?label=code+coverage&show_trend=true&token=IxW1QBGxZ_fS27YFYgJ81_u6)](https://app.deepsource.com/gh/navneet-singh2907/pixelrag-langchain/)
+
+[![CI](https://github.com/navneet-singh2907/pixelrag-langchain/actions/workflows/ci.yml/badge.svg)](https://github.com/navneet-singh2907/pixelrag-langchain/actions/workflows/ci.yml)
+
+# pixelrag-langchain
+
+A LangChain `Tool` + `Retriever` for [PixelRAG](https://github.com/StarTrail-org/PixelRAG) — visual, screenshot-native search for AI agents.
+
+**This is an unofficial, community-built integration.** It is not affiliated with, endorsed by, or maintained by StarTrail-org or the PixelRAG paper authors. All credit for the underlying research and the `pixelrag-serve` engine belongs to them — see [Credits](#credits).
+
+## Why visual search
+
+Most agent pipelines fetch a page, strip the HTML down to text, chunk it, and embed the chunks. That throws away tables, charts, and layout — information that's often exactly what answers the question. [PixelRAG](https://arxiv.org/abs/2606.28344) (Wang, Li, Wang, Teiletche, Jin, Zaharia, Gonzalez, Min — UC Berkeley, Princeton, EPFL, Databricks) skips the text step entirely: it renders documents to screenshot tiles, retrieves over the images directly, and hands the retrieved tiles to a vision-language model to read.
+
+This package doesn't reimplement any of that. It just wraps the existing `pixelrag-serve` search API so it drops into a LangChain (or LangGraph) agent as a normal tool/retriever.
+
+## Architecture
+
+```mermaid
+flowchart LR
+    A["Page / PDF"] --> B["pixelrag-render\n(screenshot tiles)"]
+    B --> C["pixelrag-embed\n(Qwen3-VL-Embedding)"]
+    C --> D["FAISS index\n(pixelrag-serve)"]
+    D -- "this package talks to D" --> E["PixelRAGRetriever /\nPixelRAGSearchTool"]
+    E --> F["Your LangChain / LangGraph agent"]
+    F --> G["VLM reader\n(your model of choice)"]
+```
+
+Everything left of the dotted line (`render` → `embed` → `index` → `serve`) is upstream PixelRAG, run by you. This package is the box on the right: a client + LangChain adapters.
+
+## Install
+
+```bash
+pip install pixelrag-langchain
+```
+
+This installs only the LangChain adapter (`httpx`, `langchain-core`, `pydantic`) — no GPU, no `torch`, no CUDA. It talks to a `pixelrag-serve` instance over HTTP; it doesn't run embedding or indexing itself.
+
+## Quickstart — zero setup
+
+Upstream PixelRAG runs an official public production endpoint — no API key or index download required. Nothing to install beyond this package:
+
+```python
+from pixelrag_langchain import PixelRAGRetriever, PixelRAGConfig
+
+retriever = PixelRAGRetriever(config=PixelRAGConfig.hosted())
+docs = retriever.invoke("What is the capital of France?")
+
+for doc in docs:
+    print(doc.metadata["score"], doc.metadata["source"])
+    print(doc.metadata["image_url"])  # fetchable PNG URL
+```
+
+`PixelRAGConfig.hosted()` points at `https://api.pixelrag.ai`, upstream's documented production endpoint. It uses their infrastructure and Wikipedia index, not your documents. PixelRAG does not publish throughput, latency, or rate-limit guarantees; self-host when you need guaranteed capacity or private data.
+
+## Retrieving the actual tile images
+
+Search results identify each screenshot with `article_id`, `tile_index`, and `chunk_index`. Version 0.2 turns those coordinates into the upstream-supported `/tile/...` URL; the server-relative `path` field is informational and should not be used to construct URLs.
+
+Use the client when you want explicit control over image downloads:
+
+```python
+from pixelrag_langchain import PixelRAGClient, PixelRAGConfig
+
+with PixelRAGClient(PixelRAGConfig.hosted()) as client:
+    tile = client.search("diagram of a transformer architecture", n_docs=1)[0]
+    print(client.tile_url(tile))
+    png_bytes = client.fetch_tile(tile)
+```
+
+The separate `GET /tile/{article_id}/{tile_index}/{chunk_index}` request is the recommended default because it keeps search responses small. If you need every image inline, explicitly request base64 data:
+
+```python
+with PixelRAGClient(PixelRAGConfig.hosted()) as client:
+    tiles = client.search("quarterly revenue chart", include_images=True)
+    print(tiles[0].image_base64)
+```
+
+`PixelRAGRetriever` exposes the coordinate URL in both `metadata["tile_url"]` and `metadata["image_url"]`. `PixelRAGSearchTool` provides `tile_image_url()` and `fetch_tile_bytes()` for structured LangGraph workflows. See [`examples/langgraph_node.py`](examples/langgraph_node.py) for a complete multimodal-message handoff.
+
+The API is currently pre-1.0. Its authoritative machine-readable contract is served at [`/openapi.json`](https://api.pixelrag.ai/openapi.json); this package parses additive hit fields leniently.
+
+For operational checks, `client.health()` returns a best-effort boolean and `client.status()` returns the server's current index diagnostics.
+
+## Quickstart — your own documents
+
+For real use — your own PDFs, internal docs, scraped pages — run `pixelrag serve` yourself:
+
+```bash
+pip install 'pixelrag[serve]'
+
+# Download a pre-built Wikipedia index, or build your own from `pip install 'pixelrag[index]'`
+huggingface-cli download StarTrail-org/pixelrag-faiss-indexes \
+  --repo-type dataset --include "search_index_normed_v2/*" --local-dir ./index
+
+pixelrag serve --index-dir ./index/search_index_normed_v2 --port 30001
+```
+
+```python
+from pixelrag_langchain import PixelRAGRetriever, PixelRAGConfig
+
+retriever = PixelRAGRetriever(config=PixelRAGConfig(base_url="http://localhost:30001"))
+docs = retriever.invoke("What is the capital of France?")
+```
+
+## As an agent tool
+
+```python
+from langchain.agents import create_agent
+from pixelrag_langchain import PixelRAGSearchTool, PixelRAGConfig
+
+tool = PixelRAGSearchTool(config=PixelRAGConfig.hosted())  # or base_url="http://localhost:30001"
+agent = create_agent(model="your-model", tools=[tool])
+```
+
+Nothing here defaults to a single host silently — `PixelRAGConfig(base_url=...)` accepts any pixelrag-serve-compatible URL, including your own deployment; `.hosted()` is one explicit opt-in line, not a hidden default.
+
+## What this package is *not*
+
+- Not a reimplementation of PixelRAG's render/embed/index/serve pipeline — install the [upstream project](https://github.com/StarTrail-org/PixelRAG) for that.
+- Not a hosted service itself. `PixelRAGConfig.hosted()` is a convenience pointer to *upstream's* public endpoint, not infrastructure we run.
+- Not affiliated with StarTrail-org, Berkeley Sky Computing Lab, BAIR, or the Berkeley NLP Group, who built the actual PixelRAG engine and research this wraps.
+
+## Benchmarks (from the upstream paper, not measured by this package)
+
+Reported in [PIXELRAG: Web Screenshots Beat Text for Retrieval-Augmented Generation](https://arxiv.org/abs/2606.28344) (arXiv:2606.28344):
+
+| Metric | Text-based RAG | PixelRAG | Context |
+|---|---|---|---|
+| Prompt tokens, agentic benchmark | 37.5M | 3.6M | MoNaCo multi-step agent benchmark |
+| Accuracy gain | baseline | up to +18.1% | vs. text-based RAG baselines, across NQ/SimpleQA/MMSearch/LiveVQA |
+| Token cost via image compression | — | up to 3x reduction | lower-resolution tiles, accuracy preserved |
+
+These are the paper's own reported numbers, not something we independently reproduced — treat them as a starting point for your own evaluation on your workload, not a guarantee.
+
+## Development
+
+```bash
+git clone https://github.com/navneet-singh2907/pixelrag-langchain
+cd pixelrag-langchain
+pip install -e ".[dev]"
+pytest
+```
+
+## Credits
+
+- **PixelRAG** engine, paper, and research: Yichuan Wang, Zhifei Li, Zirui Wang, Paul Teiletche, Lesheng Jin, Matei Zaharia, Joseph E. Gonzalez, Sewon Min. [Paper](https://arxiv.org/abs/2606.28344) · [Code](https://github.com/StarTrail-org/PixelRAG) (Apache-2.0).
+- This package: an independent, unofficial LangChain adapter around that work.
+
+## License
+
+Apache-2.0 — see [LICENSE](LICENSE). This is a new work built to talk to PixelRAG over HTTP; it does not vendor or redistribute any upstream PixelRAG code, so no NOTICE carryover is required, but attribution is given above regardless because it's their research this is built on.

