Skip to main content

Update LLM Parameters

Overview​

The admin-update-llm-params tool inserts or updates parameter sets that agents reference when joining a deployment. It enforces the capability catalog, drops unsupported knobs, and applies default values so downstream agents never receive invalid combinations.

Tool ID: admin-update-llm-params

The examples use the direct runtime/operator route. Replit/Admin UI browser code should call the Admin UI BFF instead: local /admin-api/*, testing /dashboard/admin-api/*.

Input Schema​

{
  deploymentName: string;          // Required Azure deployment identifier
  modelName?: string;              // Optional fallback model name
  temperature?: number;            // 0-2
  topP?: number;                   // 0-1
  frequencyPenalty?: number;       // -2 to 2
  presencePenalty?: number;        // -2 to 2
  maxOutputTokens?: number;        // Max tokens in response
  actor?: {
    id?: string;                   // Optional actor ID
    name?: string;                 // Optional actor name, used as createdBy fallback
  };
  createdBy?: string;              // Audit-friendly identifier
  sourceSystem?: string;           // Defaults to admin-dashboard
}

Only knobs supported by the registered capability catalog entry survive; unsupported keys are silently dropped to match the deployment’s allowed parameters.

Example Usage​

cURL​

curl -s -X POST "${MASTRA_BASE_URL}/admin/tools/update-llm-params" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer ${BEARER_TOKEN}" \
  -d '{
    "data": {
      "deploymentName": "gpt-4o-mini",
      "temperature": 0.4,
      "topP": 0.95,
      "maxOutputTokens": 1200,
      "actor": {"name": "ops-bot"}
    }
  }'

Node.js​

const response = await fetch(
  `${MASTRA_BASE_URL}/admin/tools/update-llm-params`,
  {
    method: 'POST',
    headers: {
      'Content-Type': 'application/json',
      'Authorization': `Bearer ${process.env.BEARER_TOKEN}`,
    },
    body: JSON.stringify({
      data: {
        deploymentName: 'gpt-4o-mini',
        temperature: 0.4,
        maxOutputTokens: 1200,
        actor: { name: 'ops-bot' }
      }
    })
  }
);

const result = await response.json();
console.log('Stored parameter set:', result);

Output​

The tool returns the newly created LlmParam record, including:

  • deploymentName – Normalized deployment identifier
  • modelName – Catalog model name used for validation
  • temperature, topP, frequencyPenalty, presencePenalty, maxOutputTokens – Stored knobs
  • createdBy, sourceSystem – Audit fields
  • createdAt – ISO timestamp

Use Cases​

  • Seed parameter sets for new agent deployments
  • Align agent defaults with capability catalog changes
  • Update response length or randomness per deployment without editing code
  • Automate overrides when diagnosing behavioral regressions

Implementation Notes​

  • The tool consults server/model-capabilities to determine which knobs the deployment supports
  • Missing knobs receive capability-specific defaults via applyCapabilityDefaults
  • The actor object serves as a friendly audit fallback for createdBy