Personal AI, On Personal Devices
Python
# NativeReActAgent on 5 use-case benchmarks.
# Thought-Action-Observation loop — classic ReAct paradigm.
[meta]
name = "use-case-react"
description = "Use-case benchmarks — NativeReActAgent (Thought-Action-Observation)"
[defaults]
temperature = 0.0
max_tokens = 2048
[judge]
model = "gpt-5-mini-2025-08-07"
[run]
max_workers = 4
output_dir = "results/use-cases-react/"
seed = 42
[[models]]
name = "qwen3:8b"
engine = "ollama"
[[benchmarks]]
name = "email_triage"
backend = "jarvis-agent"
agent = "native_react"
tools = ["think"]
max_samples = 30
[[benchmarks]]
name = "morning_brief"
backend = "jarvis-agent"
agent = "native_react"
tools = ["think", "web_search"]
max_samples = 15
[[benchmarks]]
name = "research_mining"
backend = "jarvis-agent"
agent = "native_react"
tools = ["think", "web_search"]
max_samples = 30
[[benchmarks]]
name = "knowledge_base"
backend = "jarvis-agent"
agent = "native_react"
tools = ["think"]
max_samples = 30
[[benchmarks]]
name = "coding_task"
backend = "jarvis-agent"
agent = "native_react"
tools = ["think", "code_interpreter"]
max_samples = 29