I recently pulled down both EvoScientist and OmniScientist and read through the source code. Both come with their own arXiv technical reports and both claim end-to-end scientific research capability. But after reading the source, it's clear they are not solving the same problem at all. One is about making AI act more like a seasoned researcher over time; the other is about stopping AI from fabricating results over second-hand tables.
What follows is my comparison based on both READMEs and source code — not an official write-up, with some of my own judgment mixed in.
EvoScientist: A Research Partner That Evolves Itself
EvoScientist (v0.2.8, Apache 2.0) is built on LangChain's DeepAgents framework. Its core selling point is "self-evolution": after each research run, the system records observations into a knowledge graph whose edges are labeled as complementary, contradictory, or superseding. Even more aggressive is the AutoSkills mechanism, which periodically mines recurring patterns from its own memory, drafts them into reusable skills, and suspends them pending review via /autoskills. In other words, the system writes its own tools.
Supporting this are 6 sub-agents — planning, research, coding, debugging, data analysis, and writing — plus an orchestrator. Humans are placed "on the loop" rather than "in the loop": by default every shell execution requires approval, and agents proactively ask when uncertain. It integrates Telegram, Feishu, WeChat, Slack and other channels, with WebUI, TUI, and CLI interfaces, even cron-style scheduled tasks via /schedule. Maturity is high: it ranked first in both versions of the DeepResearch Bench and won Best Paper at ICAIS 2025.
OmniScientist: A Judge-Style System That Reads Raw Data Directly
OmniScientist (v0.1.1, MIT) is driven by a completely different obsession. Its README opens by noting that most agents reason only over text and code — raw observations like microscopy images, seismic waveforms, or CT volumes must first be compressed into tables by humans before being fed in. It aims to let agents read the raw material itself.
Implementation-wise, it is a self-contained ReAct loop with three stages: Ideation (hypothesis) → Experiment (run code, read images) → Writeup (write the paper), each stage gated by a checkpoint. That gate is code-driven and reads execution records, not the draft: every number appearing in the manuscript must have actually appeared in some run's stdout to pass. Empty experimental results send the work back to the ideation stage, with at most two rollbacks, and it remembers the previous failed approach. References are checked in real time against OpenAlex and Crossref — if not found, nothing is fabricated, which distinguishes it from many "AI paper writing" projects.
It supports 12 modalities spanning images, waveforms, audio, video, point clouds, trajectories, tables, and formulas, with sample cases in seismology, chemistry, genomics, astronomy, pathology, CT, superconductivity, and birdsong. The desktop app strips credentials out of the environment before handing code execution to the agent — a detail that shows real security awareness.
How Similar Are They?
A quick checklist:
| Dimension | EvoScientist | OmniScientist | |---|---|---| | Primary goal | Continuous cross-session evolution | Raw-data perception with strict traceability | | Agent structure | 6 collaborating sub-agents | Single ReAct loop, three stages | | Multimodality | Weak, text/code-centric | Strong, 12 native modalities | | Verification | Human approval + ask_user | Code gates + stdout number back-linking | | Domain coverage | General research assistant | Multi-disciplinary STEM samples | | Deployment | Python stack + WebUI | Desktop App / CLI, local-first | | Maturity | v0.2.8, benchmark first place | v0.1.1, early stage | | License | Apache 2.0 | MIT |
They share quite a bit too: both can produce an end-to-end paper with figures and tables; both insist the agent runs its own analysis code and reads back results rather than memorizing; both place humans at critical points — one via approvals, the other via gates; and both picked the most permissive licenses.
Their Respective Weaknesses
EvoScientist's "self-evolution" sounds impressive, but its working style is text/code-centric. It cannot "see" an electron micrograph or a seismic trace — raw observations still need to be converted into something it can read. Its hallucination defense relies mainly on humans: approvals, dangerous-mode flags, auto-allow lists. The system itself lacks OmniScientist's hard constraint that "numbers must come from real stdout." Also, the full stack requires a langgraph dev backend, and the WebUI needs Node 24 — deployment is not lightweight.
OmniScientist, conversely, has strong verification and perception, but it follows a single-task, single-run mindset with no cross-session memory or skill accumulation, and a much smaller ecosystem: few built-in skills, no messaging channels, no scheduling. v0.1.1 means interfaces will still change, and Windows lacks real-machine feedback. Expecting it to "understand you better over time" is not realistic yet.
My Take
These two are not competitors — they are two pieces of the same puzzle.
If you want a long-term research partner that accumulates experience, builds its own tools, and can take tasks through messaging channels, EvoScientist is usable today and works smoothly. It bets that the process gets more familiar with each run.
If you want to throw a specific dataset in and get a paper where every number is grounded in actual observation of raw data, OmniScientist's design fits better. It bets that "seeing reality, no fabrication" is more fundamental than "evolution."
One personal concern: both are still stuck in a "building a god" narrative. EvoScientist tells a rousing story of self-evolution, but whether cross-session memory becomes a self-reinforcing echo chamber is unverified. However hard OmniScientist's gates are, the hypotheses and directions are still chosen by humans. Getting to genuine paper-writing assistance still requires a chunk of interpretability and accountability engineering.
That's it. Both repos are on GitHub — links and technical reports below. Reading the source yourself is faster than reading this post.
References
- EvoScientist: https://github.com/EvoScientist/EvoScientist — technical report arXiv:2603.08127
- OmniScientist: https://github.com/Omni-Scientist/OmniScientist — technical report arXiv:2608.13558