Free AI Research Tools That Replace $20/Month Subscriptions — The Complete Arsenal
Free AI Research Tools That Replace $20/Month Subscriptions — The Complete Arsenal124 tools. Zero subscriptions. Your personal research army — runs on your computer, costs nothing, sees everything.
One-Line Flow:
You ask a question → AI searches the entire internet + scientific papers + Reddit + YouTube → gives you a full answer with sources → all for free, all private, no subscriptions.
Why this matters:
Perplexity costs $20/month and still limits you. ChatGPT Pro is $200/month for “deep research” that hallucinates half its sources. Google’s AI Overviews are mid at best.
Meanwhile, open-source tools do the SAME thing — sometimes dramatically better — completely free, unlimited, and nobody sees what you search. Some scored higher than ChatGPT and Perplexity in blind tests. Some came from Stanford, Tsinghua, ByteDance, and Alibaba labs. Some are single Python files a teenager could run.
This is the complete map. Every tool that exists. Ranked, categorized, and explained for someone who’s never opened a terminal.
PICK YOUR FIGHTER — The Starter Pack
🏆 THE EASY ONES — 5 Minutes, No Skills Needed
Perplexica — The “Just Works” Option
You type a question. It searches the web. It gives you an answer with links. Like Google but it actually answers you.
Modes for research papers, Reddit opinions, YouTube videos. Upload PDFs and ask questions about them. 100% runs on your machine.
docker run -d -p 3000:3000 -v perplexica-data:/home/perplexica/data --name perplexica itzcrazykns1337/perplexica:latest
Open localhost:3000. That’s it.
github.com/ItzCrazyKns/Perplexica — 18k+ stars
Open WebUI — The Swiss Army Knife
Chat interface that connects to ANY AI model. 15+ search engines built in. Works with free Ollama models. DuckDuckGo search — no API keys. Upload files, generate images, voice chat.
github.com/open-webui/open-webui — 60k+ stars
LLocalSearch — Zero API Keys, Zero BS
Searches the web, reads the results, then searches AGAIN based on what it learned. Recursive brain mode. Shows its thinking in real-time.
github.com/nilsherzig/LLocalSearch — 5.5k+ stars
MindSearch — Thinks Like a Human
Breaks your question into sub-questions, searches them ALL simultaneously, combines everything. Processes 300+ web pages per search. In blind tests, preferred over ChatGPT and Perplexity. Free with DuckDuckGo.
git clone https://github.com/InternLM/MindSearch
cd MindSearch && pip install -r requirements.txt
python -m mindsearch.app --lang en --search_engine DuckDuckGoSearch
github.com/InternLM/MindSearch — 6.7k+ stars
Khoj — Your Second Brain
Searches the web AND your personal documents (PDFs, Notion, emails). Remembers context across conversations. Works offline. Apps for WhatsApp, Obsidian, desktop.
github.com/khoj-ai/khoj — 25k+ stars
Deep Research Agents — The Nuclear Options
These aren’t search tools. They’re autonomous research agents that plan, search, read, analyze, and write full reports. Some scored higher than GPT-4 on benchmarks.
🔥 Tier 1 — Research Lab Heavyweights
From Stanford, Tsinghua, ByteDance, Alibaba, HuggingFace. Not hobby projects — open-source versions of what OpenAI charges $200/month for.
