Skip to content
CAI
Software that uses CAICheck a score

assafelovic/gpt-researcher

44.9

Weak · 16 November 2026

36.2k

lines of production code

Python

with TypeScript, JavaScript

5

measurements over time

CAI band scale
CAI trend line
CAI lens gauges

What this system is

Features

Add Bing Search Retriever

Introduces a new Bing Search retriever component that queries the Bing Web Search API (v7.0) using the BING\_API\_KEY environment variable. The implementation includes robust error handling for network failures and JSON parsing issues, normalizes API responses into a standard list of result dictionaries (title, href, body), and filters out YouTube links and malformed entries to prevent crashes during research.

_gpt\_researcher/retrievers/arxiv, gpt\researcher/retrievers/bing · high confidence

Add BoCha Search Retriever with resilient error handling

A new BoChaSearch retriever has been added to the system, enabling users to fetch search results via the BoCha AI API. The implementation includes robust error handling to gracefully manage network issues, API failures, and malformed or missing data in provider payloads, ensuring that such errors result in an empty response rather than crashing the research process.

_gpt\researcher/retrievers/bocha · high confidence

Add Brave Search as a new retrieval source

Users can now utilize Brave Search for retrieving information by setting the BRAVE\_API\_KEY environment variable. This new retriever integrates with the Brave Search API to fetch web results, normalizing them into the standard title, href, and body format used by other search providers in the system.

_gpt\researcher/retrievers/brave · high confidence

Add FireCrawl scraper with concurrency controls and SDK v4.6+ compatibility

A new FireCrawl scraper implementation has been added to the gpt\_researcher project. This update introduces a dedicated module that integrates with the FireCrawl Python SDK, specifically adapting to API changes in version 4.6.0 and later (e.g., using \scrape()\ instead of \scrape\_url()\ and accessing metadata via attributes rather than dictionary keys). To prevent rate-limit errors during deep research, the scraper includes a configurable concurrency limit (defaulting to 2 concurrent calls, adjustable via the \FIRECRAWL\_CONCURRENCY\ environment variable) using an async semaphore. It also safely handles optional HTTP sessions for image extraction and guards against null responses.

_gpt\researcher/scraper/firecrawl · high confidence

A new retriever has been added that allows GPT Researcher to search for tweets via the GetXAPI service. Users must set the GETXAPI\_API\_KEY environment variable to enable this feature. The retriever queries the GetXAPI advanced search endpoint, normalizes various field names from the API response (such as author usernames and engagement metrics), and returns results in the standard title, href, and body format used by other retrievers.

_gpt\researcher/retrievers/getxapi · high confidence

Add GroundRoute multi-engine search retriever

A new GroundRoute retriever is available, which routes queries across multiple web-search engines (Serper, Brave, Exa, Tavily, Firecrawl, Perplexity) to select the cheapest option that meets a quality bar, with caching for repeated queries. It requires the GROUNDROUTE\_API\_KEY environment variable or the groundroute\_api\_key header.

_gpt\researcher/retrievers/groundroute · high confidence

Add OpenAlex scholarly works retriever

Users can now retrieve information from the OpenAlex open catalog of scholarly works. This new retriever supports searching by query with configurable sorting (relevance, citation count, or publication date) and handles optional authentication via environment variables for improved rate limits. It robustly parses API responses, reconstructing abstracts from inverted indices and selecting the best available link (PDF, landing page, or work ID) for each result.

_gpt\researcher/retrievers/openalex · high confidence

Add SearchApi retriever with domain filtering support

A new SearchApi retriever has been added to the gpt\_researcher system, allowing users to perform internet searches via the SearchApi service. This implementation supports the \query\_domains\ parameter, enabling users to restrict search results to specific domains by appending site-specific filters to the query. The retriever handles API key configuration via the \SEARCHAPI\_API\_KEY\ environment variable and includes robust error handling for network failures and malformed responses.

_gpt\researcher/retrievers/searchapi · high confidence

Add SearxNG API Retriever with snippet length guard

A new SearxSearch retriever has been added to the gpt\_researcher package, allowing users to fetch search results via a SearxNG instance configured via the SEARX\_URL environment variable. The implementation explicitly marks results as requiring scraping (requires\_scraping = True) and enforces a 100-character limit on the prefetched snippet body. This cap prevents the system from incorrectly treating SearxNG's search snippets as full article text, ensuring that actual page content is fetched and cited correctly.

_gpt\researcher/retrievers/searx · high confidence

Add Semantic Scholar retriever with API response validation

Introduces a new Semantic Scholar search retriever that queries the Semantic Scholar API for academic papers. The implementation includes validation for sort criteria (relevance, citationCount, publicationDate) to ensure correct camelCase formatting expected by the API, and adds robust error handling and type checking for API responses, including guards for open access PDF structures to prevent crashes on malformed data.

_gpt\_researcher/retrievers/semantic\scholar · high confidence

Add SerpApi search retriever with query domain filtering

Users can now use SerpApi as a search provider via the new SerpApiSearch class. The retriever supports the query\_domains parameter to restrict results to specific sites (e.g., site:example.com) and includes robust handling for malformed or missing API response fields, ensuring that missing organic results or invalid entries do not cause errors.

_gpt\researcher/retrievers/serpapi · high confidence

Add Xquik X/Twitter search retriever with robust null-handling

A new Xquik retriever is available for searching X (Twitter) to provide real-time perspectives and breaking news. It requires the XQUIK\_API\_KEY environment variable and returns results in the standard title, href, and body format. The implementation includes defensive coding to handle malformed API payloads, specifically guarding against null or non-list values for tweets, authors, and text fields to prevent silent failures or crashes.

_gpt\researcher/retrievers/xquik · high confidence

Add fastCRW search retriever

Users can now use the fastCRW API as a search retriever by setting the CRW\_API\_KEY environment variable (or passing it via headers). The retriever supports self-hosted deployments via the CRW\_API\_URL environment variable, defaults to the managed cloud, and gracefully skips individual search results that lack a URL instead of failing the entire request.

