Major fixes

What's Changed

• fix: restore main, add CI, and land 97 queued PRs by @assafelovic in #2074
• fix(readme): restore broken star history chart by @Dessalines39394 in #2066
• docs: document scholarly retriever configuration by @pikaqiu2333 in #1863
• chore(deps): update python-pptx requirement from >=1.0.0 to >=1.0.2 by @dependabot[bot] in #1927
• chore: remove the dead poetry dependency block, test the declared Python floor by @assafelovic in #2075
• Add agent observability with Monocle by @imohammedansari in #1873
• fix: make the empty-source-urls notification reachable (#1978) by @assafelovic in #2076
• fix(scraper): keep extracted content when web_base_loader enrichment fetch fails by @AmirF194 in #2078
• feat(retrievers): declare whether results need scraping, instead of guessing by @assafelovic in #2079

New Contributors

• @Dessalines39394 made their first contribution in #2066
• @pikaqiu2333 made their first contribution in #1863
• @imohammedansari made their first contribution in #1873
• @AmirF194 made their first contribution in #2078

Full Changelog: v3.6.0...v3.6.1

v3.6.0

Introducing a super stable version with enhanced improvements across security, stability and performance. We're also introducing a collaboration with Langchain with the latest deep agents framework here - https://github.com/assafelovic/gpt-researcher/tree/main/deep_agents

Thank you and welcome to all the new contributors!

What's Changed

• security: sanitize untrusted content, add SECURITY.md, pin brotli by @assafelovic in #1820
• Fix silently dropped token limits, real usage-based cost tracking, and research pipeline robustness by @assafelovic in #1861
• Add Deep Agents example: GPT Researcher as the research engine in a LangChain deep agent by @assafelovic in #1857
• Resolve post-merge conflicts: 4 retriever/context fixes (supersedes #1895-1898) by @assafelovic in #1906
• Retriever hardening: 20 guard fixes against malformed results (supersedes #1837…#1890) by @assafelovic in #1907
• Scraper robustness: 6 fixes (title/PDF-detection/temp-file/dimensions) (supersedes #1824…#1842) by @assafelovic in #1908
• Multi-agent robustness: bound revision loops + exact sentinels (supersedes #1883/#1885/#1886) by @assafelovic in #1909
• Core/misc hardening: 11 fixes (costs/query/llm/agent/mcp/config) (supersedes #1822…#1902) by @assafelovic in #1910
• Reconcile master→main: GetXAPI retriever, Claude 4.x temp fix, MiniMax M3, curate_sources fixes by @assafelovic in #1911
• feat: add Nebius Token Factory as LLM and embedding provider by @teionarr in #1891
• chore(multi_agents): nest AG2 variant under multi_agents/ag2 by @assafelovic in #1912
• docs(skill): fix drift in .claude reference docs (retrievers, multi-agents, api) by @assafelovic in #1913

New Contributors

• @teionarr made their first contribution in #1891

Full Changelog: v3.5.1...v3.6.0

v3.5.1

What's Changed

• fix: add MAX_REVISIONS guard to prevent infinite revision loop by @varshiniml7 in #1783
• Improve SimpleQA eval observability with structured logging and latency metadata by @SeanHe727 in #1790
• feat(retrievers): add Brave Search support by @agbocsardi in #1793
• Abstain instead of fabricating a report when no content was gathered by @aaravanmay in #1799
• fix(deep-research): keep oversized context from collapsing to empty by @XiaojuCH in #1802
• feat(retrievers): add fastCRW search retriever by @us in #1806
• feat: Add FactChecker and Visualizer Agents to Multi-Agent Workflow by @KasraBashirioskooei in #1809
• Add GroundRoute retriever (multi-engine search, cached) by @jp0xz in #1812
• feat: add Atlas Cloud as an OpenAI-compatible LLM provider by @lucaszhu-hue in #1813
• Update web_scraping.py by @Joly0 in #1816
• Fix Azure blob loader path handling by @konnisan in #1817