| Agent | What Makes It Special | Link |
|---|---|---|
| WebThinker | NeurIPS 2025. Reasoning LLMs (QwQ-32B) with autonomous search within the thinking process. The AI thinks and searches simultaneously. | GitHub |
| AgentCPM | Tsinghua’s 4B/8B agent. SOTA on GAIA benchmark. Small enough for consumer hardware. Fully open-source OpenAI Deep Research alternative. | GitHub |
| MiroThinker | 80.8% GAIA. 256K context. Up to 400 tool calls per task. 30B and 235B scales. | GitHub |
| DeerFlow | ByteDance. Separate Researcher/Coder/Reporter agents. MCP support. RAGFlow integration. Text-to-speech for reports. | GitHub |
| Tongyi DeepResearch | Alibaba. 30.5B params. Automated synthetic data pipeline. SOTA on HLE/BrowseComp/GAIA. | GitHub |
| Auto-Deep-Research | One-click Docker. Ranked #3 globally on GAIA (best open-source). Multiple LLMs via LiteLLM. | GitHub |
| HuggingFace Open Deep Research | Built in 24 hours. 55% on GAIA with ~1,000 lines of code. Proof you don’t need a massive codebase. | GitHub |
| DeepResearchAgent | Hierarchical multi-agent. GPT-4.1, Gemini, local Qwen via vLLM. Image/video generation in reports. | GitHub |
| STORM (Stanford) | Multiple perspectives, AI “expert” interviews, full Wikipedia-style articles with citations. 70K+ live demo users. Co-STORM collaboration mode. | GitHub |
| Onyx | Deep research across 40+ workplace apps — Slack, Drive, Salesforce, Confluence. RAG + knowledge graphs. | GitHub |
| AI-Researcher | NeurIPS 2025 Spotlight. Autonomous research from ideation to full paper writing. | GitHub |
🔬 Tier 2 — Self-Hosted Research Agents (Your Hardware, Your Rules)
| Agent | What It Does | Link |
|---|---|---|
| Local Deep Research | 10+ sources at once (Google, arXiv, PubMed, Wikipedia + your docs). 95% factual accuracy. Military-grade encryption. | GitHub |
| Automated AI Web Researcher | Fully local Ollama. Auto-generates focus areas, 100+ web searches, compiled reports with citations. | GitHub |
| GPT-Researcher | 5-6 page reports. Multi-agent planning + search + writing. MCP server for Claude/Cursor. Works with Ollama. | GitHub |
| Shandu | Skips Firecrawl API — built-in scraper. LM Studio/Ollama local models. | GitHub |
| Deep-Searcher | Private data + web research using Milvus vector DB. Obsidian integration. Ollama/OpenAI. | GitHub |
| NanoSage | Monte Carlo exploration for depth vs breadth. Gemma 2B summarization. Tavily search. | GitHub |
| dzhng/deep-research | Minimalist (<500 LoC). Recursive research, configurable depth/breadth, markdown reports. | GitHub |
| LangChain Open Deep Research | Official LangChain + LangGraph Studio GUI. Supervisor-researcher multi-agent. | GitHub |
| Argo | One-click Ollama downloads. Offline RAG. Deep research. Cross-platform. | GitHub |
| Local RAG Researcher | LangGraph + DeepSeek R1 + Ollama + Tavily search. Adaptive RAG. | GitHub |
| SurfSense | 20+ service connectors (Slack, Notion, Jira, GitHub, Discord, Gmail, Drive, YouTube). Browser extension. Generates podcasts from research. 150+ AI models. | GitHub — 10k+ stars |
| Scira | “Extreme Search” mode. Academic, YouTube, Reddit modes. Memory for preferences. | GitHub — 11k+ stars |
docker run -d -p 5000:5000 --network host \
--volume 'deep-research:/data' \
localdeepresearch/local-deep-research
SearXNG-Based Private Research
Fully Local Search — Nothing Touches the Cloud
SearXNG is a privacy-respecting metasearch engine you self-host. These bolt AI research on top — zero cloud, zero tracking.
