Module 09Research Tools
§ why this exists
The AI tools landscape moves fast. By the time you read this, some tools listed here will have changed names, merged, or disappeared. That is expected. This directory is not a consumer guide — it is a curated map of the tool categories that matter for academic research, with representative examples in each.
Every tool listed here is either free, open-source, or has a free tier sufficient for academic use. No tool on this list requires a paid subscription to evaluate. Tools marked [oss] are open-source; tools marked [free] have workable free tiers.
apply the four questions
1. What is being delegated? You are delegating tool discovery, not tool selection. This list is a starting point, not a recommendation.
2. What would make a tool choice believable? Test it on your own task, with your own data, against an explicit criterion. Do not adopt a tool because it appears on a curated list.
3. What could be lost? Time spent evaluating tools that don't fit your workflow. The research question should drive tool choice, not the reverse.
4. What must remain human? The decision of which tool to use, under what data-governance constraints, and with what verification.
Ⅰ · coding agents & terminal harnesses
AI tools that write, edit, and explain code — from your terminal, editor, or browser.
Anthropic's terminal-native coding agent. Deep codebase understanding, git-aware, with Model Context Protocol (MCP) for extensible tool use. Free tier via Claude Max subscription; also available via OpenRouter.
OpenAI's lightweight open-source coding agent. Cloud sandboxed execution, git worktree isolation. Runs via OpenAI OAuth or API key. The open-source reference implementation for terminal coding agents.
Open-source terminal AI agent from SST with 150K+ GitHub stars. Supports 75+ model providers. Free, privacy-first, with LSP integration. GitHub Copilot auth partnership for free model access.
Terminal-based AI pair programming tool. Every AI edit is auto-committed to git. Model-agnostic — works with Claude, GPT, Gemini, and local models. Strong for iterative editing of existing codebases.
Google's open-source terminal agent. Free tier includes Gemini 2.5 Pro access with 1M token context window. Being succeeded by Antigravity CLI but remains functional and widely used.
AI-first VS Code fork. Multi-file edits, agent mode, codebase-wide chat. Currently the category leader for AI-augmented IDEs. Free tier available; Pro subscription for extended usage.
Autonomous coding agent that runs inside VS Code. Can create/edit files, run commands, browse the web, and use MCP servers. Open-source. Uses your own API keys.
Cline fork with multi-mode agent loop (code, architect, debug). VS Code extension. Supports configurable modes for different task types.
Open-source IDE assistant for VS Code and JetBrains. Self-hostable. Configurable models. Good for researchers who want an AI coding assistant without vendor lock-in.
Ⅱ · agentic systems & orchestrators
Systems that coordinate multiple AI agents, manage workflows, or operate as autonomous research assistants.
Self-improving AI agent by Nous Research. Built-in learning loop — creates skills from experience, improves them during use, searches past conversations. Runs on any LLM provider. Connects via Telegram, Slack, Discord, WhatsApp, Signal, and CLI. Scheduled automations, subagent parallelism. Research-ready with batch trajectory generation.
Open-source platform for AI software developers as generalist agents. Docker-sandboxed execution, multi-agent delegation, evaluation framework. 64K+ GitHub stars. Strong for prototyping agent workflows in a safe environment.
Open-source software factory by LangChain. Turns engineering work into repeatable systems: investigate, implement, validate, deliver PRs. Reviews code, monitors CI, responds to feedback. Deployable in your own infrastructure.
Cognition's autonomous AI software engineer. Browser, shell, and editor in one sandbox. Runs tasks asynchronously. Commercial product; academic pricing may be available.
Role-based multi-agent framework. Define agents with roles, goals, and tools; orchestrate them as a crew. Python library. Useful for designing research workflows with specialized AI roles.
Microsoft's conversational multi-agent framework. Agents can chat with each other, use tools, and incorporate human feedback. Strong for designing collaborative research assistant workflows.
Stateful multi-agent graph framework from LangChain. Durable execution, thread state, checkpointing. Good foundation for building research pipelines that need reliable agent state management.
Ⅲ · model access & registries
Where to find, access, and run language models — from cloud APIs to local hardware.
The primary repository for open-weight models. 500K+ models, datasets, and demos. Free model hosting and inference API. Essential for discovering and accessing research models.
