Context Answerer Agent¶
The context answerer retrieves stored context and synthesizes evidence-backed answers for agents and users.
The graph below is generated from the compiled LangGraph runtime.
---
config:
flowchart:
curve: linear
---
graph TD;
__start__([<p>__start__</p>]):::first
plan_retrieval(plan_retrieval)
execute_retrieval(execute_retrieval)
synthesize_answer(synthesize_answer)
__end__([<p>__end__</p>]):::last
__start__ --> plan_retrieval;
execute_retrieval --> synthesize_answer;
plan_retrieval --> execute_retrieval;
synthesize_answer --> __end__;
classDef default fill:#f2f0ff,line-height:1.2
classDef first fill-opacity:0
classDef last fill:#bfb6fc
Inputs¶
- user or agent question
- allowed project ids
- retrieval budget from config
- context-store records and search indexes
Flow¶
plan_retrievalchooses the context reads needed for the question.execute_retrievalruns the bounded deterministic store reads.synthesize_answerwrites the final answer from retrieved evidence.
Output¶
The answer includes cited context when evidence exists. When the store cannot support a claim, the answer should say that instead of inventing context.