AI coding tools have advanced rapidly in recent years. Code generation, bug fixing, and document summarization have become increasingly accessible, even to users without extensive technical expertise. But a more significant shift is taking place in enterprise development environments: AI is becoming part of actual code repositories and collaborative development workflows.
Developers are moving beyond simply asking AI questions. They increasingly need AI to work with source code, design documents, operational configurations, and test results. As AI agents gain the ability to explore repositories, reference relevant files, and execute multi-step workflows, AI is beginning to function as an integrated component of enterprise development systems.
Why Local LLMs Matter
Early enterprise adoption of generative AI largely centered on external SaaS-based services. These solutions offered immediate access to powerful models without requiring organizations to build and maintain their own infrastructure.
However, integrating AI into enterprise development environments introduces a different set of challenges.
Development systems contain sensitive information, including proprietary source code, credentials, internal documentation, operational configurations, and incident logs. As AI agents interact with repositories, execution results, and connected systems, the scope of data they access can expand significantly.
Without appropriate access controls and execution boundaries, sensitive information may be transmitted through external networks. For organizations operating in restricted or air-gapped environments, external connectivity may not be permitted at all.
The key consideration is therefore no longer just which AI model to use, but where AI runs and what data and systems it is authorized to access.
Local LLMs offer a practical approach to addressing these requirements. By running models within enterprise infrastructure, organizations can maintain greater control over data flows and prevent source code and internal documents from being sent to external AI services for inference. External connectivity and system access can also be restricted according to organizational security policies.
Addressing Enterprise Local LLM Operations with Putron
Deploying a Local LLM is one challenge. Operating it reliably across an enterprise is another.
Running a model for individual use differs significantly from providing an AI environment that supports multiple development teams. Enterprise deployments must account for inference performance, GPU resource utilization, concurrent users, security policies, and operational stability.
In the webinar, Putron was introduced as an infrastructure solution for operating Local LLMs in enterprise environments.
Putron AI provides an integrated environment for managing key components of enterprise AI operations, including fine-tuning, Retrieval-Augmented Generation (RAG), LLM serving, and Model Context Protocol (MCP) connectivity.
Its architecture also incorporates KV cache management during inference to reduce GPU memory pressure, supporting more efficient and stable AI operations within available hardware resources.
Connecting Development and Collaboration: CodeCenter IDE and GitOn
A central focus of the webinar was connecting development workflows through MCP.
Software development extends beyond writing code. It involves issue tracking, branch management, commits, Pull Requests (PRs), code reviews, and approvals. Yet many AI coding tools primarily operate within the IDE, leaving the broader development and collaboration process disconnected.
The webinar demonstrated how CodeCenter IDE and GitOn can be connected through MCP to extend AI-assisted development beyond the IDE.
Within this workflow, developers can work on code associated with existing issues, create branches, analyze and modify code, and connect their changes to commits and Pull Requests without constantly switching between tools.
The demonstration also introduced GitOn AI's Lite Model and AI User capabilities, illustrating how role-based AI collaboration can help reduce repetitive development tasks.
Each component serves a distinct purpose:
- CodeCenter IDE: The development environment where developers analyze, modify, and work with source code.
- GitOn: The Git-based platform for managing code changes, collaboration, reviews, and development workflows.
- MCP: The integration protocol connecting these environments to support AI-driven workflows.
From IDE-Based Development to Connected Git Workflows
The live demonstration began with a sample repository containing a malfunctioning feature.
The developer described the issue in natural language within CodeCenter IDE. The AI then retrieved project information through MCP and created a corresponding Issue in GitOn.
Using that Issue as the starting point, the workflow proceeded with branch creation, code structure analysis, and the development of a modification plan within CodeCenter IDE.
The demonstrated workflow included:
- Branch creation
- Code structure analysis and modification
- Commit creation
- Pull Request creation
- Review verification
Traditionally, coding, issue tracking, code review, and team collaboration often take place across separate tools and interfaces.