Unified API for 300+ models across providers. Single API key accesses OpenAI, Anthropic, Google, Meta, DeepSeek, and dozens more. Free tier available. Good for comparing models without managing multiple accounts.
Run LLMs locally on your own hardware. Simple CLI: `ollama run llama3.2`. Supports Llama, Mistral, Qwen, DeepSeek, Gemma, and more. No data leaves your machine. Essential for working with sensitive research data.
Desktop application for running LLMs locally. Browse HuggingFace models, download, and run with a chat interface. No terminal required. Good for researchers who want a GUI for local model experimentation.
Cloud API for running open-source models. Pay-per-inference. Cog framework for packaging models. Good for running models that won't fit on your hardware without managing cloud infrastructure.
Unified gateway for 100+ LLM providers with an OpenAI-compatible API. Self-hostable. Load balancing, fallbacks, spend tracking. Useful for research projects that need provider-agnostic model access.
C/C++ inference for LLMs in GGUF format. Runs on CPU, GPU (CUDA, Metal, Vulkan). The engine behind Ollama and most local LLM tools. Essential for researchers who need to run models on consumer hardware.
Ⅳ · literature, writing & academic tools
AI-powered tools for literature search, synthesis, writing, and academic workflow.
Open catalog of 250M+ scholarly works. Free REST API with no rate limits for academic use. Replaces Microsoft Academic Graph. Essential for programmatic literature search. Used in Module 3 of this course.
AI-powered academic search engine. 200M+ papers with citation graph, TLDR summaries, and influence scores. Free API with generous rate limits. Strong for citation network exploration.
Open-source reference manager. Browser connector for one-click paper saving. PDF annotation, cloud sync (free up to 300MB). MCP server available for AI agent integration. The standard for academic reference management.
Open-access preprint repository. 2.5M+ papers across physics, math, CS, econ, stats, and more. Free API for programmatic access. RSS feeds for new submissions in your field.
Visual citation graph explorer. Enter a paper, get a graph of related work with similarity-based layout. Good for getting oriented in a new field. Free tier with limited graphs per month.
AI research assistant. Ask a research question, get a table of relevant papers with extracted findings, methods, and outcomes. Free tier available. Good for rapid literature scoping — but verify every extracted claim.
Online LaTeX editor with real-time collaboration. Git integration. Many journals provide direct Overleaf submission. Free plan available.
Local-first markdown knowledge base. Plain text files on your filesystem — no vendor lock-in. Graph view, backlinks, plugin ecosystem. Good for research notes, literature synthesis, and project management.
Ⅴ · research infrastructure & evaluation
Sandboxes, evaluation frameworks, and infrastructure for reliable AI-assisted research.
Open standard for connecting AI agents to external tools and data sources. Supported by Claude Code, Codex CLI, Cursor, OpenCode, and others. Growing ecosystem of MCP servers for filesystems, databases, APIs, and research tools.
Experiment tracking and model evaluation platform. Log hyperparameters, metrics, artifacts, and model checkpoints. Free tier for individuals. Useful for recording LLM evaluation runs with provenance.
Standard framework for evaluating language models. 200+ benchmark tasks (MMLU, GSM8K, HellaSwag, etc.). Used by most model release papers. Essential if you need to benchmark a model for research claims.
Open-source browser automation for AI agents. Lets LLMs navigate websites, fill forms, extract data. Good for web-based research tasks like systematic data collection from online sources.
Convert any URL to clean, LLM-readable markdown. Prefix any URL with `r.jina.ai/` to get extracted text. Free tier. Useful for feeding web content into AI tools for analysis.
Turn websites into LLM-ready data. Crawls pages, extracts clean markdown, handles JavaScript rendering. Free tier available. Good for systematic web data collection for research.
Data framework for LLM applications. Connect models to your data — documents, databases, APIs. Good for building retrieval-augmented generation (RAG) pipelines over research corpora.
Microsoft's tool for converting files to markdown. Handles PDF, Word, Excel, PowerPoint, images, audio, HTML, CSV, JSON, XML, ZIP. Good for preparing documents for AI processing.
Container platform for reproducible environments. Run AI tools in isolation. Essential for ensuring that your research code runs identically on different machines. Many AI coding tools run inside Docker sandboxs.