_gpt\researcher/retrievers/crw · high confidence

Added Next.js API action handlers and Nginx configuration

The frontend now includes server-side action handlers in \frontend/nextjs/actions/apiActions.ts\ that communicate with backend endpoints (\/api/getSources\, \/api/getAnswer\, \/api/getSimilarQuestions\, \/api/generateLanggraph\) to support question answering and streaming responses. Additionally, an Nginx configuration file (\frontend/nextjs/nginx/default.conf\) has been added to serve the Next.js application on port 3000.

frontend/nextjs/actions, frontend/nextjs/nginx · high confidence

Added TypeScript type definitions for data models and analytics

This change introduces new TypeScript declaration files to the frontend Next.js application to provide type safety for core data structures and third-party libraries. The new \data.ts\ file defines interfaces for various content types (such as basic text, Langgraph buttons, differences, questions, and chat), configuration settings for chat boxes and MCP (Model Context Protocol) support, and structures for chat messages and research history. Additionally, \react-ga4.d.ts\ adds type definitions for the Google Analytics 4 library, specifically typing the initialization and event tracking methods to ensure correct usage of the analytics integration.

frontend/nextjs/types · high confidence

Added memory state definitions for draft and research workflows

New Python modules have been introduced in the backend memory package to define the data structures for the drafting and research phases. The \draft.py\ module establishes a \DraftState\ TypedDict containing fields for the task, topic, draft content, review status, and revision notes. Similarly, \research.py\ defines a \ResearchState\ TypedDict that structures research-related data, including initial findings, sections, research data, and the final report layout components such as title, headers, table of contents, and sources.

backend/memory · high confidence

Backend containerization and new report export formats

The backend is now containerized with a Dockerfile and Procfile, specifying Python 3.12 as the runtime and exposing port 8000. Additionally, the backend now supports exporting reports as DOCX files (in addition to existing Markdown and PDF formats) by integrating the python-docx and htmldocx libraries, and fixes a bug where empty filenames previously resulted in invalid PDF output paths.

backend · high confidence

GPT Researcher agent restructured with new capabilities and configuration options

The GPT Researcher module has been reorganized into a new package structure (gpt\_researcher) with a refactored main agent class. This update introduces several new capabilities: deep research support via a dedicated skill, integration with Model Context Protocol (MCP) servers for extended tool use, and image generation for reports. Users can now configure search scope with query domains, complement source URLs, and select from different prompt families. The agent also tracks per-step costs, supports UTF-8 encoding for output files, and includes a structured logging system.

_gpt\researcher · high confidence

Initial deployment infrastructure for gpt-researcher service

This change introduces the complete Terraform configuration to deploy the 'gpt-researcher' application on AWS. It establishes an immutable ECR repository for container images, an ECS service running on the 'general-cluster' with CloudMap service discovery, and the necessary IAM roles for GitHub Actions OIDC authentication. The setup includes remote state management via S3, specific security groups for EFS NFS access, and integration with AWS Secrets Manager for application credentials.

terraform · high confidence

Initial project scaffolding and configuration

The repository has been initialized with essential configuration files and documentation. This includes a Python version file (3.12.1), environment variable examples (.env.example) for API keys and scraper settings, and standard ignore files (.gitignore, .dockerignore, .cursorignore). It also introduces Cursor AI configuration (.cursorrules) to guide development, an MCP server configuration (.mcp.json) for tool integration, and standard community files like CODE\_OF\_CONDUCT.md and CONTRIBUTING.md. Additionally, Dockerfiles are provided to containerize the application.

(repo-wide) · high confidence

Introduce CustomRetriever for external API integration

Users can now connect to a custom external search API by setting the RETRIEVER\_ENDPOINT environment variable, with additional configuration passed via RETRIEVER\ARG\ prefixed variables. The new CustomRetriever class handles the HTTP request, enforces a strict list-based response contract, and robustly filters malformed payloads to prevent pipeline crashes, while signaling that results are returned with raw content and do not require further scraping.

_gpt\researcher/retrievers/custom · high confidence

Introduce GPT-Researcher evaluation framework with SimpleQA, quality, and hallucination assessments

Added a new \evals\ directory containing three complementary evaluation systems for assessing GPT-Researcher performance. The \simple\_evals\ module implements factual accuracy testing using the OpenAI SimpleQA dataset and methodology, tracking metrics like accuracy, F1 score, cost, and latency with structured JSON logging. The \quality\_eval\ module assesses report structure and content quality (citation faithfulness, source diversity, authority, subtopic coverage) without requiring ground-truth answers. The \hallucination\_eval\ module detects fabricated content by comparing reports against scraped sources. The framework includes configuration options for grading models and researcher behavior, with both human-readable console output and machine-readable JSON logs for trend analysis.

evals · high confidence

Introduce LangGraph-based multi-agent research workflow

Adds a new multi-agent system under the \multi\_agents\ directory that leverages LangGraph to coordinate a team of specialized agents (Chief Editor, Researcher, Editor, Reviewer, Revisor, Writer, Publisher) for in-depth research tasks. This feature allows users to generate comprehensive research reports (PDF, Docx, Markdown) by automating planning, data collection, review, and writing stages. The implementation includes a \main.py\ entry point, a \task.json\ configuration file for defining queries and output formats, and optional observability via Monocle tracing. It also supports human-in-the-loop feedback and integrates with the existing GPT-Researcher configuration for model selection.

_multi\agents · high confidence

Introduce MCP-based research retriever with two-stage tool selection

Users can now leverage Model Context Protocol (MCP) servers for research via a new \MCPRetriever\. This component implements a two-stage intelligent approach: first, it selects the most relevant tools from available MCP servers, and second, it executes research using only those selected tools. The implementation is modular, relying on dedicated classes for client management, tool selection, research execution, and streaming, and gracefully handles cases where the \langchain-mcp-adapters\ library is not installed.

_gpt\researcher/retrievers/mcp · high confidence

Introduce Model Context Protocol (MCP) integration for external tool usage

GPT Researcher now supports connecting to external tools and data sources via the Model Context Protocol (MCP). This new module allows users to configure MCP servers (stdio, WebSocket, or HTTP) to extend research capabilities with local databases, file systems, and third-party APIs. The integration includes an intelligent LLM-based tool selector to choose the most relevant tools for a query, a research skill to execute tool calls and process results, and streaming utilities for real-time progress updates.

_gpt\researcher/mcp · high confidence

Introduce PubMed Central retriever for full-text scientific article search

Adds a new PubMed Central retriever that queries the NCBI e-utilities API to find and retrieve full-text content (title, abstract, and body) from PubMed Central articles. The implementation handles API key configuration via environment variables, supports optional custom parameters, and ensures robust error handling by returning empty lists instead of None to prevent downstream type errors.

_gpt\_researcher/retrievers/pubmed\central · high confidence

Introduce VectorStoreWrapper for Langchain integration

Added a new VectorStoreWrapper component in the gpt\_researcher/vector\_store module that bridges GPT-Researcher's document format with Langchain's vector store capabilities. This wrapper handles the conversion of internal document dictionaries into Langchain Document objects, applies text splitting via RecursiveCharacterTextSplitter (1000 char chunks, 200 overlap), and provides an asynchronous similarity search interface. It includes robust input validation to skip non-dict entries and missing content, ensuring stable document loading.