New Contributors

• @varshiniml7 made their first contribution in #1783
• @SeanHe727 made their first contribution in #1790
• @agbocsardi made their first contribution in #1793
• @aaravanmay made their first contribution in #1799
• @XiaojuCH made their first contribution in #1802
• @us made their first contribution in #1806
• @KasraBashirioskooei made their first contribution in #1809
• @jp0xz made their first contribution in #1812
• @lucaszhu-hue made their first contribution in #1813
• @konnisan made their first contribution in #1817

Full Changelog: v3.5.0...v3.5.1

v3.5.0

Massive update across the stack with critical bug fixes and performance improvements including addition to new models and retrievers. Thanks to the amazing community for pushing this!

What's Changed

• feat: add ModelsLab image generation provider by @adhikjoshi in #1647
• fix: NameError when STRATEGIC_LLM lacks colon separator (#1673) by @sudabg in #1679
• fix: properly resolve report_type imports in websocket_manager and re… by @varadfromeast in #1740
• Fix: resolve unhashable dict error in detailed report context deduplication by @3mk4yl in #1700
• Add Codex CLI plugin manifest by @internet-dot in #1721
• fix: eliminate process-level env pollution from MCP retriever setup by @octo-patch in #1742
• feat(cli): generate filename from LLM and add YAML frontmatter for re… by @VVjwell in #1743
• feat: add OpenAlex retriever by @CuiweiG in #1748
• feat(server): add Agent Discovery Protocol manifest endpoint by @zhao0112 in #1749
• fix: truncate document content to prevent embedding token limit errors by @octo-patch in #1750
• fix: support MISTRAL_BASE_URL env var for custom Mistral-compatible endpoints by @octo-patch in #1754
• Add LLM cost change analysis to CI by @Jwrede in #1761
• chore: fix some typos across codebase by @box4wangjing in #1768
• Limit multi-agent human plan revisions by @CalibratedGhosts in #1769
• fix(deep_research): parse LLM output with json_repair + regex fallback by @RomainROCH in #1773
• fix(llm): raise max_tokens cap from 32k to 200k for modern long-output models by @RomainROCH in #1775
• Fix incorrect config file path in Chinese README by @wesssI in #1779
• fix(retrieval): send Brave snippets to scraper by @RomainROCH in #1780
• feat(cost): track Anthropic usage from native metadata by @RomainROCH in #1781

New Contributors

• @adhikjoshi made their first contribution in #1647
• @sudabg made their first contribution in #1679
• @varadfromeast made their first contribution in #1740
• @3mk4yl made their first contribution in #1700
• @internet-dot made their first contribution in #1721
• @VVjwell made their first contribution in #1743
• @CuiweiG made their first contribution in #1748
• @zhao0112 made their first contribution in #1749
• @Jwrede made their first contribution in #1761
• @box4wangjing made their first contribution in #1768
• @CalibratedGhosts made their first contribution in #1769
• @RomainROCH made their first contribution in #1773
• @wesssI made their first contribution in #1779

Full Changelog: v3.4.4...v3.5.0

v3.4.4

What's Changed

• #1673: Fixed the reference error in code. by @parth3083 in #1674
• fix potential data parsing issue in web scraping by @Carton in #1607
• Fix: Add context normalization for dict and string formats by @GeorgelPreput in #1668
• feat: add MiniMax as a native LLM and embedding provider (M2.7) by @octo-patch in #1677
• Update dependencies and add missing os import by @mparker404 in #1680
• fix: PubMed Central retriever returns no results by @antek-eth in #1686
• Fix missing os import in websocket manager by @sztoplover-bit in #1697
• Bug fix/pdf report by @test23techvv in #1720
• Fix issue 1712 by @test23techvv in #1717
• fix: skip empty LLM streaming chunks and flush stdout by @kiranvk-2011 in #1737
• feat: add Xquik X/Twitter search retriever by @kriptoburak in #1734

