Mastra•5d ago

network() stream doesn't forward reasoning events from sub-agents

Environment:

@mastra/core: beta
@mastra/ai-sdk: beta
Model: anthropic/claude-haiku-4-5

@mastra/core: beta
@mastra/ai-sdk: beta
Model: anthropic/claude-haiku-4-5

@mastra/core: beta
@mastra/ai-sdk: beta
Model: anthropic/claude-haiku-4-5

@mastra/core: beta
@mastra/ai-sdk: beta
Model: anthropic/claude-haiku-4-5

What I Expected: When using agent.network() with extended thinking enabled on sub-agents, I expected reasoning-delta events to be forwarded in the stream (similar to how agent.stream() works with sendReasoning: true).

What Actually Happens: The network stream emits agent-execution-event-text-delta but NOT agent-execution-event-reasoning-* events. Reasoning tokens are consumed internally, but the actual reasoning content is not accessible.

Minimal Reproduction:

// chat-agent.ts - Parent agent with sub-agent
export const chatAgent = new Agent({
  id: 'chat-agent',
  instructions: {
    role: 'system',
    content: 'You are a router...',
    providerOptions: {
      anthropic: {
        thinking: { type: 'enabled', budgetTokens: 4000 }
      }
    }
  },
  model: 'anthropic/claude-haiku-4-5',
  agents: { strategist: strategistAgent }
});
// Test script
const stream = await chatAgent.network('Hello', { memory: { thread: 'test' } });
for await (const chunk of stream) {
  console.log(chunk.type);
  // Never see: agent-execution-event-reasoning-delta
  // Do see: agent-execution-event-text-delta
}

// chat-agent.ts - Parent agent with sub-agent
export const chatAgent = new Agent({
  id: 'chat-agent',
  instructions: {
    role: 'system',
    content: 'You are a router...',
    providerOptions: {
      anthropic: {
        thinking: { type: 'enabled', budgetTokens: 4000 }
      }
    }
  },
  model: 'anthropic/claude-haiku-4-5',
  agents: { strategist: strategistAgent }
});
// Test script
const stream = await chatAgent.network('Hello', { memory: { thread: 'test' } });
for await (const chunk of stream) {
  console.log(chunk.type);
  // Never see: agent-execution-event-reasoning-delta
  // Do see: agent-execution-event-text-delta
}

// chat-agent.ts - Parent agent with sub-agent
export const chatAgent = new Agent({
  id: 'chat-agent',
  instructions: {
    role: 'system',
    content: 'You are a router...',
    providerOptions: {
      anthropic: {
        thinking: { type: 'enabled', budgetTokens: 4000 }
      }
    }
  },
  model: 'anthropic/claude-haiku-4-5',
  agents: { strategist: strategistAgent }
});
// Test script
const stream = await chatAgent.network('Hello', { memory: { thread: 'test' } });
for await (const chunk of stream) {
  console.log(chunk.type);
  // Never see: agent-execution-event-reasoning-delta
  // Do see: agent-execution-event-text-delta
}

// chat-agent.ts - Parent agent with sub-agent
export const chatAgent = new Agent({
  id: 'chat-agent',
  instructions: {
    role: 'system',
    content: 'You are a router...',
    providerOptions: {
      anthropic: {
        thinking: { type: 'enabled', budgetTokens: 4000 }
      }
    }
  },
  model: 'anthropic/claude-haiku-4-5',
  agents: { strategist: strategistAgent }
});
// Test script
const stream = await chatAgent.network('Hello', { memory: { thread: 'test' } });
for await (const chunk of stream) {
  console.log(chunk.type);
  // Never see: agent-execution-event-reasoning-delta
  // Do see: agent-execution-event-text-delta
}

Verification:

agent.stream() with toAISdkStream({ sendReasoning: true }) → reasoning events visible

agent.network() → no reasoning events in stream

MastraAgentNetworkStream.usage has reasoningTokens property, suggesting it should be tracked

Question: Is this intentional? If so, is there a recommended way to access sub-agent reasoning in network streams?

Mastra Triager•12/12/25, 12:02 PM

Created GitHub issue: https://github.com/mastra-ai/mastra/issues/11123

If you're experiencing an error, please provide a minimal reproducible example whenever possible to help us resolve it quickly.