In this demonstration, work initiated within the IDE extended directly into the Git-based collaboration workflow, connecting individual development activities with the broader software delivery process.
This approach illustrates how MCP can help bridge the gap between AI-assisted coding and enterprise development operations.
Reducing Repetitive Tasks with GitOn AI Lite Model
GitOn AI's Lite Model focuses on fast, lightweight assistance rather than computationally intensive reasoning tasks.
In everyday development, repetitive activities such as locating functions, identifying relevant symbols, searching for changes, and preparing Pull Request descriptions can consume considerable time.
The Lite Model is designed to simplify these tasks.
For example, its natural language search capability converts user requests into GitOn's structured search syntax. Developers can locate relevant code and change history without manually constructing complex search queries.
The Lite Model also supports context-aware symbol search.
Traditional keyword searches often return multiple functions or variables with matching names, leaving developers to determine which result is relevant.
By analyzing the current code context, the Lite Model can prioritize symbols most relevant to the developer's current task, helping narrow down the search results.
A similar approach applies to Pull Request title and description generation. By using information from code changes, GitOn AI can help prepare the documentation needed for collaborative reviews.
Rather than attempting to replace complex development reasoning, the Lite Model focuses on reducing the time developers spend on frequent, repetitive tasks.
AI Users: Role-Based AI Collaboration
Another capability introduced in the webinar was GitOn AI User, which allows AI to operate as a role-specific participant within development workflows.
Unlike a single general-purpose AI assistant, AI Users can be configured for different responsibilities, with separate system prompts and AI models assigned to each role.
For example:
- Security Reviewer AI: Reviews code for potential security vulnerabilities.
- Performance Reviewer AI: Evaluates code from a performance optimization perspective.
- QA Reviewer AI: Identifies potential gaps in testing and examines edge cases.
Because each AI User can have its own model and system prompt, the same code changes can be evaluated against different review criteria.
Organizations can also control which users are authorized to access specific AI Users and which projects those AI Users can operate within.
This creates a more structured approach to AI collaboration, where AI responsibilities and access permissions can be aligned with organizational roles and project boundaries.
Rather than relying on one AI assistant for every task, development teams can establish specialized AI participants within their existing collaboration processes.
Extending GitOn AI into CI/CD Workflows
GitOn AI's capabilities also extend beyond code understanding and review into CI/CD pipeline configuration and build failure analysis.
During the webinar, the demonstration showed how developers could request CI job creation using natural language.
It also illustrated how AI could analyze failed build logs, identify potential causes, and suggest corrective actions.
These capabilities highlight another direction for AI-assisted development: extending AI support from code creation into the operational processes required to build, test, and deliver software.
Conclusion: AI as Part of the Enterprise Development Workflow
The central theme of the webinar was how to integrate AI into real-world development workflows while operating within enterprise infrastructure.
The demonstration brought together three complementary components:
Puteron AI provides the infrastructure for running and managing Local LLMs within enterprise environments.
CodeCenter IDE serves as the development environment where AI-assisted code analysis and modification take place.
GitOn AI extends these capabilities into Git-based collaboration, code review, and CI/CD workflows.
Together, these technologies illustrate an approach in which AI is not limited to answering questions or generating code. Instead, it becomes an execution layer connected to the systems and processes that development teams already use.
For enterprises, this shift introduces considerations beyond model performance. Infrastructure control, system integration, access permissions, and workflow governance are becoming increasingly important to successful AI adoption.
The webinar demonstrated how Local LLM infrastructure, MCP connectivity, and role-based AI collaboration can work together to support this transition toward more integrated, AI-assisted development.
SLEXN will continue sharing practical insights into Local LLMs, AI-powered development, agentic workflows, and enterprise AI infrastructure.
[WEBINAR] Git-based Development Collaboration Structure Implemented with Multi-LLMs and AI Agents
This webinar explores how to build a secure, Git-based AI development collaboration structure using local infrastructure.