New Contributors

• @parth3083 made their first contribution in #1674
• @Carton made their first contribution in #1607
• @octo-patch made their first contribution in #1677
• @mparker404 made their first contribution in #1680
• @antek-eth made their first contribution in #1686
• @sztoplover-bit made their first contribution in #1697
• @test23techvv made their first contribution in #1720
• @kiranvk-2011 made their first contribution in #1737
• @kriptoburak made their first contribution in #1734

Full Changelog: v3.4.3...v3.4.4

v3.4.3

Huge release with major fixes and performance improvements. Welcome to all the incredible new contributors and shoutout to @Br1an67 with major contributions!

What's Changed

• fix: Read all pages in PyMuPDFScraper instead of just first page by @MattBenesch in #1623
• docs: fix invalid Discord invite link in LLM docs by @Br1an67 in #1649
• fix: invalidate import caches after runtime package install by @Br1an67 in #1650
• fix: fallback to OPENAI_API_VERSION for Azure embeddings by @Br1an67 in #1651
• fix: increase PyMuPDF download timeout and add SSL fallback by @Br1an67 in #1652
• fix: use all configured retrievers in deep research planning by @Br1an67 in #1653
• fix: add table styles to FastAPI frontend report container by @Br1an67 in #1654
• feat: track cost per step in research process by @Br1an67 in #1655
• feat: add scrape count option to FastAPI frontend by @Br1an67 in #1656
• feat: add pondering questions section to FastAPI frontend by @Br1an67 in #1657
• fix: dynamically retrieve agent role if missing in ReportGenerator by @kiwabb in #1658
• fix docs links and add ag2 pipeline diagram by @qingyun-wu in #1662
• fix: improve retry handling in create_chat_completion by @jhyz in #1664
• Add support for connection headers in MCP client/server config by @GeorgelPreput in #1665

New Contributors

• @MattBenesch made their first contribution in #1623
• @Br1an67 made their first contribution in #1649
• @kiwabb made their first contribution in #1658
• @qingyun-wu made their first contribution in #1662
• @jhyz made their first contribution in #1664

Full Changelog: v3.4.2...v3.4.3

v3.4.2

This release is packed with performance and quality improvements! Welcome to the new and amazing contributors, such a pleasure as always!

What's Changed

• Update FireCrawl setup instructions for self hosted instance by @technot80 in #1639
• fix(server): resolve multi-agent run_research_task NameError by @AlexanderCGO2 in #1640
• perf: optimize context compression with smart fast-path for small documents by @maanavagrawal in #1641
• perf: add URL deduplication to prevent redundant scraping by @maanavagrawal in #1642
• fix: replace 9 bare except clauses with except Exception by @haosenwang1018 in #1643
• fix: nodriver scraper Docker compatibility and browser pool deadlock by @mareurs in #1645
• feat: add Avian as LLM provider by @avianion in #1646

New Contributors

• @AlexanderCGO2 made their first contribution in #1640
• @maanavagrawal made their first contribution in #1641
• @haosenwang1018 made their first contribution in #1643
• @mareurs made their first contribution in #1645
• @avianion made their first contribution in #1646

Full Changelog: v3.4.1...v3.4.2

v3.4.1

Huge release packed with incredible improvements and welcome to all the new contributors! See details below:

What's Changed

• Update server_utils.py - fixed file name via hash by @MaxAkaAltmer in #1611
• add bocha search tool by @weijintaocode in #1618
• Add openrouter embeddings support by @Joly0 in #1620
• Claude/analyze project structure md q kw by @aiiscoding in #1621
• feat: Add Forge LLM provider support by @Yiiii0 in #1626
• Added AG2 based example by @VasiliyRad in #1627
• Attempt to load config path from env var if no path provided at runtime by @GeorgelPreput in #1630
• fix: handle list context in deep research to prevent AttributeError by @technot80 in #1637

New Contributors

• @MaxAkaAltmer made their first contribution in #1611
• @weijintaocode made their first contribution in #1618
• @Joly0 made their first contribution in #1620
• @aiiscoding made their first contribution in #1621
• @Yiiii0 made their first contribution in #1626
• @VasiliyRad made their first contribution in #1627
• @GeorgelPreput made their first contribution in #1630
• @technot80 made their first contribution in #1637

Full Changelog: v3.4.0...v3.4.1

v3.4.0

Such an exciting release with powerful new features.