Thank you for helping us improve Mastra!

_roamin_•12/12/25, 5:53 PM

Hey @Sunil Kumar K S ! This looks like a bug, I'm looking into it

_roamin_•12/12/25, 6:27 PM

Hey @Sunil Kumar K S ! I do see the sub-agent's agent-execution-event-reasoning-deltaagent-execution-event-reasoning-delta chunks on my end, are you sure you have thinking/reasoning enabled on your sub-agent?
This is the code I used to try to reproduce:

import { Agent } from "@mastra/core/agent";
import { Memory } from "@mastra/memory";
import { LibSQLStore } from "@mastra/libsql";

export const calculationAgent = new Agent({
  id: "calculationAgent",
  name: "sampleAgent",
  instructions: "You are a helpful assistant that is used to do calculation.",
  model: "openrouter/anthropic/claude-haiku-4.5",
  defaultOptions: {
    providerOptions: {
      openrouter: {
        reasoning: {
          max_tokens: 2000,
          enabled: true,
        },
      },
    },
  },
});

const networkAgent = new Agent({
  id: "networkAgent",
  name: "networkAgent",
  instructions:
    "You have helper agents to assist you in doing calculations. Use them when necessary.",
  model: "openrouter/anthropic/claude-haiku-4.5",
  agents: {
    calculationAgent,
  },
  memory: new Memory({
    storage: new LibSQLStore({
      id: "memory-storage",
      url: ":memory:",
    }),
  }),
  defaultOptions: {
    providerOptions: {
      openrouter: {
        reasoning: {
          max_tokens: 2000,
          enabled: true,
        },
      },
    },
  },
});

const result = await networkAgent.network("what is 4+4?", {});

for await (const chunk of result) {
  console.log(chunk);
}

import { Agent } from "@mastra/core/agent";
import { Memory } from "@mastra/memory";
import { LibSQLStore } from "@mastra/libsql";

export const calculationAgent = new Agent({
  id: "calculationAgent",
  name: "sampleAgent",
  instructions: "You are a helpful assistant that is used to do calculation.",
  model: "openrouter/anthropic/claude-haiku-4.5",
  defaultOptions: {
    providerOptions: {
      openrouter: {
        reasoning: {
          max_tokens: 2000,
          enabled: true,
        },
      },
    },
  },
});

const networkAgent = new Agent({
  id: "networkAgent",
  name: "networkAgent",
  instructions:
    "You have helper agents to assist you in doing calculations. Use them when necessary.",
  model: "openrouter/anthropic/claude-haiku-4.5",
  agents: {
    calculationAgent,
  },
  memory: new Memory({
    storage: new LibSQLStore({
      id: "memory-storage",
      url: ":memory:",
    }),
  }),
  defaultOptions: {
    providerOptions: {
      openrouter: {
        reasoning: {
          max_tokens: 2000,
          enabled: true,
        },
      },
    },
  },
});

const result = await networkAgent.network("what is 4+4?", {});

for await (const chunk of result) {
  console.log(chunk);
}

import { Agent } from "@mastra/core/agent"; import { Memory } from "@mastra/memory"; import { LibSQLStore } from "@mastra/libsql"; export const calculationAgent = new Agent({ id: "calculationAgent", name: "sampleAgent", instructions: "You are a helpful assistant that is used to do calculation.", model: "openrouter/anthropic/claude-haiku-4.5", defaultOptions: { providerOptions: { openrouter: { reasoning: { max_tokens: 2000, enabled: true, }, }, }, }, }); const networkAgent = new Agent({ id: "networkAgent", name: "networkAgent", instructions: "You have helper agents to assist you in doing calculations. Use them when necessary.", model: "openrouter/anthropic/claude-haiku-4.5", agents: { calculationAgent, }, memory: new Memory({ storage: new LibSQLStore({ id: "memory-storage", url: ":memory:", }), }), defaultOptions: { providerOptions: { openrouter: { reasoning: { max_tokens: 2000, enabled: true, }, }, }, }, }); const result = await networkAgent.network("what is 4+4?", {}); for await (const chunk of result) { console.log(chunk); }