| Tool | What It Does | Link |
|---|---|---|
| OpenDeepResearcher-via-SearXNG | Deep researcher + fully local Ollama inference + chain-of-thought reasoning | GitHub |
| VCRA | Autonomous SearXNG + local LLMs. Semantic duplicate prevention. Multi-format reports. | GitHub |
| Search-Scrape | Rust-native MCP for SearXNG. Intelligent scraping + noise filtering. Qdrant research history. | GitHub |
| Web Research Assistant | MCP server for SearXNG. 13 production tools. GitHub integration + API docs discovery. | GitHub |
| Arsalion AI Agent | Free super agent — Gemini Pro, Groq, LocalAI, SearXNG. Zero subscription. | GitHub |
THE PRIVATE ONES — For the Paranoid (Respect)
Nothing Leaves Your Machine
| Tool | What It Does | Link |
|---|---|---|
| Cognito AI Search | Local AI + private search engine. Conversations cannot leave your machine. True offline after setup. | GitHub |
| WebLLM | Full AI chatbot running entirely IN YOUR BROWSER. No server. No installation. Uses your GPU through WebGPU. Works offline after first load. Just visit webllm.mlc.ai. | GitHub |
| Secret-LLama | Private chatbot in your browser. Llama 3, Mistral. No server needed. Zero install. | GitHub |
docker run -d -p 3000:3000 \
-e OLLAMA_API_URL="http://localhost:11434" \
-e SEARXNG_API_URL="http://localhost:8080" \
kekepower/cognito-ai-search:latest
Academic & Scientific Research
Tools Built for Actual Researchers — Papers, Patents, Systematic Reviews
| Tool | What It Does | Link |
|---|---|---|
| ASReview | Utrecht University. Active learning for systematic reviews. Published in Nature Machine Intelligence. | GitHub |
| LatteReview | Multi-agent systematic literature review. OpenAI, Claude, Groq, Ollama. | GitHub |
| LitLLM | ServiceNow + Mila. Auto-generates literature review sections with citations. | GitHub |
| OpenPaper | Research library workbench. AI assistant for paper reading + citation-grounded responses. | GitHub |
| OpenAlex | 260M+ works from CrossRef/PubMed/arXiv. Full API. CC0 license. Free forever. | openalex.org |
| Lens.org | FREE — patents (100+ jurisdictions) + 272M+ scholarly works with citation linkages. | lens.org |
| ProjectPQ | AI patent search (AT&T/Georgia IP Alliance). Semantic search on 11M+ US patents. | projectpq.ai |
| CiteTrue | AI citation verifier. 15+ databases — CrossRef, arXiv, CORE, PubMed, IEEE, Springer. | citetrue.com |
| SwanRef | AI hallucination detector. Verifies citations against 150M+ papers. | GitHub |
| SR-Accelerator | Bond University. Free SR tools — Polyglot (search translation) + Deduplicator. | sr-accelerator.com |
| RobotReviewer | Cochrane. Semi-automated evidence synthesis + bias assessment. | robotreviewer.net |
| Decipher Research Agent | Topics/links/files → research notebooks + FAQ Generator + Mindmap Creator. | GitHub |
| Radiology Research Agent | LangGraph + SearXNG for radiology evidence retrieval. | GitHub |
MCP Servers — Supercharge Claude/Cursor/VS Code
Plug Deep Research Directly Into Your AI Editor
MCP (Model Context Protocol) lets you give Claude, Cursor, or VS Code direct access to research tools. Install one of these and your AI assistant can search the web, read papers, and write reports without you leaving the editor.