🖼️ Inline Image Generation

GPT Researcher can now automatically generate and embed AI-created illustrations directly in your research reports using Google's Gemini models.

Key features:

• Pre-generation flow - Images are generated during research, not after, for seamless UX
• Context-aware - LLM analyzes your research to identify the best visualization opportunities
• Dark mode styling - Generated images match the app's aesthetic by default
• Fully configurable - Control max images, style (dark/light/auto), and model

Quick setup:

IMAGE_GENERATION_ENABLED=true
GOOGLE_API_KEY=your_key
IMAGE_GENERATION_MODEL=models/gemini-2.5-flash-image
IMAGE_GENERATION_STYLE=dark

📖 Image Generation Documentation

---

🔍 LangSmith Integration

Added native support for LangSmith tracing and observability. Monitor your research workflows, debug LLM calls, and analyze performance with full visibility into the research pipeline.

Quick setup:

LANGCHAIN_TRACING_V2=true
LANGCHAIN_API_KEY=your_langsmith_key
LANGCHAIN_PROJECT=gpt-researcher

What you can track:

• LLM calls and token usage
• Research planning and sub-query generation
• Report generation steps
• Cost analysis per research task

📖 LangSmith Logs Documentation

---

🤖 Claude Code Skills Integration

Added comprehensive .claude/skills/ directory that enables Claude Code to understand, use, and extend GPT Researcher effectively.

What's included:

• SKILL.md (1,500+ lines) - Complete architecture, workflows, method signatures, data flow diagrams, and the 8-step feature pattern for adding new features
• REFERENCE.md - Quick lookup for all config vars, API endpoints, and WebSocket events
• Real case studies including the Image Generation implementation as a reference

Benefits:

• Faster contributor onboarding with AI assistance
• Consistent code patterns across contributions
• End-to-end feature development following established patterns

📖 AI-Assisted Development Documentation

---

📚 Documentation Updates

• ✨ New: Image Generation guide
• ✨ New: AI-Assisted Development guide for Claude Code users
• 📝 Updated README with new features section

---

Installation

pip install --upgrade gpt-researcher

Or with Docker:

docker pull gptresearcher/gpt-researcher:latest

---

Full Changelog: v0.14.5...v0.14.6

What's Changed

• Add aggregated summary flag to quick_search by @TheSpaceGod in #1604
• feat: enable LangSmith tracing for enhanced observability by @tiandee in #1599
• added image generation with nano banana by @assafelovic in #1608

Full Changelog: v3.3.9...v3.4.0

Performance improvements

What's Changed

• Fix CORS defaults: remove wildcard origins when allow_credentials=true by @PriscaAmajuoyi in #1595
• docs: add docstrings to memory and context compression modules by @kga245 in #1589
• refactor: Implement ECR and GitHub Actions Terraform setup with necessary configurations and outputs by @huytran-gg in #1590
• feat: Implement WebSocket chat command handling & make Tavily optional by @tiandee in #1593
• Persist /api/reports history to local JSON store (enables Next.js history sync) by @PriscaAmajuoyi in #1594
• Remove hardcoded authentication token by @E-G-C in #1596

New Contributors

• @huytran-gg made their first contribution in #1590
• @tiandee made their first contribution in #1593
• @E-G-C made their first contribution in #1596

Full Changelog: v.3.3.8...v3.3.9