_gpt\_researcher/vector\store · high confidence

Introduce draft and research state schemas for multi-agent workflow

New state definitions (DraftState and ResearchState) have been added to the memory module to structure data for the multi-agent planning and reporting process. DraftState tracks individual section drafts, review feedback, and revision counts, while ResearchState manages the broader report structure, including sections, research data, human feedback, layout details, and fact-checking notes.

_multi\agents/memory · high confidence

Introduce generic LLM provider abstraction

A new \GenericLLMProvider\ class has been added to the \gpt\_researcher/llm\_provider/generic\ module, providing a unified interface for interacting with a wide variety of Large Language Model providers. This change introduces support for numerous providers including OpenAI, Anthropic, Azure OpenAI, Cohere, Google Vertex AI, Google GenAI, Fireworks, Ollama, Together, Mistral, Hugging Face, Groq, AWS Bedrock, DashScope, xAI, DeepSeek, LiteLLM, GigaChat, OpenRouter, vLLM, Aimlapi, Netmind, Forge, Avian, MiniMax, Atlas Cloud, and Nebius. The implementation handles provider-specific initialization, supports custom base URLs for OpenAI-compatible endpoints, manages temperature settings for models that do not support them (such as O1/O3/O4 series and Claude 4.x), and includes features like chat logging and usage metadata capture for cost tracking.

_gpt\_researcher/llm\provider/generic · high confidence

Introduce optional web search in chat via Tavily

The backend/chat module now supports a new ChatAgentWithMemory class that enables web search capabilities within the chat interface. This feature relies on the optional Tavily client; if the TAVILY\_API\_KEY environment variable is set and the package is installed, the agent can perform live searches using a 'quick\_search' tool. If the key is missing or the package is absent, the chat degrades gracefully to using only the provided report context, ensuring the feature remains optional and non-breaking for existing deployments.

backend/chat · high confidence

Introduce split-pane research interface with interactive chat copilot

The research view now supports a side-by-side layout where the main research results are displayed in a resizable ResearchPanel alongside a new CopilotPanel for interactive follow-up questions. This new interface includes a dedicated CopilotResearchContent component that manages the split-pane state, auto-shows the chat panel upon research completion, and handles mobile responsiveness. Additionally, a NotFoundContent component is added to handle missing research reports gracefully.

frontend/nextjs/components/research · high confidence

Introduction of BasicReport class with configurable research parameters

The backend now exposes a BasicReport class that initializes the GPTResearcher with support for query domains, tone selection, and MCP (Model Context Protocol) configurations. Users can now pass specific headers, MCP configs, and strategy settings to tailor the research process, while the system automatically generates unique research IDs for tracking.

_backend/report\_type/basic\report · high confidence

Introduction of STORM-inspired detailed report architecture

A new detailed report generation mode has been added, implementing a multi-step architecture inspired by the STORM paper. This approach first conducts initial research, then generates subtopics, and finally constructs a comprehensive report by writing an introduction, a table of contents, and distinct subtopic sections while actively preventing information duplication across sections. The implementation includes the \DetailedReport\ class, context normalization for handling mixed data formats, and support for passing configuration parameters like tone and MCP settings to sub-researchers.

_backend/report\_type/detailed\report · high confidence

New AG2-based multi-agent research example

Added a new example under \multi\_agents/ag2\ that uses the AG2 framework to orchestrate the GPT Researcher multi-agent workflow. This variant mirrors the existing LangGraph flow with the same set of agents (Human, Chief Editor, Researcher, Editor, Reviewer, Reviser, Writer, Publisher) and stages (Planning, Data collection, Review, Writing, Publication), but implements the orchestration layer using AG2's \ConversableAgent\ and \GroupChat\ components. Users can run this variant via \python -m multi\_agents.ag2.main\ and customize the research query, model, and other parameters via \task.json\.

_multi\agents/ag2 · high confidence

New BeautifulSoup scraper with robust content extraction

A new BeautifulSoup-based scraper has been introduced to handle web page fetching and content parsing. This component improves reliability by implementing retry logic for transient HTTP errors (429, 500, 502, 503, 504) and skipping pages that exceed a 10MB size limit. It also enhances data quality by correctly handling character encodings based on Content-Type headers and extracting page titles alongside cleaned text and relevant images.

_gpt\_researcher/scraper/beautiful\soup · high confidence

New Deep Agents example using GPT Researcher as the research engine

This location introduces a new example that integrates GPT Researcher into LangChain's Deep Agents harness. The example features a Chief Editor agent that plans reports and delegates parallel research tasks to subagents, which use GPT Researcher's \quick\_search\ and \deep\_research\ tools to produce cited drafts. It also includes a comprehensive benchmark suite (\BENCHMARK.md\, \benchmark.py\, and a realistic private-docs corpus) demonstrating that this setup significantly increases verified citations and private-document fact coverage compared to a raw search baseline.

_deep\agents · high confidence

New Deep Research capability with recursive exploration

Added a new 'deep' report type that performs recursive, tree-like research by generating multiple search queries at each level and following leads to greater depths. This feature introduces configurable parameters for breadth (parallel paths), depth (recursion levels), and concurrency, along with real-time progress tracking via callbacks. The implementation utilizes reasoning-capable LLMs (such as o3-mini) for complex analysis and question generation, requiring specific model permissions and resulting in higher API usage and longer execution times compared to standard research.

_backend/report\_type/deep\research · high confidence

New GPTResearcher React component for embedding

The frontend now includes a new \GPTResearcher\ React component (exported via \frontend/nextjs/src\) that allows the application to be embedded as a library in other React projects. This component accepts configuration props such as \apiUrl\, \apiKey\, and \defaultPrompt\, and manages the research workflow using WebSocket connections and API calls. It includes necessary styling imports (Tailwind, markdown, settings) and type definitions to support integration into external React applications.

frontend/nextjs/src · high confidence

New Google Serper retriever with configurable base URL and filtering

Added a new SerperSearch retriever that supports country, language, and time-range filtering, as well as domain and site exclusion. The retriever's base API URL is now configurable via the SERPER\_API\_BASE environment variable (defaulting to https://google.serper.dev), allowing use of proxies or regional endpoints. It also includes robust error handling, returning an empty list instead of None on failures or malformed responses, and uses a configurable timeout to prevent hanging.

_gpt\researcher/retrievers/serper · high confidence

New PWA and embed capabilities for the frontend

The frontend now supports Progressive Web App (PWA) functionality and an embeddable widget. A new \manifest.json\ configures the app as a standalone PWA with icons and shortcuts, while a service worker (\sw.js\) using Workbox precaches assets and defines caching strategies for offline resilience. Additionally, a new \embed.js\ script allows the application to be embedded in other sites via an iframe, supporting custom API URL configuration and height adjustments. Also in this area: 2 commits (0 fixes) modifying frontend.

frontend/nextjs/public · medium confidence · unverified

New SimpleQA evaluation framework with structured logging and metrics

The \evals/simple\_evals\ module now provides a dedicated evaluation framework for assessing model performance. It introduces structured JSON logging for individual run results, capturing detailed metrics such as accuracy, F1 score, latency percentiles, and cost per query. The module includes a multi-domain Simple QA test set (covering topics like Science, Politics, and Sports) and a README documenting the evaluation methodology and historical performance benchmarks.

