Discover AI agents built for Research: Comprehensive directory of solutions with open-source code verified by AI agency professionals focused on the Research sector.

Open-Source AI Agents for Research | AI Agent Store

Agent S is an open-source framework for building and running “computer use” GUI agents that can autonomously operate a computer through an Agent-Computer Interface. It’s designed to let an LLM plan and execute multi-step tasks by observing the screen and taking actions (e.g., clicking, typing, navigating) on supported platforms (Linux, macOS, Windows). The project provides a CLI to run the agent, supports multiple model providers (including OpenAI, Anthropic, Gemini, OpenRouter, and vLLM), and recommends pairing a main LLM with a separate grounding model for UI understanding. Agent S also includes evaluation assets and reports results on established computer-use benchmarks such as OSWorld.

Agent S

agent-s

Open-source GUI agent framework that lets an LLM use your computer like a human via an Agent-Computer Interface.

Desktop Commander MCP is an open-source Model Context Protocol (MCP) server that gives AI clients controlled access to your local development environment. It enables running terminal commands with streaming output and interactive process/session management, plus full filesystem operations like reading/writing files, creating directories, and recursive search. It includes diff-based “surgical” edits and pattern-based replacements, powered by ripgrep for fast repo-wide search. The tool also adds native Excel support (read/write/edit/search) and PDF tooling (extract text, create PDFs from markdown, and modify PDFs), along with comprehensive audit logging of tool calls for traceability.

Desktop Commander MCP

desktop-commander-mcp

MCP server that lets AI run terminal commands, manage processes, and search/edit files (diff-based) with Excel/PDF support and audit logs.

FLAMEHAVEN FileSearch is an open-source, self-hosted RAG document search engine you can run locally (Docker or Python). It supports multi-format ingestion (PDF, DOCX, TXT, MD) and provides keyword, semantic, and hybrid search modes (with typo correction). Results include source attribution, and the API includes production features like API-key authentication with permissions, rate limiting, audit logging, batch search, and an admin dashboard. It integrates with Google Gemini (via GEMINI_API_KEY) and supports a lightweight local store plus optional components such as Redis caching and Postgres/pgvector for higher-performance deployments.

FLAMEHAVEN FileSearch

flamehaven-filesearch

Self-hosted RAG file search engine (FastAPI) with keyword/semantic/hybrid modes, citations, API keys, and Docker quickstart.

Gobii is a platform for creating “digital workers” (always-on browser agents) that can perform repetitive web-based tasks automatically. You can create custom workers or spawn pretrained ones, run them on schedules, and connect them to workflows through triggers and APIs. For developers, Gobii provides a cloud browser automation stack (persistent agents/sessions, webhooks, and integrations) and also supports self-hosting through its open-source codebase for teams that want more control over infrastructure and data residency.

Gobii

gobii

Always-on browser workers that automate web tasks via schedules, triggers, and API—run in the cloud or self-host open source.

Together Open Data Scientist is an open-source AI data analysis agent built around the ReAct (Reasoning + Acting) pattern. It automates common end-to-end data science workflows: loading and exploring datasets, running EDA, engineering features, training baseline models, creating plots, and writing a structured report. The agent can execute Python code either locally (via Docker/internal execution) or in the cloud using Together Code Interpreter (TCI), and it supports uploading a data directory for analysis. It’s positioned as experimental software—outputs should be reviewed and validated—making it most useful for exploration, learning, and fast first-pass analysis rather than unattended production analytics.

Together Open Data Scientist

together-open-data-scientist

Open-source data scientist agent that runs Python (locally or in the cloud) to explore datasets, build models, and generate detailed analysis reports using the ReAct framework.

