a digital person for bluesky
cameron.stream
1"""
2Letta Multi-Agent Financial Analysis Team Example
3================================================
4
5This example demonstrates how to create a team of autonomous AI agents that work together
6to analyze financial markets, each with their own specialized knowledge and shared data.
7
8Key Concepts:
9------------
101. **Memory Blocks**: Letta agents use "memory blocks" to store and access information.
11 Think of these as persistent knowledge bases that agents can read and write to.
12
132. **Shared vs Private Memory**:
14 - Shared blocks (like market data) are attached to multiple agents
15 - Private blocks (like proprietary models) are only attached to specific agents
16
173. **Agent Communication**: Agents can send messages to each other using tools,
18 allowing them to share findings and coordinate analysis.
19
20Architecture:
21------------
22- 3 Quant Agents: Each analyzes markets using different strategies
23 - Momentum Quant: Uses price momentum signals
24 - Value Quant: Uses fundamental value metrics
25 - ML Quant: Uses machine learning predictions
26
27- 1 Portfolio Manager: Receives and consolidates findings from all quants
28
29Memory Design:
30-------------
31- Global Shared Memory: "market-data" - All agents see the same market data
32- Local Private Memory: Each quant has their own model/strategy block
33- PM Aggregation Memory: "aggregated-signals" - Stores consolidated findings
34
35Workflow:
36--------
371. All agents read shared market data
382. Each quant analyzes using their unique approach
393. Quants send findings to the Portfolio Manager
404. PM aggregates insights and produces final report
41"""
42
43from letta_client import Letta
44import os
45
46client = Letta(token=os.environ["LETTA_API_KEY"])
47project_id = "your-project-id"
48
49# 1. Create shared memory blocks (global data)
50market_data = client.blocks.create(
51 project_id=project_id,
52 label="market-data",
53 value="S&P500: 4800, VIX: 15.2, DXY: 102.5...",
54 description="Shared financial market data"
55)
56
57# 2. Create individual quant memory blocks
58quant1_model = client.blocks.create(
59 project_id=project_id,
60 label="quant1-momentum-model",
61 value="Momentum factor model: 12-month return signals...",
62 description="Quant 1's proprietary momentum model"
63)
64
65quant2_model = client.blocks.create(
66 project_id=project_id,
67 label="quant2-value-model",
68 value="Value factor model: P/E ratios, book values...",
69 description="Quant 2's value investing model"
70)
71
72quant3_model = client.blocks.create(
73 project_id=project_id,
74 label="quant3-ml-model",
75 value="LSTM predictions, feature importance weights...",
76 description="Quant 3's ML model"
77)
78
79# 3. Create agents with mixed memory (shared + individual)
80quant1 = client.agents.create(
81 project_id=project_id,
82 name="quant-momentum",
83 block_ids=[market_data.id, quant1_model.id] # Shared + Individual
84)
85
86quant2 = client.agents.create(
87 project_id=project_id,
88 name="quant-value",
89 block_ids=[market_data.id, quant2_model.id]
90)
91
92quant3 = client.agents.create(
93 project_id=project_id,
94 name="quant-ml",
95 block_ids=[market_data.id, quant3_model.id]
96)
97
98# 4. Create portfolio manager with access to all findings
99pm_findings = client.blocks.create(
100 project_id=project_id,
101 label="aggregated-signals",
102 value="", # Will be populated by quant reports
103 description="Consolidated findings from all quants"
104)
105
106portfolio_manager = client.agents.create(
107 project_id=project_id,
108 name="portfolio-manager",
109 block_ids=[market_data.id, pm_findings.id],
110 tags=["pm"] # Tag for message routing
111)
112
113# 5. Give quants ability to send findings to PM
114send_to_pm_tool = client.tools.list(name="send_message_to_agents_matching_tags")[0]
115for quant in [quant1, quant2, quant3]:
116 client.agents.tools.attach(agent_id=quant.id, tool_id=send_to_pm_tool.id)
117
118# 6. Run analysis
119for quant in [quant1, quant2, quant3]:
120 response = client.agents.messages.create(
121 agent_id=quant.id,
122 messages=[{"role": "user", "content": "Analyze market and send findings to PM (tag: pm)"}]
123 )