_evals/simple\evals · high confidence

New Task UI components with domain filtering and report rendering

The frontend now includes a new set of components for the Task interface: an Accordion for displaying structured logs with markdown support, an AgentLogs component for raw agent output, a Report component that renders markdown research reports as HTML, and a ResearchForm that allows users to configure report type, source, tone, layout, and MCP settings. A key addition is the DomainFilter component, which lets users add and remove specific domains to filter web research sources, with these preferences persisted to local storage.

frontend/nextjs/components/Task · high confidence

New Tavily search retriever with domain filtering and robust error handling

A new TavilySearch retriever has been added to the gpt\_researcher package, enabling web searches via the Tavily API. This implementation supports filtering results by specific domains through the query\_domains parameter and automatically translates Google-style 'site:' operators in search queries into compatible domain filters. It also includes resilience improvements, such as handling malformed API responses and guarding against missing fields in search results to prevent crashes.

_gpt\researcher/retrievers/tavily · high confidence

New browser-based web scraper with Selenium and zendriver support

A new browser scraping module has been added to the gpt\_researcher package, introducing two distinct scraping backends: a synchronous \BrowserScraper\ powered by Selenium (supporting Chrome, Firefox, and Safari) and an asynchronous \NoDriverScraper\ powered by zendriver. The Selenium scraper handles cookie management, headless mode, and basic text/image extraction, while the zendriver scraper adds advanced features like browser tab/window management, domain-based rate limiting, and load balancing across multiple browser instances. Both scrapers include utilities for extracting hyperlinks, processing HTML, and scraping PDF content via PyMuPDF and Arxiv.

_gpt\researcher/scraper/browser · high confidence

New context filtering system with multiple ranking strategies

The context module now supports configurable relevance filtering for scraped content via the CONTEXT\_FILTER setting. Users can choose between 'jev' (TypeSafe's Jev model for direct usefulness scoring), 'keyword' (local BM25 ranking), 'embeddings' (vector similarity), or 'none' (no filtering). The 'auto' mode defaults to Jev if a TYPESAFE\_API\_KEY is present, otherwise falling back to keyword. This replaces the previous embedding-only approach, allowing users to reduce costs or improve relevance without relying on embedding models.

_gpt\researcher/context · high confidence

New document loading infrastructure with Azure and LangChain support

The document processing module has been restructured to introduce three new loader classes: \DocumentLoader\ for local files, \OnlineDocumentLoader\ for remote URLs, and \LangChainDocumentLoader\ for pre-loaded LangChain documents. \OnlineDocumentLoader\ now includes SSRF protection via URL validation and sends a standard Mozilla User-Agent header during downloads. \DocumentLoader\ adds support for HTML files using \BSHTMLLoader\ and improves robustness by tolerating missing metadata sources. A new \AzureDocumentLoader\ enables fetching documents directly from Azure Blob Storage containers. The module's public API is updated in \\_\init\\_.py\ to export these new loaders.

_gpt\researcher/document · high confidence

New evaluation framework and Jev-based content filtering

The system now includes a quality evaluation suite (evals/quality\_eval) that measures report faithfulness, answer relevancy, source quality, and citation accuracy using a standardized metric interface. Additionally, scraped content is filtered using TypeSafe's Jev model to score chunk usefulness directly against the query, replacing reliance on embeddings alone for relevance ranking. A new JSON schema generator utility has also been added to structure data based on Pydantic models.

gpt-researcher · high confidence

New hallucination evaluation tool for model outputs

A new \HallucinationEvaluator\ has been added to the \evals/hallucination\_eval\ directory, allowing users to assess whether model-generated summaries contain hallucinations by comparing them against source documents. The tool uses the \HaluEvalDocumentSummaryNonFactual\ judge to return a binary hallucination flag and reasoning. This evaluation suite is pre-configured with a multi-domain dataset of 105 search queries covering healthcare, legal, history, climate, finance, education, and science, along with sample aggregate results.

_evals/hallucination\eval · high confidence

New helper utilities for object diffing, host resolution, and secure markdown rendering

The frontend now includes three new helper modules in the \frontend/nextjs/helpers\ directory. \findDifferences.ts\ provides a recursive utility to compare two objects and return their structural differences, supporting nested objects, arrays, and primitive values. \getHost.ts\ centralizes backend URL resolution by checking \localStorage\, URL parameters, environment variables (\NEXT\_PUBLIC\_GPTR\_API\_URL\, \REACT\_APP\_GPTR\_API\_URL\), and specific purpose flags (like \langgraph-gui\) before falling back to the current window host or a default localhost port. \markdownHelper.ts\ introduces secure markdown-to-HTML conversion using \remark\ and \remark-gfm\, featuring automatic sanitization via \DOMPurify\ to prevent XSS attacks on untrusted content, fixes for list item paragraph formatting, and the addition of \target="\_blank"\ and \rel="noopener noreferrer"\ attributes to all links for security and usability.

frontend/nextjs/helpers · high confidence

Added ImagesAlbum and ImageModal components to the frontend, providing a responsive grid gallery with lazy loading and a full-screen modal viewer. The modal supports keyboard navigation (arrow keys, Escape) and mobile swipe gestures, while the album automatically filters out broken images to prevent display errors.

frontend/nextjs/components/Images · high confidence

New image generation providers for Gemini and ModelsLab

Users can now generate images within GPT Researcher using two new providers: Google's Gemini/Imagen models (via the \google-genai\ SDK) and ModelsLab (supporting Flux, SDXL, and other community models). The \ImageGeneratorProvider\ class handles Google models, defaulting to \gemini-2.5-flash-image\ and supporting landscape cropping for report layouts, while \ModelsLabImageGeneratorProvider\ integrates with the ModelsLab API for alternative model access. Both providers save generated images to the \outputs/images/\ directory and require their respective API keys (\GOOGLE\_API\_KEY\/\GEMINI\_API\_KEY\ or \MODELSLAB\_API\_KEY\) to be set.

