Skip to main content

Coder mode

Coder agents are specialised AI assistants designed to aid developers and data scientists in their programming tasks. These agents are enhanced with programming documentation and code execution capabilities, making them powerful tools for coding assistance and productivity enhancement.

Key Features​

  1. Code Assistance: Coder agents can help with writing, debugging, and optimising code across various programming languages.

  2. Knowledge Hub Access: Similar to the Retriever agents, they have access to the Knowledge Hub, allowing them to provide accurate and up-to-date information on language syntax, libraries, and best practices using the uploaded information as source of truth.

  3. Code Execution: When enabled, Coder agents can execute code, allowing for real-time testing and demonstration of code snippets.

  4. Code Upload and Analysis: Users can upload source code files for the agent to analyse, providing insights, suggestions for improvements, or help with debugging.

  5. Language Versatility: Coder agents are typically proficient in multiple programming languages, making them versatile assistants for various development tasks.

Execution Environments and Code Upload Settings​

  • Code environments: Coder agents can leverage predefined docker images and code snippets to execute code. ToothFairyAI automatically detects and uses the most suitable environment for the task at hand leveraging the description of the environment. You can find more details here.

  • Code upload: Allows users to upload up to 10 source code files at a time for analysis or reference. To review which file types are supported see here.

Use Cases​

  1. Code Writing Assistance: Helping developers write efficient and clean code by providing suggestions and examples.

  2. Debugging Support: Analysing code to identify and help fix bugs or errors.

  3. Process automation: Leveraging Code execution, Coder agents are extremely powerful at automating tasks by running executable python code. See more details here.

  4. Language and Library Queries: Answering questions about programming language syntax, library usage, and best practices.

  5. Code Explanation: Breaking down complex code snippets and explaining their functionality.

  6. Algorithm Implementation: Assisting with the implementation of various algorithms and data structures.

  7. Code Review: Providing insights and suggestions for improving uploaded code.

  8. Development Environment Setup: Offering guidance on setting up development environments and tools.

Process Automation with Coder Agents​

Coder agents can automate tasks by running executable Python code in three distinct modes.

Fully Dynamic Mode​

In this case, the agent has access to no execution environments. The agent will attempt to accomplish the task solely with the available libraries and the inner-knowledge of the LLM used.

  • Code Generation: Code is generated on-the-fly by the agent.
  • Limited Access: No access to external resources requiring credentials.
  • Best for: One-off operations with limited production resource access.
  • Use Case: Ideal for quick, ad-hoc tasks that don't require extensive resource access.

Guided Mode​

In this case, the agent would have access to one or more execution environments with code snippets and execution guidelines to steer the agent towards a more predictable behaviour.

  • Code Execution Environment: Leverages a code execution environment in the workspace.
  • Access to Credentials: Access to credentials (e.g., API keys) and guidance on code structure, output style, and guidelines.
  • Versatility: Most versatile mode, but thorough testing is recommended before deploying to production.
  • Use Case: Suitable for tasks that require a balance between flexibility and reliability.

Scripted Mode​

  • Pre-Defined Scripts: Executes pre-defined scripts without inferring any parameters or dependencies.
  • Reliable Execution: Ideal for repeated, predictable tasks that require extremely reliable execution every time.
  • Use Case: Perfect for automated workflows with well-defined inputs and outputs.

To review which libraries are available and how execution environments can be customised see here

Advanced Features​

  • Show code blocks: When set to true (default), the agent displays code snippets in its responses. This can be toggled off if code display is not needed.

  • Knowledge Settings: Coder agents can be configured to use specific documentation sources, enhancing their knowledge in particular areas or technologies.

Limitations​

  • While Coder agents can execute code, they operate in a controlled environment and may have limitations on the complexity or resource requirements of the code they can run as the only supported language at this time is python.
  • When Code execution is invoked, the agent will behind the scenes operate with our internal instructions and guidelines to ensure no misuse or abuse of the agent's capabilities. Therefore, if you need a specific behaviour (e.g. what the agent has already in the instructions settings) you have to re-enter the same in the chat itself prior to the execution running.

Best Practices​

  1. Be Specific: When asking for coding help, provide as much context as possible, including the programming language, relevant libraries, and specific requirements.

  2. Use Code Blocks: When sharing code with the agent, use proper formatting (such as markdown code blocks) to ensure the agent can accurately read and analyse the code.

  3. Iterative Approach: For complex problems, break them down into smaller parts and work with the agent iteratively to build a complete solution.

  4. Verify Outputs: While Coder agents are highly capable, always review and test the code they provide to ensure it meets your specific needs and works in your environment.

  5. Leverage Documentation: Use the agent's access to documentation by asking for explanations or clarifications on specific programming concepts or library usage.

  6. Utilise Code Execution: When available, use the code execution feature to test and refine code snippets in real-time.

  7. Explore Different Modes: Familiarise yourself with the different automation modes (Fully Dynamic, Guided, and Scripted) to choose the most appropriate one for your task.

  8. Keep Security in Mind: When using Guided or Scripted modes with access to credentials, ensure proper security measures are in place to protect sensitive information.

  9. Regularly Update Knowledge: If using custom documentation sources, keep them updated to ensure the agent provides the most current information.

  10. Combine with Other Agent Types: Consider using Coder agents in conjunction with other agent types (e.g., Retriever agents) for more comprehensive problem-solving capabilities.