mirror of
https://github.com/tiennm99/DocsGPT.git
synced 2026-10-04 16:13:23 +00:00
Alex
6b9b193337
A one-shot graph ranking diffuses over the whole neighbourhood; a question whose answer sits two hops away is better served by following the edges. The graph_search tool gives an agent search_entities, get_relationships and read_entity_pages over its graph sources. On a multi-hop corpus where the bridging entity is never named, answers went from 0/10 with classic vector retrieval to 10/10 with the tool, end to end through /stream. The tool has no setting of its own. It is offered exactly where the agent can already search: agentic and research agents always, a classic agent only for sources exposed as a search tool. A graph source left at prefetch is used for ranking only. Tests now pin GRAPHRAG_ENABLED to its shipped default, as CI has: with a dev .env enabling it, every agent test's graph check read the developer's real database and left a pool to it behind.
61 lines
2.0 KiB
Python
61 lines
2.0 KiB
Python
import logging
|
|
from typing import Dict, Generator, Optional
|
|
|
|
from docsgpt.agents.base import BaseAgent
|
|
from docsgpt.agents.tools.graph_search import add_graph_search_tool
|
|
from docsgpt.agents.tools.internal_search import add_internal_search_tool
|
|
from docsgpt.agents.tools.wiki import add_wiki_tool
|
|
from docsgpt.logging import LogContext
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
class AgenticAgent(BaseAgent):
|
|
"""Agent where the LLM controls retrieval via tools.
|
|
|
|
Unlike ClassicAgent which pre-fetches docs into the prompt,
|
|
AgenticAgent gives the LLM an internal_search tool so it can
|
|
decide when, what, and whether to search.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
retriever_config: Optional[Dict] = None,
|
|
wiki_config: Optional[Dict] = None,
|
|
*args,
|
|
**kwargs,
|
|
):
|
|
super().__init__(*args, **kwargs)
|
|
self.retriever_config = retriever_config or {}
|
|
self.wiki_config = wiki_config or {}
|
|
|
|
def _gen_inner(
|
|
self, query: str, log_context: LogContext
|
|
) -> Generator[Dict, None, None]:
|
|
tools_dict = self.tool_executor.get_tools()
|
|
add_internal_search_tool(tools_dict, self.retriever_config)
|
|
add_graph_search_tool(tools_dict, self.retriever_config)
|
|
if self.wiki_config:
|
|
add_wiki_tool(tools_dict, self.wiki_config)
|
|
self._prepare_tools(tools_dict)
|
|
|
|
# 4. Build messages (prompt has NO pre-fetched docs)
|
|
messages = self._build_messages(self.prompt, query)
|
|
|
|
# 5. Call LLM — the handler manages the tool loop
|
|
llm_response = self._llm_gen(messages, log_context)
|
|
|
|
yield from self._handle_response(
|
|
llm_response, tools_dict, messages, log_context
|
|
)
|
|
|
|
# 6. Collect sources from internal search tool results
|
|
self._collect_internal_sources()
|
|
|
|
yield {"sources": self.retrieved_docs}
|
|
yield {"tool_calls": self._get_truncated_tool_calls()}
|
|
|
|
log_context.stacks.append(
|
|
{"component": "agent", "data": {"tool_calls": self.tool_calls.copy()}}
|
|
)
|