_gpt\_researcher/llm\provider/image · high confidence

New modular FastAPI server with agent discovery and WebSocket support

The backend/server directory has been restructured into a new modular FastAPI application. This introduces an agent discovery manifest endpoint at /.well-known/agent-discovery.json that advertises HTTP and WebSocket services (research, reports, chat, and real-time streaming) to external agents. The server now supports real-time interactive research via WebSocket connections, managed by a new WebSocketManager, and persists report history to a local JSON file via a new ReportStore. Additionally, it includes a structured JSON logging system for research events and supports multi-agent execution through a dedicated runner.

backend/server · high confidence

New modular configuration system with deep research and image generation support

The configuration system has been restructured into a modular setup (base, default, and local overrides) to support new capabilities and finer control. Users can now enable Deep Research with configurable breadth, depth, and concurrency settings, and optionally generate images within reports using Google or ModelsLab providers. The system also introduces Model Context Protocol (MCP) integration for tool selection, allows setting LLM and embedding kwargs via environment variables, and updates default models to the GPT-5.4 family with adjusted token limits and reasoning effort controls.

_gpt\researcher/config/variables · high confidence

New modular scraper architecture with multiple backend support

The scraping subsystem has been restructured into a modular package with a central \Scraper\ class that orchestrates content extraction using pluggable backends. New supported scrapers include Tavily Extract, FireCrawl, NoDriver, and Browser-based scraping, alongside existing ones like BeautifulSoup and PyMuPDF. The central scraper now automatically installs missing dependencies (e.g., \tavily-python\, \firecrawl-py\) at runtime if they are not present. It also introduces URL deduplication to prevent redundant scraping, improved PDF detection for extensionless URLs, and better handling of anti-bot block pages and word-list noise. Image extraction logic has been refined to prioritize relevant images based on class and dimensions, and title extraction now robustly handles empty or nested title tags.

_gpt\researcher/scraper · high confidence

New quality evaluation suite for autonomous research reports

A new \quality\_eval\ module has been added to the \evals\ directory to assess the quality of GPT-Researcher reports without requiring ground-truth answers. This suite evaluates reports across four dimensions: Faithfulness (claim grounding), Answer Relevancy (subtopic coverage), Source Quality (authority and diversity of retrieved sources), and Citation (reference accuracy). It provides a standardized framework for scoring, including zero-cost structural metrics and optional LLM-based checks, along with runners for single-query evaluation and multi-configuration benchmarking.

_evals/quality\eval · high confidence

New research history, analytics, and WebSocket management hooks

This change introduces several new React hooks in the frontend to enhance user experience and observability. The \useResearchHistory\ hook and its associated context provider enable persistent tracking of research sessions, syncing local storage with the server to maintain a searchable history of past reports. A new \useAnalytics\ hook integrates Google Analytics (react-ga4) to track research queries, allowing for better insight into user behavior. Additionally, \useWebSocket\ manages real-time communication with the backend, implementing a heartbeat mechanism (ping/pong every 30 seconds) to maintain stable connections and handle streaming data updates. Finally, \useScrollHandler\ provides a utility for managing scroll-to-bottom behavior in content-heavy views.

frontend/nextjs/hooks · high confidence

New static frontend with WebSocket monitoring and conversation history

The frontend now includes a new static HTML/JS interface (served via FastAPI) that replaces or supplements the previous UI. This interface introduces a WebSocket status panel for real-time connection monitoring, a conversation history panel with import/export capabilities, and a redesigned research form supporting report types like 'Deep Research' and tone selection. It also features a new dark gradient theme, Font Awesome icons, and fixes for WebSocket socket disposal to prevent duplicate connections.

frontend · high confidence

New utility module for multi-agent file handling, LLM calls, and output formatting

The multi\_agents/agents/utils package has been introduced to centralize shared capabilities for the multi-agent workflow. It provides asynchronous file I/O utilities (file\_formats.py) that write text to Markdown, convert Markdown to PDF (using a new Libre Baskerville CSS theme) and DOCX, and ensure UTF-8 encoding. It includes an LLM adapter (llms.py) that routes requests through the existing configuration and handles JSON parsing. To improve reliability in human-in-the-loop steps, it adds strict sentinel parsers (none\_sentinels.py) that accurately detect 'None' and 'no' responses without false positives. Finally, it offers filename sanitization (utils.py) and colored console output formatting (views.py) for better cross-platform compatibility and readability.

_multi\agents/agents/utils · high confidence

New utility modules for cost tracking, structured logging, and URL security

This change introduces a suite of new utility modules in the \gpt\_researcher/utils\ package. \costs.py\ adds LLM cost estimation capabilities, including support for Anthropic usage metadata and prompt caching discounts. \logger.py\ and \logging\_config.py\ implement a structured logging system with colorized console output and JSON-based research event logging. \url\_security.py\ adds SSRF and local-file-read protections by validating URLs against disallowed schemes and private IP addresses. Additionally, \rate\_limiter.py\ provides a global singleton rate limiter to control scraper request frequency across all worker pools, and \workers.py\ introduces a \WorkerPool\ class that enforces both concurrency limits and the global rate limit.

_gpt\researcher/utils · high confidence

Next.js frontend with API proxying and research history

The application has moved to a Next.js frontend structure under frontend/nextjs/app, introducing server-side API routes that proxy requests to the backend service (configured via NEXT\_PUBLIC\_GPTR\_API\_URL). These routes handle chat interactions (/api/chat), report management (CRUD operations at /api/reports and /api/reports/\[id\]), and report-specific chat (/api/reports/\[id\]/chat). The frontend now includes a ResearchHistoryContext for managing research state and history, supports both desktop and mobile layouts, and integrates Google Analytics and Plausible for tracking. The UI features a new sidebar experience, improved mobile responsiveness, and a premium visual design with background gradients and custom loaders. Also in this area: 1 commit (0 fixes) modifying (repo-wide).

frontend/nextjs/app · medium confidence · unverified

Next.js frontend with PWA and React component library support

The Next.js frontend now supports Progressive Web App (PWA) capabilities and can be distributed as a React component library (gpt-researcher-ui). Users can install the UI as an npm package to embed the research interface in their own React applications, while the standalone Next.js app provides a premium-feel interface with image generation, Google Analytics integration, and a redesigned UI. The build system uses Rollup for library bundling and Next.js for the web application, with Docker support for both development and production environments.

frontend/nextjs · high confidence

Self-hosted frontend vendor assets

The frontend now serves pinned versions of Bootstrap (4.5.2), Font Awesome Free (6.5.1), Showdown (1.9.1), and DOMPurify (3.2.6) locally from the \frontend/vendor\ directory instead of loading them from third-party CDNs. This change ensures that self-hosted instances of the application do not make external network requests during page load, improving privacy and reliability in restricted environments. Corresponding license files for these dependencies have also been included.

frontend/vendor · high confidence

Unified embedding provider management via new Memory class

The \gpt\_researcher/memory\ module now provides a centralized \Memory\ class that manages document embeddings across a wide range of providers, including OpenAI, Azure OpenAI, Cohere, Google Vertex AI, Ollama, HuggingFace, AWS Bedrock, and many others. This change introduces a unified interface for generating embeddings, allowing users to switch between different embedding services by simply specifying the provider name and model, with support for custom OpenAI-compatible APIs and local models like Ollama.

