Skip to main content
Deep Research gives agents the ability to autonomously search the web, read sources, extract knowledge, and synthesize findings across multiple research waves.

​
Enable Deep Research

from definable.agent import Agent

agent = Agent(model="gpt-4o", deep_research=True)

output = await agent.arun("What are the latest advances in quantum computing?")
print(output.content)  # Grounded response with citations

​
Custom Configuration

from definable.agent.research import DeepResearchConfig

agent = Agent(
    model="gpt-4o",
    deep_research=DeepResearchConfig(
        depth="deep",                  # "quick", "standard", or "deep"
        search_provider="duckduckgo",  # or "google", "serpapi"
        max_sources=30,
        max_waves=5,
        include_citations=True,
    ),
)

​
Depth Presets

PresetWavesSourcesUse case
quick18Fast fact-checking
standard315Balanced research (default)
deep530Comprehensive analysis

​
How It Works

  1. Wave 1: Generate search queries from the user’s question.
  2. Search: Execute queries via the configured search provider.
  3. Read: Fetch and parse source pages concurrently.
  4. Extract: Use a compression model to extract Compact Knowledge Units (CKUs).
  5. Evaluate: Score relevance, detect contradictions, identify gaps.
  6. Repeat: Generate follow-up queries and research further (waves 2+).
  7. Inject: Final CKUs are injected into the agent’s system prompt.

​
Configuration Reference

​
depth
str
default:"standard"
Research depth: "quick", "standard", or "deep".
​
search_provider
str
default:"duckduckgo"
Search backend: "duckduckgo", "google", or "serpapi".
​
max_sources
int
default:"15"
Maximum unique sources across all waves.
​
max_waves
int
default:"3"
Maximum research waves.
​
include_citations
bool
default:"true"
Include source citations in the research context.
​
trigger
str
default:"always"
When to research: "always", "auto" (model decides), "tool" (explicit call).

​
Imports

from definable.agent.research import DeepResearchConfig