Divar is a practical example of a multi-agent system applied to the critical problem of technical documentation. By deploying specialized agents for analysis and synthesis, the framework avoids the context-window limitations that often hinder simpler AI tools. This allows for a more granular understanding of codebases that are too large for a single prompt to handle.
In the AI agent stack, Divar occupies the developer productivity layer. They are championing a shift from conversational AI to background agentic processes that integrate directly with CI/CD pipelines. This matters to the ecosystem because it demonstrates how multi-agent architectures can be used to solve specific enterprise-scale problems, such as maintaining technical clarity across dozens of microservices without manual intervention.
Divar is widely recognized as the primary classifieds platform in Iran, offering a service that mirrors the core functionality of Craigslist or OLX for the Persian-speaking market. However, the organization's technical arm, represented on GitHub under the organization handle divar-ir, has expanded into the AI agent space. Their primary contribution to the ecosystem is an AI-powered multi-agent system designed to address the persistent problem of stale or missing technical documentation in large software projects.
Maintaining documentation is a historically difficult task because codebases often change faster than the prose describing them. Divar addresses this by using a multi-agent architecture that treats documentation as a living byproduct of development. Instead of relying on a single large language model to hold an entire repository in its context—an approach that often leads to errors or missed nuances in complex systems—the multi-agent model distributes tasks among specialized roles. One agent might be responsible for mapping internal dependencies, while another focuses on functional summaries, and a third ensures the final output follows specific style requirements.
This concurrency is a requirement for scale. For a platform like Divar, which manages a massive digital marketplace with millions of users and a sprawling microservices architecture, documentation must be both deep and broad. The multi-agent system allows for the parallel processing of these services, making it possible to document a complex codebase in a fraction of the time required by a human team. This project sits at the intersection of agentic AI and DevOps, a category often called DevAI, where the goal is to remove the cognitive load of routine maintenance from engineers.
Integration is the core of the Divar value proposition. The system is designed to exist within the existing developer workflow, specifically through GitLab integration. By hooking into version control, the agents can trigger documentation updates automatically as code changes are committed. This effectively turns technical documentation into a continuous process rather than a periodic chore.
In the broader market, Divar competes with emerging AI documentation platforms and built-in features from major cloud providers. Their differentiator is the multi-agent approach, which suggests a more modular and thorough analysis than single-agent or simple RAG-based solutions. By open-sourcing these tools, the Divar engineering team contributes to a growing library of agentic frameworks that are moving beyond basic chat interfaces and into autonomous, specialized work. The result is a tool that understands the code not just as a set of files, but as a functioning system that requires constant, automated translation for the humans who maintain it.
An AI framework that analyzes codebases and generates technical documentation.
Divar is hiring.