Is it possible to make parallel call same sub agent (by default planner makes call one by one)?

[gsagrawal-binocs]

I have a root agent which has a web-search tool. Based on user input (like list of queries), root agent usages web-search tool to search. For my queries like "search a and b and c", it makes the web-search calls in sequence. Is there any way to make it parallel ?

[adk-bot]

Response from ADK Answering Agent (experimental, answer may be inaccurate)

TLDR: You can use a ParallelAgent to run multiple agents concurrently. Each agent can perform a search and write its result to a unique key in the session state. A subsequent agent in a SequentialAgent can then aggregate these results.

---

Hello! As an agent assisting with the ADK repository, I can provide some information on how to achieve parallel execution.

You've correctly observed that a standard SequentialAgent or a planner will execute tools one by one. To run them in parallel, you should use a ParallelAgent. [1]

A ParallelAgent allows you to define a set of sub-agents that will be executed concurrently. For your use case, you could create multiple agents, each responsible for a single web search. These agents would run in parallel, and their outputs can be collected and processed by a subsequent agent.

Here is a conceptual example based on the documentation of how you could structure this:

from adk.agents import LlmAgent, ParallelAgent, SequentialAgent

# An agent that performs a search and writes the result to a specific key.
search_agent_a = LlmAgent(
    # ... your LLM and tool setup ...
    output_key="response_a",
)

# A second agent that does the same, but for a different query and output key.
search_agent_b = LlmAgent(
    # ... your LLM and tool setup ...
    output_key="response_b",
)

# The ParallelAgent will run both search agents at the same time.
parallel_searches = ParallelAgent(
    sub_agents=[search_agent_a, search_agent_b]
)

# An agent to process the combined results from the parallel searches.
gather_agent = LlmAgent(
    # ... your LLM setup ...
    prompt_template="Summarize these two search results: {{response_a}} and {{response_b}}",
)

# The SequentialAgent orchestrates the end-to-end process.
root_agent = SequentialAgent(
    sub_agents=[parallel_searches, gather_agent]
)

# When you run the root_agent, the searches will be performed in parallel.

In this structure, parallel_searches will execute search_agent_a and search_agent_b concurrently. Once both have completed, the gather_agent will run, having access to the outputs of both searches via the response_a and response_b keys in the session state.

This approach allows you to parallelize your web search tasks as requested.

---

[1] https://google.github.io/adk-docs/agents/multi-agents/

[gsagrawal-binocs]

but in this case, as my query is dynamic, i cant make fix number of search agents. So i want something similar but with unknown number of search agents