_gpt\researcher/memory · high confidence

Web base scraper now extracts page title and images alongside content

The web\_base\_loader scraper has been refactored to return a tuple of (content, image\_urls, title) instead of just content. It now uses a utility function to extract the page title and relevant images from the HTML, providing richer metadata to users consuming the scraped data.

_gpt\_researcher/scraper/web\_base\loader · high confidence

Architecture

Introduce modular skills architecture for research, curation, and generation

The \gpt\_researcher/skills\ package has been introduced to replace the monolithic agent structure with a modular, component-based architecture. This change introduces dedicated manager classes for distinct research phases: \ResearchConductor\ for orchestrating the research workflow and query planning, \ContextManager\ for retrieving and compressing context, \SourceCurator\ for ranking and filtering sources, \BrowserManager\ for parallel web scraping, \ImageGenerator\ for AI-generated visuals, and \ReportGenerator\ for writing the final report. This restructuring improves code clarity and maintainability while preserving the existing research capabilities.

_gpt\researcher/skills · high confidence

Behavioural changes

4 commits (0 fixes) modifying evals/simple\_evals/logs

A change to existing behaviour in evals/simple\_evals/logs — 4 commits, 4 files: “feat(evals): Add SimpleQA evaluation framework and initial results”; “docs: fix broken README language and evals links”; “move evals back to root”; and 1 more.

_evals/simple\evals/logs · medium confidence · unverified

ArXiv scraper now uses the official Client API and includes metadata

The ArXiv scraper has been refactored to use the maintained \arxiv\ Client API instead of the deprecated \Search.results()\ method, fixing compatibility issues with arxiv\>=2.2. It now extracts and returns the publication date and author names alongside the abstract, and gracefully handles invalid or non-matching URLs by returning empty results rather than raising errors.

_gpt\researcher/scraper/arxiv · high confidence

Automatic image URL rewriting for npm package compatibility

A new image transformation utility has been added to the frontend utilities to ensure images render correctly when the code is consumed as an npm package. This plugin automatically rewrites local image paths (such as \/img/...\ or \img/...\) to absolute URLs pointing to \https://gptr.app/img/...\, resolving issues where relative paths would otherwise break in external package contexts.

frontend/nextjs/src/utils · high confidence

Exa Search retriever now supports domain filtering and handles malformed responses

The Exa retriever has been updated to accept a \query\_domains\ parameter, which is passed to the Exa API via \include\_domains\ to restrict search results to specific domains. Additionally, the implementation now robustly handles malformed or incomplete API responses by safely extracting URLs and text/summary fields using \getattr\, skipping any result objects that lack a valid URL or ID, and returning an empty list on API errors instead of crashing.

_gpt\researcher/retrievers/exa · high confidence

Introduction of generic LLM and image generation provider interfaces

The \gpt\_researcher/llm\provider\ package now exposes a unified entry point via \\\init\\_.py\, making \GenericLLMProvider\ and \ImageGeneratorProvider\ directly importable from the package root. This change consolidates the provider abstractions, allowing users to access the core LLM and image generation interfaces without needing to know the specific internal module paths (e.g., \.generic\ or \.image\).

_gpt\_researcher/llm\provider · high confidence

Langgraph research component initialization with empty authentication token

The Langgraph research component in the frontend Next.js application has been introduced to facilitate AI-driven research tasks. This component initializes a Langchain LangGraph client to search for assistants and stream run responses based on user queries. Notably, the component currently includes a hardcoded, empty string for the API authentication token ('X-Api-Key'), meaning the client is configured without valid credentials for the specified host URL.

frontend/nextjs/components/Langgraph · medium confidence

Multi-agent research workflow with bounded revision loops and new agents

The multi-agent research system now uses a LangGraph-based workflow orchestrated by ChiefEditorAgent, introducing FactCheckerAgent and VisualizerAgent alongside existing roles. The EditorAgent manages parallel section research and enforces a bounded draft revision loop (default 3 rounds) via draft\_review.py to prevent infinite cycles. A HumanAgent allows user feedback on the research plan, also bounded by a configurable max\_plan\_revisions limit. The FactCheckerAgent validates factual accuracy with a separate bounded revision loop (max\_fact\_check\_revisions), and the VisualizerAgent generates Mermaid.js diagrams from research data. The PublisherAgent now handles output formatting to PDF, DOCX, and Markdown.

_multi\agents/agents · high confidence

New dedicated styling for rendered markdown content

A new CSS file (markdown.css) has been introduced to define the visual appearance of markdown content within the Next.js frontend. This stylesheet establishes a consistent theme for reports and other text-based outputs, specifying typography (Georgia serif, 18px base), spacing, and formatting for elements like headings, paragraphs, links, code blocks, and blockquotes. It also includes specific styling for tables (with dark theme matching, alternating row colors, and hover effects), images (including special handling for generated illustrations/diagrams), and task lists, ensuring that rendered markdown aligns with the application's dark UI design.

frontend/nextjs/styles · high confidence

New modular configuration system with multi-retriever and LLM separation

The configuration module has been refactored into a new structure (\gpt\_researcher/config\) that introduces a \Config\ class for centralized management of settings. This change allows users to configure multiple retrievers simultaneously via the \RETRIEVER\ environment variable (defaulting to 'tavily') instead of a single provider. It also separates LLM settings into distinct \fast\_llm\ and \smart\_llm\ configurations, enabling different models for different research steps. The system now supports loading custom JSON configuration files via \CONFIG\_PATH\, merges them with defaults, and handles deprecated environment variables (like \LLM\_PROVIDER\) with warnings.

_gpt\researcher/config · high confidence

New research interface components with premium visual styling

The research block UI now includes new components for the input experience and result display. InputArea and ChatInput provide text entry with auto-resizing textareas, debounced input handling, and a consistent 'premium' aesthetic featuring teal-to-cyan gradient glows and backdrop blurs. SubQuestions renders interactive suggestion chips for follow-up angles. LogMessage processes and displays structured logs, including markdown rendering for 'differences' and accordion views. SourceCard displays source attribution with domain extraction and favicon fallbacks.

frontend/nextjs/components/ResearchBlocks/elements · high confidence

Normalize DuckDuckGo search results and fix snippet length handling

The DuckDuckGo retriever now normalizes search results to the shared \{href, body, title}\ contract, ensuring compatibility with downstream research steps that expect \href\ keys regardless of whether the underlying \ddgs\ library returns \link\, \url\, \body\, \snippet\, or \description\. Additionally, the retriever explicitly declares that results require scraping (\requires\_scraping = True\) and caps the prefetched \body\ snippet at 100 characters to prevent the system from incorrectly treating standard search snippets as full article text, which previously caused citation failures.

