Agents

Agent Models

Pick the agent model that fits your use case, from fast high-volume chat to HIPAA-compliant healthcare.

Link AI uses agent models: combinations of LLMs from several providers, tuned for specific use cases. Each agent model routes every request to the best underlying LLM for the task, balancing quality, speed, and cost.

Agent Models

Available agent models

ModelOptimized forQualitySpeed
Link Core SmartComplex reasoning and accuracy. Handles nuanced questions.HighestModerate
Link Core FastSpeed and efficiency for high-volume, real-time conversationsGoodFastest
Link Health SmartHIPAA-compliant, accurate, thoughtful answers for healthcareHighModerate
Link Health FastHIPAA-compliant, fast answers for high-volume healthcare conversationsGoodFast

The Health models use a selected mix of LLMs that meet healthcare compliance requirements.

Choose a model

ChooseWhen
Link Core SmartAccuracy is critical, you need complex reasoning, or quality matters more than speed
Link Core FastYou need fast replies, handle high volume, and standard accuracy is enough
Link Health SmartYou're in healthcare, need HIPAA compliance, and quality matters more than speed
Link Health FastYou're in healthcare, need HIPAA compliance, and handle high volume

Start with Link Core Fast for most use cases. It gives good quality at the best cost and speed. Move to Link Core Smart only if you need better accuracy and reasoning.

Select a model

Open the Brain tab

Open the agent from My Agents and go to the Brain tab.

Pick the model

Select the agent model from the dropdown.

Save

Save your changes. The model routes each request to the right mix of LLMs from then on.

Settings

Temperature

Temperature controls how random responses are.

RangeBehaviorUse it for
0.1 to 0.3Focused and deterministicFactual information
0.5 to 0.7Balanced creativity and consistencyMost cases (recommended)
0.8 to 1.0More creative and variedOnly when creativity is needed

Token limits

LimitWhat it controlsTypical range
Prompt tokensHow much context the agent considers4,000 to 8,000
Completion tokensMaximum response length1,000 to 2,000

Higher limits allow more context and longer replies, but cost more. Agent models already optimize token usage. Monitor usage in analytics and adjust limits to match.

Get better results

  • Write clear instructions. Be specific, give examples, set boundaries, and include relevant context. Better Instructions improve results more than switching models. See Prompting Tips.
  • Fill your knowledge base. A complete Knowledge Base means the model generates less from scratch, which improves accuracy and lowers cost.
  • Compare models. Run the same set of real customer questions against each model. Compare quality, speed, and cost, then pick the best fit.
  • Test before you switch. Test a new model in the live test chat before you use it with customers.
  • Iterate on data. Track answer quality, response time, token usage, and error rates. Switch models if quality falls short, adjust temperature if replies are too random or too rigid, and tune token limits to real usage.

Troubleshooting

Answer quality is poor
  1. Try a different model, such as Link Core Smart instead of Link Core Fast.
  2. Improve the Instructions.
  3. Add more content to the knowledge base.
  4. Adjust temperature.
Replies are slow
  1. Switch to a faster model, such as Link Core Fast or Link Health Fast.
  2. Lower the completion token limit.
  3. Make the Instructions more concise.
  4. Check for network or API issues.

Next steps

Configure Agent

Set instructions, persona, and every agent tab.

Knowledge Base

Give the model accurate content to answer from.

Prompting Tips

Write instructions that get better answers.

Actions

Let your agent take real steps for customers.