| Server | What It Does | Link |
|---|---|---|
| Claude-Deep-Research | Web + academic search for Claude Desktop. | GitHub |
| mcp-DEEPwebresearch | Deep web research with intelligent search queuing + content extraction. | GitHub |
| MCP Deep Research | Full workflow — question elaboration, subquestion generation, report synthesis. | GitHub |
| deep-research-mcp | Research assistant generating reports with citations as MCP tool. | GitHub |
| PubMed MCP | Biomedical literature research via NCBI E-utilities. | GitHub |
| Contextual MCP | RAG agent for Cursor/Claude with document grounding + citations. | GitHub |
| Research Powerpack MCP | Web crawling, Reddit mining, URL scraping, dynamic token budgeting. | GitHub |
| Finance Deep Research MCP | Investment report generation with deep research. | GitHub |
Browser Extensions & Automation
AI Research From Your Browser — No Terminal Required
| Tool | What It Does | Link |
|---|---|---|
| NanoBrowser | Multi-agent AI web automation with Ollama/Groq. Free alternative to OpenAI Operator. | GitHub |
| BrowserBee | “Cline for web browsing” — deep research on companies, jobs, academic publications. Playwright. | GitHub |
| NativeMindExtension | Fully private on-device AI running local LLMs via Ollama in your browser. | GitHub |
| extensionOS | Access any LLM from any website via right-click. Mixture of Agents feature. | GitHub |
| CatalyzeX | Find implementation code for papers on ArXiv, Scholar, PubMed, IEEE. Firefox extension. | Firefox Add-on |
Telegram & Discord Research Bots
Research Agents in Your Chat Apps
| Bot | What It Does | Link |
|---|---|---|
| TeleChat | Multi-LLM Telegram bot with web search, ArXiv summarization, URL summarization. | GitHub |
| TG Research Agent | LangGraph-powered Telegram agent. Multi-threaded web search + reviewer agent. | GitHub |
| ArXivBot | Telegram bot for searching/sharing arXiv papers. | GitHub |
| Daily AI Paper Summary | Daily AI paper summaries from arXiv straight to Telegram. | GitHub |
Multimodal — Video, Audio, Image Research
Research Beyond Text
| Tool | What It Does | Link |
|---|---|---|
| YT-Navigator | AI YouTube explorer. Semantic search + chat. Extracts insights with timestamps. | GitHub |
| VideoSeek | Find specific moments in YouTube videos using natural language + Whisper. | GitHub |
| Transcribee | macOS transcriber for YouTube/TikTok/Instagram → self-organizing searchable knowledge base. | GitHub |
| Scriberr | Self-hosted offline audio transcription. Speaker diarization. LLM chat. Summaries. | GitHub |
| Muse Podcast Search | Free podcast search — full transcripts, speaker diarization, 17-language translation. | GitHub |
| Video Analyzer | Llama 11B vision + Whisper for frame-by-frame video analysis with descriptions. | GitHub |
| rclip | AI CLI photo search using CLIP. Text-to-image search in local libraries. | GitHub |
| CLIPPyX | System-wide AI search — content-based, text, and visual similarity using CLIP + OCR. | GitHub |
OSINT & Investigative Research
Open Source Intelligence Tools
| Tool | What It Does | Link |
|---|---|---|
| Taranis AI | Advanced OSINT using AI/NLP for intel from websites, social media, RSS. Structured reports. | GitHub |
| OSINTGPT | GPT-powered embeddings + vector search for document similarity analysis. | GitHub |
| Aleph | OCCRP’s cross-referencing search for investigative reporting. 400M+ documents. | GitHub |
| OSINT Toolkit | Self-hostable analysis — email analyzer, domain finder, AI templates. | GitHub |
| MetaOSINT | Meta-tool to quickly identify relevant OSINT tools by category. | metaosint.github.io |
Competitive Intelligence & Market Research
Financial Analysis, Trading Agents, Competitor Research
| Tool | What It Does | Link |
|---|---|---|
| FinRobot | AI agent for financial analysis. Chain-of-Thought equity research + valuation. | GitHub |
| TradingAgents | Multi-agent trading — fundamental analysts, sentiment experts, technical analysts. | GitHub |
