LLM Options

What Is the Language Model Setting?

The “Language Model” section allows you to fine-tune how your Agent thinks and communicates. It’s the core AI engine that powers your Agent’s responses, decision-making and reasoning.

Selecting the Model

Model Tier Options

  • Default model: Balanced performance and cost — ideal for most use-cases.

  • Advanced models: Improved reasoning and domain expertise — higher cost, may require more resources.

  • Specialized models: Optimised for tasks like coding, creative writing, domain-specific knowledge.

Recommendation

If you’re just getting started, stick to the default model. You can switch later as your workflow grows or as you identify the need for more capability.

Key Settings

Maximum Output Tokens

  • Definition: How much output your Agent can generate in one run — including reasoning, step-by-step thinking, and response.

  • Trade-offs:

    • Lower limit → faster results, lower cost, more concise responses.

    • Higher limit → more depth, more detail, but slower and more expensive — risk of exceeding model limits.

  • Use-case: Set higher if your Agent works on large inputs, multi-step tasks, large document reasoning.

Temperature

  • Definition: A control that influences randomness vs consistency of responses.

  • Guidance:

    • Low (0.0-0.3): Very focused, repeatable, predictable responses — good for factual, reliable workflows.

    • Medium (0.4-0.7): Balanced creative/precise mix — good for general tasks with some flexibility.

    • High (0.8-1.0): Highly creative, varied responses — good for brainstorming, ideation, narrative generation.

Reasoning / “Thinking” Effort

  • Some models (e.g., from OpenAI, Google, Anthropic) support explicit reasoning effort.

  • Definition: Enables the Agent to perform deeper chained reasoning — more deliberate, slower, more cost.

  • Only applicable if your chosen model supports it — otherwise ignored.

  • Use-cases: Complex workflows, multi-step logic, critical decision-making tasks.

Best Practices

  • Start conservative: Use default model + moderate tokens + low to medium temperature.

  • Tune over time: Increase tokens or temperature only when needed based on use-case performance.

  • Monitor cost: Higher tokens and advanced models => more compute → watch usage.

  • Align model to task:

    • Factual/data-driven tasks → lower temperature, concise, reliable model.

    • Creative/ideation tasks → higher temperature, maybe advanced/specialised model.

  • Test edge cases: Especially for tasks with high stakes or multi-step reasoning. Ensure stability before scaling.

Example Configuration

  • Model: “Default-LLM-v2”

  • Max output tokens: 3,000

  • Temperature: 0.2

  • Reasoning: Off (for factual task)

  • Use-case: Invoice data extraction and validation — needs reliable, consistent output rather than creativity.
    Later, if you adopt a “Draft newsletter generator” Agent:

    • Model: “Creative-LLM-v1”

    • Max tokens: 5,000

    • Temperature: 0.85

    • Reasoning: On (for multi-step ideation & narrative)

Summary

The Language Model settings give you control over which model your Agent uses and how it behaves in terms of output length, creativity, and reasoning depth. By tailoring these settings to your workflow you ensure your Agent is both cost-efficient and fit-for-purpose.

Would you like this saved as a .md file ready for download, or an HTML version styled for your docs-site?

Updated on
App Triggers
Run your Agent