_gpt\researcher/retrievers/duckduckgo · high confidence

Redesigned Next.js frontend with mobile-first UI and new component structure

The frontend has been completely rebuilt using Next.js components, introducing a modern, premium visual design with animated particle backgrounds, gradient glows, and responsive layouts. New components include a sticky Header with stop/new research controls, a Hero section for input, a collapsible Research Sidebar for history, and a Footer with social links. A dedicated mobile experience is provided via MobileHomeScreen, MobileResearchContent, and MobileChatPanel, ensuring full functionality on smaller screens. The ResearchResults component now intelligently filters data to show only the final report block, while HumanFeedback and LoadingDots enhance the interaction flow.

frontend/nextjs/components · high confidence

Redesigned Settings modal with new configuration components

The Settings interface has been completely redesigned, introducing a new 'Preferences' modal that consolidates configuration options. This update adds specific UI components for managing research layout (Research vs. Copilot), output tone (17 options including Objective, Formal, and Humorous), and Model Context Protocol (MCP) server configurations with JSON validation. Additionally, a new File Upload component allows users to drag-and-drop, view, and delete files directly within the settings area, while the modal itself now features animated transitions and a dark-themed UI.

frontend/nextjs/components/Settings · high confidence

Redesigned research report interface with new interactive components

The research report view has been completely redesigned to offer a more premium, interactive experience. Users can now access reports in PDF, DocX, and JSON formats via the new AccessReport component, which handles file path resolution. The interface includes a ChatInterface for asking questions about the report content, with real-time markdown rendering and loading states. Chat responses (ChatResponse) now display web search sources when available, and the main report (Report) features a copy-to-clipboard function. Additional components like ImageSection, LogsSection, and Sources provide structured views for related images, agent work logs, and source citations, all styled with a consistent dark theme and backdrop blur effects.

frontend/nextjs/components/ResearchBlocks · high confidence

Restructured actions module with improved JSON parsing and deterministic references

The \gpt\_researcher/actions\ package has been reorganized into dedicated modules (\agent\_creator\, \query\_processing\, \report\_generation\, \retriever\, \web\_scraping\, \markdown\_processing\, and \utils\) to improve code clarity and maintainability. Agent selection now uses \json\_repair\ for more robust handling of malformed LLM JSON responses, and sub-query generation defensively normalizes various LLM output formats to ensure a consistent list of strings. Report generation now applies the configured language setting to the Introduction and Conclusion sections, and the References section in the final report is sorted to ensure deterministic output. Additionally, the retriever factory supports third-party plugins via entry points and strips whitespace from header-supplied retriever lists to prevent silent failures.

_gpt\researcher/actions · high confidence

Restructured frontend layout and data processing logic

The frontend now uses a new \getAppropriateLayout\ utility to dynamically select between Mobile, Copilot, and Research page layouts based on screen size and user settings, replacing the previous static layout approach. Additionally, data processing has been modularized into \dataProcessing.ts\ and \consolidateBlocks.ts\, which now handle the aggregation of report content, deduplication of source URLs, and consolidation of image blocks before rendering, ensuring a cleaner and more consistent presentation of research results.

frontend/nextjs/utils · high confidence

Standardized retriever contract and expanded search engine support

The retrievers module has been restructured to introduce a clear contract for how search results are handled, specifically adding a \requires\_scraping\ attribute to distinguish between retrievers that return URLs needing further fetching and those that return full content (like PubMed Central). This change fixes previous bugs where snippet length heuristics incorrectly determined whether scraping was needed, ensuring reliable citation generation. Additionally, the module now supports a significantly wider range of search engines, including new integrations for Arxiv, Bing, Brave, Exa, GroundRoute, OpenAlex, PubMed Central, Semantic Scholar, and several others, while also enabling the loading of third-party retrievers via entry points.

_gpt\researcher/retrievers · high confidence

Fixes

Fixes Google Custom Search retrieval robustness and domain filtering

The Google Search retriever now properly URL-encodes query parameters to prevent parsing errors with special characters (e.g., '&', '\#') and supports filtering results by specific domains via the \query\_domains\ parameter. Additionally, the implementation guards against malformed API responses by safely handling JSON parsing errors, skipping non-dict items, and excluding YouTube links, ensuring that search results are returned in a consistent format or an empty list on failure rather than crashing.

_gpt\researcher/retrievers/google · high confidence

PyMuPDF scraper now extracts content from all pages and handles SSL/temp-file edge cases

The PyMuPDF scraper has been refactored to read and return content from all pages in a PDF document, rather than just the first page, ensuring that documents with cover pages or multi-page structures are fully processed. Additionally, the scraper now includes an SSL verification fallback (retrying without verification on SSL errors) and guarantees that temporary downloaded files are cleaned up even if PDF loading fails, improving reliability and preventing disk clutter.

_gpt\researcher/scraper/pymupdf · high confidence

Test coverage

8 commits adding/updating tests in tests/skills; Added backend tests for report chat, self-hosted assets, WebSocket cleanup, and PDF export; Added tests for chat RAG context retrieval and search source tracking; Context filter benchmark suite and documentation; Expanded test coverage for retrievers, scrapers, and agent behavior.

Dependencies

Initial dependency manifests for GPT Researcher v0.16.0

This change introduces the initial set of dependency manifests (requirements.txt, pyproject.toml, and package.json files) for the project, establishing the baseline for version 0.16.0. It defines the core Python stack, including FastAPI, Pydantic, and the LangChain v1 suite (langchain, langgraph, langchain-openai), alongside specific libraries for document processing (pymupdf, python-docx), search (tavily-python, ddgs), and LLM providers (openai, ollama, litellm). The manifests also include security pinning for brotli to address [CVE redacted], frontend dependencies for the Next.js UI and Docusaurus documentation site, and configuration for evaluation and multi-agent modules.

(dependencies) · high confidence

Written by cca.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

How this codebase got here

This is the PUBLIC form of this artifact. Findings are listed in full, but the details of SECURITY findings — which rule fired, in which file, on which line, and how to fix it — are deliberately withheld, and any secret-scanner results are excluded entirely. Where detail is absent here it was REMOVED FOR PUBLICATION; it is not missing from the analysis. The complete artifact is available from the repository owner.

Score

  • CAI 42 → 45 (+3.1)
  • Rubric changed (rubric-2026.11.2 → rubric-2026.11.4) — scores are not directly comparable.

