Parallel Works
View as Markdown
Share:

Best Practices

This page covers recommended practices for getting the most out of AI Chat on the ACTIVATE platform.

Writing Effective Prompts

The quality of AI responses depends heavily on how you frame your requests.

Be Specific

Vague prompts produce vague answers. Include relevant details, constraints, and the desired format in your prompt.

Instead ofTry
"Explain Kubernetes""Explain how Kubernetes pod scheduling works, including the role of the scheduler and node affinity rules, in about 200 words."
"Write a script""Write a Bash script that finds all CSV files in a directory and combines them into a single file with one header row."

Provide Context

Give the model the background it needs to produce a useful response:

  • Describe the environment or system you are working with.
  • Mention relevant constraints (language, framework, version).
  • Reference earlier parts of the conversation if building on previous responses.

Use System Messages

If your provider supports system messages, use them to set the model's behavior for the entire conversation. For example, you can instruct the model to respond as a domain expert, use a particular coding style, or avoid certain topics.

Iterate and Refine

If the first response is not quite right, refine your prompt rather than starting from scratch. Add clarifications, ask for a different format, or request that the model focus on a specific aspect.

Choosing the Right Model

Different models have different strengths. Selecting the right model for your task improves both response quality and efficiency.

Standard Models

Standard models (such as GPT-4o, GPT-4o-mini) are best for:

  • General-purpose questions and conversation
  • Text summarization and generation
  • Code generation for straightforward tasks
  • Tasks where speed matters more than deep analysis

Reasoning Models

Reasoning models (such as o1, o3) are best for:

  • Complex math, logic, and multi-step problem-solving
  • Detailed code review and debugging
  • Architectural decisions that require evaluating tradeoffs
  • Tasks where accuracy is more important than speed

See Using Reasoning Models for details on reasoning effort configuration.

Cost Considerations

Larger and more capable models consume more tokens and may have higher usage costs. When working on routine tasks, consider using a smaller or faster model to conserve resources. Reserve more powerful models for tasks that genuinely benefit from their capabilities.

Working with File Attachments

File attachments allow you to provide the model with additional context beyond what fits in a text prompt.

Keep Documents Focused

Rather than uploading a large document and asking a broad question, extract the relevant section or upload a smaller, targeted file. This helps the model focus on the content that matters and reduces token usage.

Use the Right File Type

  • Code files — Attach source code files directly for code review, debugging, or analysis tasks.
  • Text and Markdown — Use plain text or Markdown for structured information, notes, or specifications.
  • PDFs and documents — Upload reports, papers, or documentation that the model can reference when answering your questions.

Be Aware of Size Limits

Regular files have a maximum size of 25 MB, and documents have a maximum size of 100 MB. If a file exceeds these limits, split it into smaller parts or extract the relevant sections.

See Attaching Files for supported formats and detailed usage instructions.

Collaboration Tips

AI Chat's sharing and branching features support team workflows.

Share with Appropriate Permissions

  • Use View permission for stakeholders who need to review the conversation without modifying it.
  • Use Collaborate permission for team members who need to actively contribute messages.
  • Share at the group level rather than managing individual access, so permissions stay current as team membership changes.

See Sharing Conversations for detailed instructions.

Use Branching for Exploration

When evaluating different approaches, create branches from the same message rather than cluttering a single thread. Each branch maintains its own context, allowing you to compare model responses to different prompts or explore alternative solutions side by side.

See Branching Conversations for details.

Document Decisions

Use the conversation itself as a record of your analysis. When you reach a conclusion, summarize the decision in a follow-up message so that anyone reviewing the shared conversation can quickly understand the outcome.

Security Considerations

Sensitive Data

Be mindful of the data you include in prompts and file attachments. Messages are sent to the configured AI provider endpoint, which may be hosted externally.

  • For sensitive or proprietary data, use a session tunnel provider that routes requests to a model running on your own compute infrastructure. See Session Tunnels.
  • Avoid including credentials, API keys, or personally identifiable information in prompts.

API Key Management

If you manage AI Chat providers, follow these practices for API key security:

  • Rotate keys regularly — Update API keys in the provider configuration on a regular schedule.
  • Use the minimum required permissions — Configure provider API keys with only the permissions needed for chat completions.
  • Monitor usage — Review token usage and access patterns through your provider's dashboard to detect unexpected activity.

Provider Permissions

Limit provider access to the groups and users who need it. Review and audit provider permissions periodically to ensure that only authorized teams have access.

See Managing Provider Permissions for details.

Related Documentation