| Comperator | AI competitor analysis — crawling, extracting, classifying, comparing competitor websites. | GitHub |
| Market-Research-Agent | Multi-agent financial/market analysis. CrewAI + Gemini. Company reports. | GitHub |
Knowledge Graphs & RAG Systems
Build Your Own Searchable Brain
| Tool | What It Does | Link |
|---|---|---|
| Graphiti | Real-time knowledge graphs for AI agents. Temporal data + historical context. | GitHub |
| AI Knowledge Graph | Entity extraction, relationship inference, visualization. | GitHub |
| GraphRAG Workbench | 3D interactive knowledge graph visualization using Microsoft GraphRAG. | GitHub |
| RAGFlow | Leading RAG engine. Deep document understanding. GraphRAG support. Ollama integration. | GitHub |
| SiYuan | Privacy-first self-hosted personal knowledge management. Block-level references. | GitHub |
| Casibase | Enterprise AI knowledge base. MCP protocol. ChatGPT, Claude, Ollama support. | GitHub |
Local LLM Research Assistants
Ollama-Powered Research — No Cloud, No API Keys
| Tool | What It Does | Link |
|---|---|---|
| DeepSearch (Rust) | Rust CLI. Decomposes queries into sub-queries. Wikipedia + DuckDuckGo. | GitHub |
| Ollama Agents | Agent framework with JSON graph knowledgebase, web search, fact-checking. | GitHub |
| llama-cpp-agent | Agent framework for llama.cpp. RAG, function calling, structured output, agent chains. | GitHub |
| aichat | All-in-one CLI — Shell Assistant, Chat-REPL, RAG, AI tools & agents. Ollama support. | GitHub |
| Insights-LM | Private NotebookLM alternative. Ollama. Document chat. Audio summaries. | GitHub |
| Ollama-RAG | Customizable RAG with static PDF memory + dynamic web search conversation. | GitHub |
| Open Notebook | Full NotebookLM alternative. 16+ AI providers. Podcast generation. Web ingestion. | GitHub |
Chinese & Non-English Community Tools
From Alibaba, Baidu, Tsinghua, and the EU
| Tool | What It Does | Link |
|---|---|---|
| Qwen-Agent | Alibaba. Function Calling, MCP, Code Interpreter, RAG, Chrome extension. | GitHub |
| ERNIE | Baidu. ERNIE 4.5 series (10 models up to 424B params) + ERNIEKit toolkit. | GitHub |
| Paper2GUI | Converts AI papers to desktop GUI apps for non-programmers. 50+ AI models. | GitHub |
| ExtAgents | Tsinghua multi-agent framework for scaling external knowledge beyond LLM context. | GitHub |
| ProactiveAgent | LLM agent that proactively predicts user tasks without explicit requests. | GitHub |
| AI Productivity Tool | Local LLM desktop with DeepSeek/Phi/Qwen for batch media processing. | Gitee |
| OSAI Index | European Open Source AI Index — community-driven openness evaluation. | osai-index.eu |
BUILD YOUR OWN PIPELINE — Web Crawling Tools
Turn Any Website Into AI Food
| Tool | What It Does | Link |
|---|---|---|
| Crawl4AI | #1 trending GitHub repo. 50k+ stars. 6x faster than paid alternatives. No API keys. Deep crawling with smart strategies. | GitHub |
| Firecrawl | URL → crawls whole site → clean data. Self-hostable. | GitHub |
| LLM-Reader | Webpages → LLM-friendly text. Open-source Firecrawl alternative. Fully free. | GitHub |
| AI Research Agent (STREAM) | Wikipedia topic phrase graphs + SEEKTOPIC keyphrase extraction. | GitHub |
| Media Agent | Scrapes Twitter/Reddit, indexes in ChromaDB, chat with social media content. | GitHub |
pip install -U crawl4ai
crawl4ai-setup
SINGLE-FILE SCRIPTS — Hacky But They Work
One Python File = Web Search + AI
| Tool | What It Does | Link |
|---|---|---|
| Ollama Internet Search | One Python file. Adds web search to any Ollama model. DuckDuckGo. | GitHub |
| Web-LLM-Assistant | Single file. Self-improving search queries. Any Ollama model. | GitHub |
| Ollama-Search-Agent | REACT-style agent. Human-in-the-loop — you approve each search. | GitHub |
| AgentSearch | Ollama + Firefox + DuckDuckGo learning agent. Multiple search queries per question. | GitHub |
FREE API SOURCES — When You Need Cloud Power
Free Tiers That Actually Work
| Provider | What You Get | Link |
