A New Way to Think About AI
Artificial intelligence moves fast. New tools appear often that can write, research, and plan. Many of these tools are closed and owned by a few large companies. They work well but limit control for developers, researchers, and the open-source community.
Sentient open AGI Foundation aims to change that. It is an open-source ecosystem that helps developers build intelligent agents that can think, search, and reason with transparency and safety. The main parts of Sentient include ROMA, Open Deep Search, the OML-1.0 fingerprinting approach, the Enclaves framework, and the Dobby family of models. Together, these pieces form a full, open stack for building powerful AI services.
ROMA The Brain of the System
ROMA stands for Recursive Open Meta-Agent. Think of ROMA as the coordinator that breaks big tasks into smaller steps and assigns each step to a suitable tool or sub-agent. ROMA does this in a recursive way: small tasks may be split further until they reach a step that a single model or tool can handle.
How ROMA works: it first checks whether a task is simple or complex, then plans sub-tasks, executes them using chosen tools or models, and finally combines results into a clear final answer. This method helps with long, multi-step problems like research summaries, data analysis, and complex decision support.
ROMA matters because it is transparent, flexible, and scalable. Developers can plug in different models and tools, inspect each step of reasoning, and improve individual parts. This offers more control than relying on a single, closed model.
Open Deep Search Smarter, Open Search for Everyone
Open Deep Search (ODS) is Sentient’s open search and reasoning agent. It finds information across sources, scores and ranks results, and builds reasoned answers rather than just returning links.
Key features of Open Deep Search include semantic search (understanding meaning, not just keywords), query refinement, and multi-hop reasoning that ties together evidence from multiple documents. ODS is designed to work well with ROMA and other agent layers, making it useful as a knowledge engine inside larger systems.
OML-1.0 Fingerprinting Open Access with Fair Rewards
A long-standing question for open-source AI is how to reward creators while keeping models open. The OML-1.0 fingerprinting idea helps address this need. It lets models be public while tracking their origin and use. Fingerprinting supports attribution, helps detect misuse, and opens paths for fair monetization without closing the model.
Sentient Enclaves Framework Secure AI Execution
When an agent runs code or accesses sensitive data, it must operate safely. The Enclaves framework isolates tasks in secure sandboxes, verifies tools before use, and records actions for audit. This isolation protects data and keeps system integrity when agents perform complex actions.
How the parts work together
ROMA orchestrates planning and composition. Open Deep Search gathers and reasons over data. Enclaves keep execution safe. Fingerprinting preserves trust and attribution. When combined, these elements let users ask multi-step questions and receive transparent, traceable answers.
Dobby The Loyal Model Family
A key part of the Sentient ecosystem is Dobby, a family of community-driven models that act as trusted base models for many agents. Dobby models are designed to be open and ‘loyal’ to community values such as individual freedom and decentralization. They are fingerprinted and community-governed so users can verify model origin and track changes.
Dobby appears in multiple sizes and flavors, from smaller local-friendly versions to large-capacity models. Variants include Mini and larger 70B models. The Dobby models are used inside Sentient’s agents and demos, and they can be run locally or on cloud services via standard model hosting tools.
Why Dobby matters: by offering a community-owned model that integrates closely with ROMA and ODS, Sentient gives teams a reliable default model that aligns with the broader goals of openness and traceability. Dobby also forms a natural trust layer for agents that need a consistent reasoning base.
Practical example: A full request flow
Imagine asking a Sentient-powered agent: “Create a short market report on renewable energy in Asia.” ROMA splits the task into research, data analysis, writing, and summary. Open Deep Search finds up-to-date sources. Enclaves run safe data cleaning and analysis routines. Dobby furnishes stable reasoning and natural-language synthesis. Finally, the system checks attribution via fingerprinting and returns a clear report.
Real-world applications
Sentient’s stack fits many domains. It can power research assistants that summarize academic papers, business tools that collect and analyze market data, educational tutors that plan long learning journeys, and secure systems that run code safely for finance or health tasks.
Why users and developers should care
For users: Sentient aims to deliver clear, transparent results with secure execution and verified origins. For developers: the stack offers modular components to build, test, and customize agents without losing control to a proprietary vendor.
The vision behind Sentient
Sentient’s core idea is that AI should be open, accountable, and useful for everyone. The project shows how open models, verified fingerprinting, secure enclaves, and modular orchestration can work together to give people real control over AI services.
Next steps and how to get started
If you want to try Sentient’s tools, start by exploring the ROMA and Open Deep Search repositories, then experiment with a Dobby model from Hugging Face. Try running a simple ROMA workflow with Dobby as the model and ODS as the search tool, using a safe enclave for any code execution steps.
Conclusion
Sentient proves that advanced AI does not have to live behind closed doors. By combining orchestration, search, security, attribution, and community models like Dobby, Sentient builds a complete open stack for intelligent agents. This approach helps keep innovation open, accountable, and accessible.