Letta

Python Active

Stateful agents with long-term memory

GitHub

Medium Security

14.2K

Stars

v0.6.1

Latest

~120MB

RAM

~3s

Startup

Overview News Videos Releases Benchmarks Security Community

Composite Score

Argus Report scoring across six dimensions (1-10 scale)

Performance

#13 of 15

Features

#3 of 15

Security

#11 of 15

Momentum

#3 of 15

Cost Efficiency

#13 of 15

Dev Experience

#6 of 15

Cold Start Performance

Time from process launch to first response

Cold start

~3s #9

Warm response

~45ms

Memory Usage

Peak RSS during standard workload

Idle RSS

~120MB

Active RSS (10 tools)

2.1GB

Peak (stress test)

3.4GB

Token Efficiency

Tokens consumed per equivalent task completion

Simple Q&A

~280

Multi-step task

~1200

Tool calling

~220

Methodology

Benchmarks are run weekly on standardized hardware (AWS c7g.xlarge, 4 vCPU, 8GB RAM, Ubuntu 24.04). Token efficiency is measured using equivalent prompts across Claude 4.5 Sonnet. All results are averages of 10 runs with cold cache. Last updated: Feb 17, 2026.