|---|---|---|
| Groq | Very generous free tier, stupid fast | groq.com |
| GitHub Models | Free with Personal Access Token, GPT-4/Claude access | github.com/marketplace/models |
| OpenRouter | :free models available |
openrouter.ai |
| Google AI Studio | Generous Gemini access | aistudio.google.com |
| SambaNova | Free tier, fast Llama | sambanova.ai |
| Cloudflare Workers AI | 10k requests/day free | cloudflare.com |
| Flexpilot | Open-source Copilot alternative using your own API keys | GitHub |
Full list: github.com/cheahjs/free-llm-api-resources
Curated Lists & Discovery Resources
Rabbit Holes to Fall Into
| List | What It Covers | Link |
|---|---|---|
| Awesome AI Web Search | Timeline of AI web search software including obscure projects | GitHub — 1.2k stars |
| Awesome Deep Research | Up-to-date agentic deep research resources — papers, tools, benchmarks | GitHub |
| Deep Research Agent Survey | Survey + collection of deep research agent papers | GitHub |
| Awesome AI for Science | 600+ scientific AI tools (BioDiscoveryAgent, ChemCrow, MOOSE) | GitHub |
| 500 AI Agent Projects | 500 use cases for research, finance, data interpretation | GitHub |
| Awesome MCP Servers | 100+ MCP servers including research tools | GitHub |
| Awesome Private AI | Private AI resources for self-hosted + air-gapped setups | GitHub |
| AlgorithmAudit BDT | Stanford AI Audit finalist. Unsupervised bias detection (GDPR-friendly). | algorithmaudit.eu |
QUICK START — Pick One
| Your Situation | Best Choice | Why |
|---|---|---|
| “I have 2 minutes” | WebLLM in browser | Zero install, just visit the site |
| “I have 5 minutes” | Perplexica one-liner | One Docker command and done |
| “I want the best” | Local Deep Research + SearXNG + Ollama | The full local stack |
| “I’m paranoid” | Cognito AI Search | Nothing leaves your machine |
| “I need serious research” | Auto-Deep-Research or STORM | Lab-quality agents |
| “I’m an academic” | ASReview + OpenAlex + CiteTrue | Built for real papers |
| “I use Claude/Cursor” | Any MCP server above | Plug research into your editor |
| “I have no tech skills” | Open WebUI + Ollama | Beautiful UI, works out of the box |
THE FULL ARSENAL — 124 Tools At a Glance
Master Comparison Table
| Tool | Category | Difficulty | API Keys? | Best For |
|---|---|---|---|---|
| Perplexica | Search | Easy | None | Daily searching |
| Open WebUI | Search | Easy | Optional | Swiss army knife |
| LLocalSearch | Search | Easy | None | Zero-config research |
| MindSearch | Search | Easy | None (DDG) | Multi-source research |
| Khoj | Search | Medium | Optional | Personal knowledge base |
| WebThinker | Deep Research | Advanced | Some | NeurIPS-grade research |
| AgentCPM | Deep Research | Medium | None | Consumer hardware deep research |
| MiroThinker | Deep Research | Advanced | Some | Highest GAIA scores |
| DeerFlow | Deep Research | Medium | Optional | Multi-agent reports |
| Auto-Deep-Research | Deep Research | Easy | Optional | One-click deep research |
| STORM | Deep Research | Medium | Some | Writing full articles |
| Local Deep Research | Self-Hosted | Medium | None | Academic/deep research |
| GPT-Researcher | Self-Hosted | Medium | Some | Long reports |
| SurfSense | Self-Hosted | Complex | Some | All-in-one workspace |
| Cognito AI | Privacy | Easy | None | Maximum privacy |
| WebLLM | Privacy | None! | None | Browser-only AI |
| ASReview | Academic | Medium | None | Systematic reviews |
| OpenAlex | Academic | Easy | None | 260M+ papers free |
| CiteTrue | Academic | Easy | None | Citation verification |
| Crawl4AI | Pipeline | Easy | None | Building pipelines |
| RAGFlow | Knowledge | Medium | Optional | Enterprise RAG |
| FinRobot | Market | Medium | Some | Financial analysis |
| Taranis AI | OSINT | Medium | Optional | Intelligence gathering |
| Aleph | OSINT | Complex | None | Investigative reporting |
124 tools. All free. Most people will never know they exist.
Now you do. 
!