Lenses

  • Code Health 54 → 50 (-4.0)
  • Architecture 96 → 96 (+0.1)
  • Maturity 67 → 76 (+9.5)
  • Readiness 37 → 44 (+6.9)
  • Security 36 → 37 (+1.2)
  • Accessibility 43 → 48 (+5.1)
  • Performance 85 → 85 (+0.0)

Resolved (57)

  • CI installs an unverified third-party binary (.github/workflows/docker-build.yml)
  • Change coupling: page.tsx ↔ Header.tsx (frontend/nextjs/app/page.tsx)
  • Duplicated block (15 lines × 2) (multi_agents/agents/fact_review.py)
  • Duplicated block (18–20 lines × 2) (gpt_researcher/llm_provider/generic/base.py)
  • FileTooLong: app/page.tsx (frontend/nextjs/app/page.tsx)
  • FunctionTooLong: page.Home (frontend/nextjs/app/page.tsx)
  • FunctionTooLong: page.ResearchPage (frontend/nextjs/app/research/[id]/page.tsx)
  • Further sole-owners (lower concentration)
  • High: security finding (details withheld)
  • High: security finding (details withheld)
  • High: security finding (details withheld)
  • High: security finding (details withheld)
  • High: security finding (details withheld)
  • High: security finding (details withheld)
  • High: security finding (details withheld)
  • High: security finding (details withheld)
  • High: security finding (details withheld)
  • High: security finding (details withheld)
  • High: security finding (details withheld)
  • High: security finding (details withheld)
  • …and 37 more

New (92)

  • ArchiveScreen.ArchiveScreen (cognitive 18) (frontend/nextjs/app/(gr)/_components/ArchiveScreen.tsx)
  • ArchiveScreen.ArchiveScreen (cyclomatic 17) (frontend/nextjs/app/(gr)/_components/ArchiveScreen.tsx)
  • BingSearch.search (cognitive 21) (gpt_researcher/retrievers/bing/bing.py)
  • BingSearch.search (cyclomatic 19) (gpt_researcher/retrievers/bing/bing.py)
  • Change coupling: page.tsx ↔ Header.tsx (frontend/nextjs/app/classic/page.tsx)
  • Duplicated block (12–13 lines × 3) (gpt_researcher/retrievers/exa/exa.py)
  • Duplicated block (14–16 lines × 2) (gpt_researcher/retrievers/exa/exa.py)
  • Duplicated block (17–18 lines × 2) (gpt_researcher/skills/researcher.py)
  • Edited copies of one member (3 members, 11+ corresponding lines) (gpt_researcher/retrievers/crw/crw.py)
  • Edited copies of one member (3 members, 27+ corresponding lines) (gpt_researcher/retrievers/exa/exa.py)
  • Edited copies of one member (3 members, 3+ corresponding lines) (backend/report_type/basic_report/basic_report.py)
  • Edited copies of one member (3 members, 7+ corresponding lines) (evals/quality_eval/benchmark.py)
  • Edited copies of one member (4 members, 27+ corresponding lines) (gpt_researcher/skills/writer.py)
  • Edited copy of a member (10 corresponding lines) (evals/quality_eval/suite.py)
  • Edited copy of a member (10 corresponding lines) (multi_agents/agents/researcher.py)
  • Edited copy of a member (113 corresponding lines) (gpt_researcher/skills/image_generator.py)
  • Edited copy of a member (17 corresponding lines) (gpt_researcher/config/config.py)
  • Edited copy of a member (17 corresponding lines) (gpt_researcher/prompts.py)
  • Edited copy of a member (18 corresponding lines) (gpt_researcher/context/compression.py)
  • Edited copy of a member (18 corresponding lines) (multi_agents/agents/editor.py)
  • …and 72 more

Changes since last survey

  • 78 commits — 27 feature/other, 51 fixes

By area

  • (repo) — 53 commits
  • (root) — 6 commits
  • .github/workflows — 3 commits
  • gpt_researcher/skills — 3 commits
  • docs/docs — 2 commits
  • frontend/nextjs — 2 commits
  • gpt_researcher/actions — 2 commits
  • gpt_researcher/mcp — 1 commit
  • gpt_researcher/retrievers — 1 commit
  • gpt_researcher/scraper — 1 commit
  • gpt_researcher/utils — 1 commit
  • tests/skills — 1 commit
  • tests/test_langchain_document_loader_guards.py — 1 commit
  • tests/test_report_generation_error_handling.py — 1 commit

Notable commits

  • fix: Merge PR #2164: fix(multi_agents): treat an empty console plan reply as approval, as the web form does
  • fix: Merge PR #2167: fix: revalidate PyMuPDFScraper retry output against content-quality checks
  • fix: Merge PR #2169: fix: ReportGenerator.write_report() must read researcher.websocket live
  • fix: Merge PR #2181: fix(exa): harden find_similar and get_contents against client failures
  • fix: Merge PR #2183: fix(document): harden LangChainDocumentLoader.load against malformed rows
  • fix: Merge PR #2185: docs: document generic MCP connection headers and fix Quick Start syntax
  • fix: Merge PR #2195: fix: resolve PubMed IDs before fetching PMC full text
  • fix: Merge PR #2199: fix(images): read image_generation_* config keys in the case Config stores them
  • fix: Merge PR #2200: fix(document): keep OnlineDocumentLoader url a string when metadata omits source
  • fix: Merge PR #2201: docs: fix broken README language and evals links
  • fix: Merge PR #2208: fix(deep-research): keep cited learnings when the context reaches the word limit
  • fix: Merge PR #2209: fix(deep-research): count strategic LLM calls and nested researchers in costs
  • fix: Merge PR #2213: fix(scraper): treat the "Client Challenge" interstitial as a block page
  • fix: Merge PR #2215: fix(google): match YouTube exclusions by hostname
  • fix: Merge PR #2216: fix: replace retired Google and Anthropic model defaults
  • fix: Merge PR #2217: fix: don't send temperature to Claude models that reject it
  • fix: Merge PR #2219: fix(report): raise error and log failure when generate_report LLM calls fail
  • fix: Merge PR #2220: fix(llm): bound stalled responses with configurable idle timeouts
  • fix: Merge PR #2224: fix(pubmed_central): keep words whole at inline markup and put each block on its own line
  • fix: Merge PR #2230: fix(config): honor reasoning effort from JSON configuration
  • …and 58 more

Architecture

  • Unchanged — 0 containers · 1 contexts · 0 edges

Written by cca.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

Survey your own repository

assafelovic/gpt-researcher was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.

About this page

  • The score is its most recent published measurement, taken on 16 November 2026 at a pinned commit. It is not a live figure and does not change until the project is measured again.
  • Measured at commit ddd4db36208d9c8e49f3b2436efda97baecb5433 — the exact code this score is about.
  • Scored under rubric-2026.11.4 — the same rubric and the same method as every other entry in this index.
  • Measured by Canine Code Analyzer (cca.canine.dev) using cca measure-2026.11.73+preprod-52de90b74216.