AI coding tools have advanced rapidly in recent years. Code generation, bug fixing, and document summarization have become increasingly accessible, even to users without extensive technical expertise. But a more significant shift is taking place in enterprise development environments: AI is becoming part of actual code repositories and collaborative development workflows.
Developers are moving beyond simply asking AI questions. They increasingly need AI to work with source code, design documents, operational configurations, and test results. As AI agents gain the ability to explore repositories, reference relevant files, and execute multi-step workflows, AI is beginning to function as an integrated component of enterprise development systems.
Why Local LLMs Matter
Early enterprise adoption of generative AI largely centered on external SaaS-based services. These solutions offered immediate access to powerful models without requiring organizations to build and maintain their own infrastructure.
However, integrating AI into enterprise development environments introduces a different set of challenges.
Development systems contain sensitive information, including proprietary source code, credentials, internal documentation, operational configurations, and incident logs. As AI agents interact with repositories, execution results, and connected systems, the scope of data they access can expand significantly.
Without appropriate access controls and execution boundaries, sensitive information may be transmitted through external networks. For organizations operating in restricted or air-gapped environments, external connectivity may not be permitted at all.
The key consideration is therefore no longer just which AI model to use, but where AI runs and what data and systems it is authorized to access.
Local LLMs offer a practical approach to addressing these requirements. By running models within enterprise infrastructure, organizations can maintain greater control over data flows and prevent source code and internal documents from being sent to external AI services for inference. External connectivity and system access can also be restricted according to organizational security policies.
Addressing Enterprise Local LLM Operations with Putron
Deploying a Local LLM is one challenge. Operating it reliably across an enterprise is another.
Running a model for individual use differs significantly from providing an AI environment that supports multiple development teams. Enterprise deployments must account for inference performance, GPU resource utilization, concurrent users, security policies, and operational stability.
In the webinar, Putron was introduced as an infrastructure solution for operating Local LLMs in enterprise environments.
Putron AI provides an integrated environment for managing key components of enterprise AI operations, including fine-tuning, Retrieval-Augmented Generation (RAG), LLM serving, and Model Context Protocol (MCP) connectivity.
Its architecture also incorporates KV cache management during inference to reduce GPU memory pressure, supporting more efficient and stable AI operations within available hardware resources.
Connecting Development and Collaboration: CodeCenter IDE and GitOn
A central focus of the webinar was connecting development workflows through MCP.
Software development extends beyond writing code. It involves issue tracking, branch management, commits, Pull Requests (PRs), code reviews, and approvals. Yet many AI coding tools primarily operate within the IDE, leaving the broader development and collaboration process disconnected.
The webinar demonstrated how CodeCenter IDE and GitOn can be connected through MCP to extend AI-assisted development beyond the IDE.
Within this workflow, developers can work on code associated with existing issues, create branches, analyze and modify code, and connect their changes to commits and Pull Requests without constantly switching between tools.
The demonstration also introduced GitOn AI's Lite Model and AI User capabilities, illustrating how role-based AI collaboration can help reduce repetitive development tasks.
Each component serves a distinct purpose:
- CodeCenter IDE: The development environment where developers analyze, modify, and work with source code.
- GitOn: The Git-based platform for managing code changes, collaboration, reviews, and development workflows.
- MCP: The integration protocol connecting these environments to support AI-driven workflows.
From IDE-Based Development to Connected Git Workflows
The live demonstration began with a sample repository containing a malfunctioning feature.
The developer described the issue in natural language within CodeCenter IDE. The AI then retrieved project information through MCP and created a corresponding Issue in GitOn.
Using that Issue as the starting point, the workflow proceeded with branch creation, code structure analysis, and the development of a modification plan within CodeCenter IDE.
