OpenClaw: What is the Open-Source AI Agent that is Taking the Web by Storm?
In just a few months, OpenClaw has gone from being a confidential GitHub project to a global phenomenon with 400,000 users. But is this general-purpose AI agent suitable for scientific research? Comparison withCharlie, the specialized agent for biomedical research.
Emerit Science Team
At the end of 2025, a GitHub project called Clawdbot—quickly renamed OpenClaw—captivated the global tech community. Developed by Peter Steinberger, this open-source AI agent surpassed 214,000 GitHub stars and attracted between 300,000 and 400,000 active users in just a few months. For many, it represents a breakthrough: a powerful, free AI agent that runs locally and integrates with everyday messaging apps.
But behind all the hype, one question remains: Is OpenClaw really suited to demanding scientific use? And how does Charlie, Emerit Science's AI agent specialized in biomedical research, meet fundamentally different needs? This article offers an honest analysis of both approaches.
What is OpenClaw?
OpenClaw is an open-source autonomous AI agent, launched in November 2025 under the name Clawdbot by Peter Steinberger. The principle is simple but powerful: an agent that runs locally on your machine, compatible with the major language models on the market (Claude from Anthropic, DeepSeek, GPT from OpenAI), and which acts through the applications you already use on a daily basis.
What OpenClaw can do:
- Access your emails and calendars to plan, respond, and organize
- Integration with Signal, Telegram, Discord, WhatsApp
- Independent web browsing and information searching
- Running tasks on your local file system
- Multi-LLM compatibility: Claude, DeepSeek, GPT-4o, and others
Its meteoric rise in popularity can be explained by its local-first philosophy: unlike cloud services, the data processed remains on your machine. It is also a vibrant community project, with a global base of contributors.
How does OpenClaw work?
OpenClaw is based on an autonomous agent architecture: it receives a natural language objective, breaks it down into steps, and then executes them in sequence using the available tools (email, messaging, web, files). The user chooses their preferred LLM—Claude 3.5 Sonnet, DeepSeek R1, GPT-4o—as the reasoning brain. The agent orchestrates the actions.
In practice, you can say to it: "Check my unread emails, summarize the three most important ones, and block out two hours in my calendar to respond to them." OpenClaw will execute this sequence without intervention. This operational autonomy in the general personal and professional sphere is the reason for its success.
The Strengths of OpenClaw
- Open-source and free: The code is fully accessible on GitHub. No subscription, no opaque business model.
- Local-first: The agent runs on your machine. Your data does not pass through third-party servers (except for the LLM's webAPI).
- Multi-LLM flexibility: You choose your language model. If Claude is too expensive, you switch to DeepSeek. This freedom is invaluable.
- Messaging integration: It meets users where they are: Signal, Telegram, Discord, WhatsApp. No additional interface to learn.
Limitations and Risks to Be Aware Of
The enthusiasm surrounding OpenClaw must be tempered by a careful assessment of the real risks associated with the solution.
- 512 security vulnerabilities: A security audit conducted in January 2026 identified 512 vulnerabilities in the code, including 8 classified as critical. For an agent with access to your emails, calendars, and messaging, these flaws represent a real risk of exposure of sensitive data.
- No scientific specialization: OpenClaw is designed for everyday tasks, not for biomedical research. It has no connectors to PubMed, PMC, GEO, or Espacenet. It cannot produce verifiable citations with DOI or PMID.
- Exposed research data: By giving an agent access to your emails and calendars, you potentially expose confidential data—research correspondence, unpublished results—to risks related to identified vulnerabilities.
- No citations or scientific traceability: For research, the verifiability of sources is fundamental. OpenClaw is not designed to produce answers rooted in peer-reviewed literature with precise references.
General Agent vs. Scientific Agent: Two Philosophies
The difference between OpenClaw and Charlie is not just about features—it's philosophical. OpenClaw is a digital Swiss Army knife: it does many things well for anyone. Charlie is a precision instrument: it does one thing exceptionally well for researchers.
Specifically, if you ask OpenClaw, "What are the latest promising treatments for glioblastoma?", it will perform a general web search and summarize articles from blogs, the press, and forums. If you ask the same question to Charlie, it will directly query PubMed for the 50 most recent peer-reviewed publications, PMC for open access full texts, identify ongoing clinical trials, and provide you with a structured summary with 20 verifiable references (PMID/DOI). It's not the same product.
Comparison Table: OpenClaw vs.Charlie
| Criterion | OpenClaw | Charlie (Emerit Science) |
|---|---|---|
| Type | General AI agent | Specialized AI scientist |
| Source | Open source (GitHub) | Owner (SaaS) |
| Cost | Free (+ costAPILLM) | Free / Pro / Enterprise |
| Data & execution | Local (your machine) | Sovereign cloud (data centers in France) |
| AccessPubMed/PMC | No | Yes (native, real-time) |
| Scientific citations | No | Systematics (DOI/PMID) |
| Audited security | 512 vulnerabilities (8 critical, Jan. 2026) | Certified hosting, native GDPR compliance |
| Target audience | Consumer tech users | Researchers, laboratories, biotech companies |
"I tested OpenClaw out of curiosity—it's impressive for managing my emails and organizing my meetings. But for my research on biomarkers, it referred me to popular science articles. Charlie, on the other hand, provided me with 18 recent references PubMed with full texts. They really are two different tools." — Molecular biology researcher
Who Should Choose Each Solution?
OpenClaw is suitable if you:
- - Look for a free digital personal assistant
- - Manage emails, calendars, and messaging
- - Are comfortable with technical configuration
- - No need for verifiable scientific sources
Charlie is suitable if you:
- - Conduct biomedical research
- - Need verifiable citations (DOI/PMID)
- - Conduct systematic literature reviews
- - Work with sensitive data (GDPR compliance)
- - Look for a specialized agent, not a Swiss Army knife
Both tools can coexist in your daily life: OpenClaw for personal productivity tasks, Charlie for your scientific work. The important thing is to recognize that biomedical research requires rigor—source traceability, peer review, up-to-date data—that a general-purpose agent cannot guarantee.
Conclusion: Complementary, but not Interchangeable
OpenClaw is a remarkable achievement by the open-source community. Its popularity reflects a real need: accessible, flexible AI agents that can be integrated into everyday life without a subscription. For general personal and professional use, it is a tool that deserves attention.
But scientific research is not a general-purpose application. It requires verifiable sources, direct access to reference databases, traceability of citations, and security for sensitive research data. Charlie was designed specifically to meet these requirements. They are not competitors—they are two tools designed for fundamentally different contexts.
TryCharliefor Free
Discover for yourself the difference between a general agent and a specialized agent for biomedical research. Access to PubMed, PMC, GEO, and Espacenet included.
TryCharliefor FreeRelated articles
AI Agent vs. AI Assistant: What Are the Differences for Search?
Understanding the fundamental distinction between an AI agent and an AI assistant
What is a Scientific AI Agent?
Understanding the fundamentals of AI agents applied to biomedical research
Emmy (CNRS) vsCharlie: Which AI Tool for Research?
Comparative analysis of two French approaches to scientific AI