Agentic RAG

1
from r2r import R2RClient
2
from r2r import (
3
    ThinkingEvent,
4
    ToolCallEvent,
5
    ToolResultEvent,
6
    CitationEvent,
7
    MessageEvent,
8
    FinalAnswerEvent,
9
)
10
11
# when using auth, do client.users.login(...)
12
13
# Basic RAG mode with streaming
14
response = client.retrieval.agent(
15
    message={
16
        "role": "user",
17
        "content": "What does DeepSeek R1 imply for the future of AI?"
18
    },
19
    rag_generation_config={
20
        "model": "anthropic/claude-3-7-sonnet-20250219",
21
        "extended_thinking": True,
22
        "thinking_budget": 4096,
23
        "temperature": 1,
24
        "top_p": None,
25
        "max_tokens_to_sample": 16000,
26
        "stream": True
27
    },
28
    rag_tools=["search_file_knowledge", "get_file_content"],
29
    mode="rag"
30
)
31
32
# Improved streaming event handling
33
current_event_type = None
34
for event in response:
35
    # Check if the event type has changed
36
    event_type = type(event)
37
    if event_type != current_event_type:
38
        current_event_type = event_type
39
        print() # Add newline before new event type
40
41
        # Print emoji based on the new event type
42
        if isinstance(event, ThinkingEvent):
43
            print(f"\n🧠 Thinking: ", end="", flush=True)
44
        elif isinstance(event, ToolCallEvent):
45
            print(f"\n🔧 Tool call: ", end="", flush=True)
46
        elif isinstance(event, ToolResultEvent):
47
            print(f"\n📊 Tool result: ", end="", flush=True)
48
        elif isinstance(event, CitationEvent):
49
            print(f"\n📑 Citation: ", end="", flush=True)
50
        elif isinstance(event, MessageEvent):
51
            print(f"\n💬 Message: ", end="", flush=True)
52
        elif isinstance(event, FinalAnswerEvent):
53
            print(f"\n✅ Final answer: ", end="", flush=True)
54
55
    # Print the content without the emoji
56
    if isinstance(event, ThinkingEvent):
57
        print(f"{event.data.delta.content[0].payload.value}", end="", flush=True)
58
    elif isinstance(event, ToolCallEvent):
59
        print(f"{event.data.name}({event.data.arguments})")
60
    elif isinstance(event, ToolResultEvent):
61
        print(f"{event.data.content[:60]}...")
62
    elif isinstance(event, CitationEvent):
63
        print(f"{event.data}")
64
    elif isinstance(event, MessageEvent):
65
        print(f"{event.data.delta.content[0].payload.value}", end="", flush=True)
66
    elif isinstance(event, FinalAnswerEvent):
67
        print(f"{event.data.generated_answer[:100]}...")
68
        print(f"   Citations: {len(event.data.citations)} sources referenced")

1
# Research mode with all available tools
2
response = client.retrieval.agent(
3
    message={
4
        "role": "user", 
5
        "content": "Analyze the philosophical implications of DeepSeek R1 for the future of AI reasoning"
6
    },
7
    research_generation_config={
8
        "model": "anthropic/claude-3-opus-20240229",
9
        "extended_thinking": True,
10
        "thinking_budget": 8192,
11
        "temperature": 0.2,
12
        "max_tokens_to_sample": 32000,
13
        "stream": True
14
    },
15
    research_tools=["rag", "reasoning", "critique", "python_executor"],
16
    mode="research"
17
)
18
19
# Process streaming events as shown in the previous example
20
# ...
21
22
# Research mode with computational focus
23
# This example solves a mathematical problem using the python_executor tool
24
compute_response = client.retrieval.agent(
25
    message={
26
        "role": "user", 
27
        "content": "Calculate the factorial of 15 multiplied by 32. Show your work."
28
    },
29
    research_generation_config={
30
        "model": "anthropic/claude-3-opus-20240229",
31
        "max_tokens_to_sample": 1000,
32
        "stream": False
33
    },
34
    research_tools=["python_executor"],
35
    mode="research"
36
)
37
38
print(f"Final answer: {compute_response.results.messages[-1].content}")

1
# Create a new conversation
2
conversation = client.conversations.create()
3
conversation_id = conversation.results.id
4
5
# First turn
6
first_response = client.retrieval.agent(
7
    message={"role": "user", "content": "What does DeepSeek R1 imply for the future of AI?"},
8
    rag_generation_config={
9
        "model": "anthropic/claude-3-7-sonnet-20250219",
10
        "temperature": 0.7,
11
        "max_tokens_to_sample": 1000,
12
        "stream": False
13
    },
14
    conversation_id=conversation_id,
15
    mode="rag"
16
)
17
print(f"First response: {first_response.results.messages[-1].content[:100]}...")
18
19
# Follow-up query in the same conversation
20
follow_up_response = client.retrieval.agent(
21
    message={"role": "user", "content": "How does it compare to other reasoning models?"},
22
    rag_generation_config={
23
        "model": "anthropic/claude-3-7-sonnet-20250219",
24
        "temperature": 0.7,
25
        "max_tokens_to_sample": 1000,
26
        "stream": False
27
    },
28
    conversation_id=conversation_id,
29
    mode="rag"
30
)
31
print(f"Follow-up response: {follow_up_response.results.messages[-1].content[:100]}...")
32
33
# The agent maintains context, so it knows "it" refers to DeepSeek R1