Skip to main content
The Hypothesis Agent synthesizes findings from completed tasks into scientific hypotheses with inline citations. It can create new hypotheses or update existing ones based on new evidence.

​
Function Signature

// src/agents/hypothesis/index.ts
export async function hypothesisAgent(input: {
  objective: string;
  message: Message;
  conversationState: ConversationState;
  completedTasks: PlanTask[];
}): Promise<HypothesisResult>;

type HypothesisResult = {
  hypothesis: string;
  thought?: string;
  start: string;
  end: string;
  mode: "create" | "update";
};

​
Modes

​
Create Mode

Generates a new hypothesis when none exists: 
const result = await hypothesisAgent({
  objective: "Investigate mechanisms of cellular senescence",
  message,
  conversationState: {
    values: {
      currentHypothesis: undefined  // No existing hypothesis
    }
  },
  completedTasks: [
    {
      type: "LITERATURE",
      objective: "Search for p53 and senescence",
      output: "(p53 regulates senescence via p21)[10.1038/nature.2009.1234]..."
    }
  ]
});

// Returns:
// {
//   hypothesis: "We hypothesize that p53-mediated p21 activation is a key regulator of cellular senescence...",
//   mode: "create"
// }

​
Update Mode

Updates existing hypothesis with new findings: 
const result = await hypothesisAgent({
  objective: "Investigate mechanisms of cellular senescence",
  message,
  conversationState: {
    values: {
      currentHypothesis: "We hypothesize that p53-mediated p21 activation regulates senescence..."
    }
  },
  completedTasks: [
    {
      type: "ANALYSIS",
      objective: "Analyze p21 expression levels",
      output: "Analysis shows 3.2-fold increase in p21 expression (p < 0.001)..."
    }
  ]
});

// Returns:
// {
//   hypothesis: "Our analysis confirms p53-mediated p21 activation as a key regulator. We now extend this to propose...",
//   mode: "update"
// }

​
Hypothesis Structure

Generated hypotheses follow scientific conventions:
1

Context

Brief background establishing the research question
2

Hypothesis statement

Clear, testable proposition with inline citations
3

Supporting evidence

Key findings from completed tasks
4

Predictions

Expected outcomes if hypothesis is correct

​
Example Output

Based on literature search and expression analysis, we hypothesize that 
cellular senescence is primarily regulated through the p53-p21 axis. 
(Previous studies show p53 directly activates p21 transcription)[10.1038/nature.2009.1234], 
and our analysis confirms a 3.2-fold increase in p21 expression in senescent cells (p < 0.001).

We predict that p53 knockdown will reduce p21 levels and delay senescence onset. 
Further analysis of downstream targets (p16INK4a, SASP factors) should reveal 
additional regulatory mechanisms.

​
Usage Example

// src/routes/chat.ts
const needsHypothesis = await requiresHypothesis(
  message,
  allLiteratureOutput,
  messageId
);

if (needsHypothesis && completedTasks.length > 0) {
  const hypothesisResult = await hypothesisAgent({
    objective: planningResult.currentObjective,
    message,
    conversationState,
    completedTasks
  });
  
  // Update conversation state
  conversationState.values.currentHypothesis = hypothesisResult.hypothesis;
  
  await updateConversationState(
    conversationState.id,
    conversationState.values
  );
}

​
Citation Format

The agent preserves inline citations from task outputs: 
(claim)[DOI or URL]
Examples:
  • (Rapamycin extends lifespan in mice)[10.1038/nature.2009.1234]
  • (mTOR inhibition reduces protein synthesis)[https://example.com/paper]
These citations are maintained through the entire workflow and appear in the final paper generation.

​
Configuration

​
HYP_LLM_PROVIDER
string
default:"openai"
LLM provider: openai, anthropic, google, or openrouter
​
HYP_LLM_MODEL
string
default:"gpt-5"
Model name for hypothesis generation

​
When to Generate Hypothesis

The system determines if a hypothesis is needed using an LLM check: 
// src/routes/chat.ts:requiresHypothesis
const needsHypothesis = await requiresHypothesis(
  question,
  literatureResults
);
A hypothesis IS needed if:
  • Question asks about mechanisms, predictions, or causal relationships
  • Question requires synthesizing multiple sources into novel insight
  • Question is exploratory and needs a testable proposition
A hypothesis IS NOT needed if:
  • Question asks for factual information or definitions
  • Question can be answered directly from literature
  • Question is a simple lookup or clarification

​
Integration with Workflow

The hypothesis flows through the research cycle:

Reflection Agent

Updates world state with insights

Discovery Agent

Identifies novel claims

Deep Research

Iterative research cycle

Build docs developers (and LLMs) love

Get started for freeTalk to us