Kilo Code is an open-source AI coding assistant designed to run inside VS Code, JetBrains IDEs, and web-based editors. Built as a full coding agent, it can plan features, generate new code from natural language, refactor and debug existing projects, run terminal commands, and even automate the browser with user approval. Kilo Code combines ideas from tools like Cline and Roo into a unified workflow with dedicated modes such as Architect, Coder, and Debugger. It connects to 400+ frontier and open models through OpenRouter and similar providers, offers optional hosted credits so developers can start without managing their own API keys, and supports an MCP-style server marketplace for extending the agent with custom tools. This makes it a flexible choice for individual developers and teams who want powerful AI assistance while keeping configuration transparent and community-driven.

Kilo Code

kilo-code

Open-source AI coding assistant for VS Code and JetBrains that plans, writes, and debugs code with multi-model support.

ComfyUI is a free, open-source, node-based platform for generative AI. Creators assemble visual workflows to produce images, video, 3D, and audio, extend functionality with custom nodes, and run everything locally for speed, control, and cost savings. The Comfy Org ecosystem includes a desktop app, a registry for custom nodes, continuous-integration testing, and an announced Comfy Cloud beta to run workflows in the browser without setup.

ComfyUI (Comfy.org)

comfyui-comfy-org

Open-source, node-based generative AI platform to build image, video, 3D, and audio workflows locally or in the cloud.

Wan2.2 is an open-weights family of large-scale video generative models released by the Wan-Video team as the successor to Wan2.1. It introduces expanded training data and architecture upgrades, offering multiple variants for text-to-video, image-to-video, video-to-video, and even speech-to-video generation. The models produce cinematic, higher-resolution clips with stronger temporal consistency and motion control, and are available under an Apache-2.0 license with runnable repos and Hugging Face checkpoints.

Wan2.2

wan2-2

Open-source video generation model family (5B/14B) for text/image/video-to-video, delivering 1080p clips with improved motion and control.

Cell2Sentence (C2S) is an open-source framework from the van Dijk Lab that represents single-cell RNA-seq profiles as textual “cell sentences” (genes ordered by expression), enabling large language models to natively model biology. The project includes code, docs, and open models (C2S-Scale) released with Google Research/DeepMind for state-of-the-art single-cell tasks like cell typing, perturbation prediction, and virtual cell generation.

Cell2Sentence

cell2sentence

Open-source framework that turns single-cell gene expression into 'cell sentences' so LLMs can analyze and generate biology insights.

GLM‑4.5 is an open‑source, Mixture‑of‑Experts (MoE) large‑language model from Z.ai, designed for intelligent agent applications. It features 355 B parameters (32 B active) with an efficient ’thinking/non‑thinking’ hybrid reasoning mode and 128 K context length. A smaller variant, GLM‑4.5‑Air, offers 106 B parameters (12 B active) for more efficient deployments. GLM‑4.5 excels at reasoning, coding, and tool‑calling tasks and ranks among the top performers on benchmarks compared with models like o3, Grok‑4, Claude Sonnet, and GPT‑4.1.

GLM‑4.5

glm-4-5

Open-source hybrid‑reasoning MoE foundation model optimized for intelligent agent tasks with 128K context and tool use.

TEN Framework is an open-source modular system for building autonomous AI agents capable of reasoning, memory management, and tool usage. Designed to reflect real-world constraints, it incorporates a Thought-Emission-Norm (TEN) architecture, encouraging agents to think before acting. The framework supports open-ended planning, persistent memory storage, and tool augmentation. Developers can use TEN to experiment with agent autonomy and deploy agents that simulate complex decision-making workflows.

TEN Framework

ten-framework

An open-source framework for building autonomous AI agents that interact with the real world through tools and memory.

Trae Agent is an open-source, LLM-powered command-line assistant developed by ByteDance for software development workflows. It supports multi-LLM providers, file editing, bash execution, tool integration, trajectory logging, and a modular architecture—ideal for debugging, patching, documentation, and research.

Trae Agent

trae-agent

Open-source CLI LLM agent for automating software engineering tasks via natural-language commands.

Open-Source AI Agents for "Research" Industry