The demonstrated workflow included:
- Branch creation
- Code structure analysis and modification
- Commit creation
- Pull Request creation
- Review verification
Traditionally, coding, issue tracking, code review, and team collaboration often take place across separate tools and interfaces.
In this demonstration, work initiated within the IDE extended directly into the Git-based collaboration workflow, connecting individual development activities with the broader software delivery process.
This approach illustrates how MCP can help bridge the gap between AI-assisted coding and enterprise development operations.
Reducing Repetitive Tasks with GitOn AI Lite Model
GitOn AI's Lite Model focuses on fast, lightweight assistance rather than computationally intensive reasoning tasks.
In everyday development, repetitive activities such as locating functions, identifying relevant symbols, searching for changes, and preparing Pull Request descriptions can consume considerable time.
The Lite Model is designed to simplify these tasks.
For example, its natural language search capability converts user requests into GitOn's structured search syntax. Developers can locate relevant code and change history without manually constructing complex search queries.
The Lite Model also supports context-aware symbol search.
Traditional keyword searches often return multiple functions or variables with matching names, leaving developers to determine which result is relevant.
By analyzing the current code context, the Lite Model can prioritize symbols most relevant to the developer's current task, helping narrow down the search results.
A similar approach applies to Pull Request title and description generation. By using information from code changes, GitOn AI can help prepare the documentation needed for collaborative reviews.
Rather than attempting to replace complex development reasoning, the Lite Model focuses on reducing the time developers spend on frequent, repetitive tasks.
AI Users: Role-Based AI Collaboration
Another capability introduced in the webinar was GitOn AI User, which allows AI to operate as a role-specific participant within development workflows.
Unlike a single general-purpose AI assistant, AI Users can be configured for different responsibilities, with separate system prompts and AI models assigned to each role.
For example:
- Security Reviewer AI: Reviews code for potential security vulnerabilities.
- Performance Reviewer AI: Evaluates code from a performance optimization perspective.
- QA Reviewer AI: Identifies potential gaps in testing and examines edge cases.
Because each AI User can have its own model and system prompt, the same code changes can be evaluated against different review criteria.
Organizations can also control which users are authorized to access specific AI Users and which projects those AI Users can operate within.
This creates a more structured approach to AI collaboration, where AI responsibilities and access permissions can be aligned with organizational roles and project boundaries.
Rather than relying on one AI assistant for every task, development teams can establish specialized AI participants within their existing collaboration processes.
Extending GitOn AI into CI/CD Workflows
GitOn AI's capabilities also extend beyond code understanding and review into CI/CD pipeline configuration and build failure analysis.
During the webinar, the demonstration showed how developers could request CI job creation using natural language.
It also illustrated how AI could analyze failed build logs, identify potential causes, and suggest corrective actions.
These capabilities highlight another direction for AI-assisted development: extending AI support from code creation into the operational processes required to build, test, and deliver software.
Conclusion: AI as Part of the Enterprise Development Workflow
The central theme of the webinar was how to integrate AI into real-world development workflows while operating within enterprise infrastructure.
The demonstration brought together three complementary components:
Puteron AI provides the infrastructure for running and managing Local LLMs within enterprise environments.
CodeCenter IDE serves as the development environment where AI-assisted code analysis and modification take place.
GitOn AI extends these capabilities into Git-based collaboration, code review, and CI/CD workflows.
Together, these technologies illustrate an approach in which AI is not limited to answering questions or generating code. Instead, it becomes an execution layer connected to the systems and processes that development teams already use.
For enterprises, this shift introduces considerations beyond model performance. Infrastructure control, system integration, access permissions, and workflow governance are becoming increasingly important to successful AI adoption.
The webinar demonstrated how Local LLM infrastructure, MCP connectivity, and role-based AI collaboration can work together to support this transition toward more integrated, AI-assisted development.
SLEXN will continue sharing practical insights into Local LLMs, AI-powered development, agentic workflows, and enterprise AI infrastructure.
Written by
