Budget Pre-Flight

When the assembled context exceeds the token budget (maxContextTokens − maxOutputTokens), Open Astra automatically trims tool schema descriptions before calling the model. This prevents silent truncation mid-inference and keeps the context within budget.

How It Works

• Calculated as: overBudget = estimatedTokens + maxOutputTokens - tokenBudget
• If overBudget > 0 and tool schemas exist, trimToolSchemas() is called
• Descriptions are shortened to 80 characters, freeing ~(original_length − 80) / 4 tokens per schema
• The agent proceeds with trimmed schemas; full schemas are restored next turn

Config

agents:
  my-agent:
    model:
      maxContextTokens: 200000
      maxOutputTokens: 4000