ToolAha
Submit a tool
Explore / Coding / Harness2
github.com

Harness2

Research framework for improving the harness around a fixed language model

Visit official site

Resource information

Resource type
Project
AI relationship
AI native
Repository subpath
Not verified
License
Apache-2.0
License classification
Verified Open Source
License verified date
2026-10-19
License review findings
Reviewed the fixed repository LICENSE; third-party dependencies and benchmarks retain upstream terms.
Verification status
Official statement reviewed
Last verified (UTC)
2026-10-19
Languages
  • Python
Frameworks
  • OpenCode and optional Codex executor paths documented; not independently tested
Deployment
  • Dedicated Linux experiment environment described in official setup
Self hosting
Supported
Requirements
  • Python 3.12 for benchmark runs; optional benchmark-specific dependencies
  • Model provider access and a budget for model-backed experiments
Limitations
  • Research implementation, no supported Google product guarantee
  • Runtime execution and security have not been verified by ToolAha
Maintenance status
Not verified
Reference price
Pricing not verifiedSee official plans
Free access
Not verifiedQuota and limits apply
API
Not verifiedDeveloper terms may differ
Card for free / trial
Not verifiedOfficial information only
Profile reviewed: Not verifiedFind similar-purpose tools
01 / ABOUT

What it does

Harness² is a research framework in google-research for investigating recursive agent harness improvement. It provides task-level refinement and a library of adopted harnesses and supported lessons. This listing is a documentation review, not a runtime test or a supported Google service. See the separate explainer for selection advice.

02 / CAPABILITIES

Features

  • 01

    Task-level Propose–Probe–Compose refinement

  • 02

    Retained adopted harnesses and evidence-supported lessons

  • 03

    Documented benchmark runners and execution artifacts

03 / USE CASES

Who it is for

  • Agent infrastructure research
  • Controlled experiments on repeated professional tasks

These scenarios and steps are editorial guidance based on the official product scope, not independent hands-on results.

Report a change

Something changed?

Direct alternatives

Resources of the same type and use case; review their scope before choosing.

View similar tools
LangChainCoding

Build model-backed applications with reusable integrations.

Project
WebsitePricing not verified
LangGraphCoding

Orchestrate stateful agent workflows with durable execution.

Project
WebsitePricing not verified

Related resources

AIReiterCoding

One AI API for image, video, and text generation

Tools / Apps
API
WebsiteUsage based · see official plans
AI MagicxCoding

A multi-model workspace for chat, coding and creative generation

Tools / Apps
Website$19.00 / Month
BananaCoding

Managed GPU inference with autoscaling and deployment observability

Tools / Apps
Website$1,200.00 / Month