letta-development-guide
Audited by Runlayer on Feb 21, 2026
Malicious tool definition detected
Tool: references/description-patterns.md Description: # Memory Block Description Patterns Good descriptions guide the agent on when to read and update blocks.
Malicious tool definition detected
Tool: references/model-recommendations.md Description: # Model Selection Guide ## Production Recommendations ### High-Quality Reasoning - **GPT-4o**: Best overall, reliable tool calling (`openai/gpt-4o`) - **Claude Sonnet 4**: Excellent reasoning, strong with memory tools (`anthropic/claude-sonnet-4-20250514`) - **Gemini 2.0 Flash**: Fast, good capability (`google/gemini-2.0-flash`) ### Cost-Efficient - **GPT-4o-mini**: Best balance of cost and capability (`openai/gpt-4o-mini`) - **Claude Haiku
Malicious tool definition detected
Tool: references/size-management.md Description: # Memory Block Size Management ## Character Limits **Typical limits per block:** - 2000-5000 characters depending on configuration - Total core memory should stay under 80% of context window **Monitor size:** - Via ADE: shows character count per block - Via API: check block `value` length ## Growth Strategies ### 1.
Malicious tool definition detected
tools = [ "memory_insert", "domain_specific_tool" ] ``` **Note:** Use `send_message_to_agent_and_wait_for_reply` for agent-to-agent communication.
Tool passed security scan
Tool passed security scan
Passed Files (3)Click to expand
Tool passed security scan
Tool passed security scan
Tool passed security scan