Harness2

Google Research framework that lets LLM agents recursively improve their own harness: prompts, tools, memory and skills.

Harness² (Harness2) is an Apache-2.0 framework from Google Cloud AI Research for recursive agent harnessing, where an agent refines the system around a frozen model: system prompt, guardrails, memory, tool guidance, subagents, skills, scripts and plugins. A Propose-Probe-Compose step tests harness edits on the current task without ground-truth feedback, and lessons are kept to guide future tasks. It supports OpenCode and Codex executors and reports gains of up to 11.5 points on WorkBuddy-Bench.

Category
Frameworks
Pricing
Open Source
Website
github.com
Visit Harness2 (opens in new tab)More frameworks

Related tools and alternatives

LangChain

langchain.com

Framework (and LangGraph) for building LLM agents.

FrameworksOpen Source

CrewAI

crewai.com

Framework for orchestrating role-playing multi-agent crews.

FrameworksOpen Source

AutoGen

microsoft.github.io

Microsoft's framework for multi-agent applications.

FrameworksOpen Source

LlamaIndex

llamaindex.ai

Framework for agents over your data.

FrameworksOpen Source

OpenAI Agents SDK

openai.github.io

Lightweight SDK for building multi-agent workflows.

FrameworksOpen Source

Pydantic AI

ai.pydantic.dev

Type-safe Python agent framework from the Pydantic team.

FrameworksOpen Source