// Agent profile
OpenResearch is a local-first research workspace from alphaXiv that turns existing coding agents — Claude Code, Codex, OpenCode, Cursor, Google Antigravity — into autonomous research agents. It provides parallel agent sessions with isolated git worktrees, an immutable experiment tree for reproducibility, and evidence preservation that ties logs, diffs, and results to every piece of work. The framework supports remote execution across SSH, Slurm, Kubernetes, Ray, and Modal, while keeping all data local by default. With 6.3K GitHub stars and an MIT license, OpenResearch fills the gap between ad-hoc research scripts and full lab infrastructure.
OpenResearch is alphaXiv’s open-source framework for turning coding agents into research agents. Built in Rust, it provides the infrastructure layer that lets AI agents propose hypotheses, modify code, run experiments, and decide next steps autonomously — all while maintaining a reproducible, git-native record of every step. Rather than building a new agent from scratch, OpenResearch wraps existing coding agents (Claude Code, Codex, OpenCode, Cursor, Antigravity) and gives them a structured research environment with parallel session management and experiment tracking.
The core of OpenResearch is its experiment tree — a git-native structure where every agent session runs in an isolated worktree, and every result is preserved with full provenance (logs, code diffs, output files). This means research is reproducible by default, not as an afterthought. Agents can explore multiple research directions simultaneously in parallel sessions, with the framework managing branching and merging of successful lines of inquiry. For compute-heavy workloads, OpenResearch supports remote execution across SSH hosts, Slurm clusters, Kubernetes, Ray, and Modal — while keeping the orchestration and data local.
Researchers use OpenResearch for autonomous literature review, hypothesis generation, and systematic experimentation loops. ML engineers run overnight experiment sweeps where agents propose architecture modifications, train models, evaluate results, and branch into promising directions. The framework is also used for systematic code analysis, benchmark evaluation, and any workflow where multiple parallel investigations need to be tracked and compared with full reproducibility.
OpenResearch is a CLI-first tool that assumes familiarity with git workflows and terminal-based coding agents. It requires an existing coding agent (Claude Code, Codex, etc.) with its own API credentials — OpenResearch is the orchestration layer, not the AI model. The managed compute features via openresearch.sh are optional and may involve additional costs beyond the free open-source core.
OpenResearch is built for ML researchers, scientists, and engineers who want to scale their research with AI agents while maintaining rigorous reproducibility. It appeals to teams already using coding agents who want structured experiment management, parallel exploration, and the confidence that every result can be traced back to its exact conditions.
AI agents that autonomously conduct ML research experiments on single-GPU training, running a keep-or-revert loop overnight.
Stealth headless browser for AI agents — wraps a Firefox fork with C++-level fingerprint spoofing to bypass Cloudflare, bot detection, and anti-scraping.
An AI-powered academic search engine that finds and synthesizes evidence-based answers from peer-reviewed scientific research.