Prompt to Production: Inside Orquesta's Seamless Workflow

Originally published at orquesta.live/blog/prompt-to-production-inside-orquestas-seamless-workflow

Introduction to Orquesta's Workflow

When it comes to automating coding tasks, Orquesta takes an approach rooted in transparency and efficiency. From the moment a team member submits a prompt, to the point where code is committed and deployed, Orquesta orchestrates the entire process locally on your infrastructure. Let’s explore how this works in practice.

Writing the Prompt

The journey begins with a prompt, which can be something as straightforward as "Add a new feature to the user login system" or "Optimize the database query performance." Team members can submit these prompts directly through the Orquesta dashboard or even via a Telegram bot for on-the-go interactions. Prompts provide the AI agents with the initial context needed to start working.

Picking Up the Prompt

Once a prompt is submitted, the AI agent, running locally on your machine, picks it up. This local execution is crucial for maintaining security and control over your codebase. The agent employs Claude CLI, part of our Orquesta CLI toolset, to interpret the prompt and begin the process. Claude CLI facilitates integration with various local LLMs like OpenAI, Ollama, and vLLM, ensuring the processing stays within your infrastructure.

Writing the Code

Upon understanding the task, the agent autonomously writes the required code. This is where our execution modes come into play. The agent may choose from four modes:

• Auto: The AI decides the best path between SSH, Agent, or Batuta execution.
• SSH: Direct execution of commands on the target machine.
• Agent: Using Claude CLI for nuanced code tasks.
• Batuta: A ReAct loop that thinks, acts, observes, and repeats, perfect for complex tasks.

The agent writes code with adherence to your team’s coding standards, automatically syncing with CLAUDE.md to ensure consistency.

Running Tests

Before any code is committed, Orquesta’s agents simulate changes and run tests. This step acts as a quality gate, ensuring that only validated changes make it through. The AI performs these simulations autonomously, streaming the output of each test back to the dashboard in real-time.

Committing the Code

Once tests pass successfully, the agent commits the changes to your Git repository. Each action is a real git commit, providing a full audit trail. This transparency allows team leads to review and sign off on changes before they go live, maintaining a high level of control and oversight.

Streaming Output Back to Dashboard

Throughout the entire process, Orquesta streams every line of output back to the Agent Grid. This feature allows you to monitor dozens of agents from a single screen with live terminal views. You can see how each prompt translates into executed actions, providing insights into the AI’s decision-making process.

Ensuring Security and Collaboration

Security is at the core of Orquesta’s design. With AES-256 encryption and a full audit trail of prompts, logs, diffs, and costs, your data remains protected. Additionally, team collaboration is seamless with role-based permissions that allow you to invite anyone to submit prompts while maintaining control over who can execute or approve changes.

Conclusion: No Magic, Just Orchestration

Orquesta doesn’t rely on magic to bridge the gap between prompts and production code. Instead, it provides a robust orchestration framework that leverages local AI execution, real-time monitoring, and collaborative tools to streamline your development process. By keeping everything on your infrastructure, it offers a unique blend of security, transparency, and efficiency, allowing teams to focus on innovation rather than the mechanics of execution.