pixelrag_langchain-0.2.1.dist-info/RECORD

@@ -0,0 +1,9 @@
+pixelrag_langchain/__init__.py,sha256=Gm4aBTxc-coRQhOCC6GR9QTZcCwsnQ33Clthg7NntyM,853
+pixelrag_langchain/client.py,sha256=j2kLeHoFdPwwYgEQxyCLFFw3vpIcpRc_WNDpQYFU-rs,11608
+pixelrag_langchain/config.py,sha256=TlUzEsegzmaTFHRWNmbrv7cH4egtUDhUplXz-RRkH9U,2631
+pixelrag_langchain/retriever.py,sha256=D53T4pdG18g3uelohwR3vGjMciTAwsFA-etpmZLK4Ns,3369
+pixelrag_langchain/tool.py,sha256=XkiF9tqCnM00EEOvXWLkytc0odhWhGZo44ymxEG68Lc,4467
+pixelrag_langchain-0.2.1.dist-info/METADATA,sha256=kWxGT7LCqTGC5NYzTanMj74en-iI80k0EMnsGsG-l9Y,9584
+pixelrag_langchain-0.2.1.dist-info/WHEEL,sha256=zOwg4jB6zX2kU910N-cMawjivD6tO8NEWvE12je1bVk,87
+pixelrag_langchain-0.2.1.dist-info/licenses/LICENSE,sha256=z8d0m5b2O9McPEK1xHG_dWgUBT6EfBDz6wA0F7xSPTA,11358
+pixelrag_langchain-0.2.1.dist-info/RECORD,,

pixelrag_langchain-0.2.1.dist-info/WHEEL

@@ -0,0 +1,4 @@
+Wheel-Version: 1.0
+Generator: hatchling 1.32.0
+Root-Is-Purelib: true
+Tag: py3-none-any

pixelrag_langchain-0.2.1.dist-info/licenses/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.