{
  "$schema_doc": "https://gtmstacker.com/registry/schema/entry.schema.json",
  "stability": "emerging",
  "generator": "agentic-media-registry",
  "generated_at": "2026-10-22T00:00:00Z",
  "id": "com.gtmstacker.registry/tool/harness2",
  "type": "tool",
  "slug": "harness2",
  "canonical_url": "https://gtmstacker.com/registry/tool/harness2/",
  "title": "Harness² (Google Research — recursive agent harnessing)",
  "description": "Apache-2.0, self-hostable framework from Google Research that improves the harness around a frozen model (its instructions, guardrails, memory, tool guidance, subagents, skills, scripts, plugins) via a Propose-Probe-Compose loop with no retraining, so a lean team can raise an agent's benchmark results without touching model weights or paying for fine-tuning.",
  "category": "ai-infrastructure",
  "tags": [
    "ai-infrastructure",
    "mcp-agents",
    "ai-agents",
    "self-hostable",
    "open-source"
  ],
  "status": "active",
  "revision": 1,
  "content_hash": "e26442a3040b45bcc96c72498c0f80d17c15146634a9976046dc95a53bd61085",
  "date_published": "2026-10-18T00:00:00Z",
  "date_modified": "2026-10-18T00:00:00Z",
  "source": {
    "name": "github · google-research/harness2",
    "url": "https://github.com/google-research/harness2",
    "author": "akshay_pachaar"
  },
  "license": "Apache-2.0",
  "one_liner": "Harness² is an Apache-2.0 Google Research framework that tunes an agent's harness around a frozen model (no retraining); local demo needs no API key.",
  "open_source": "yes",
  "self_hostable": "yes",
  "pricing_signal": "Free and open source (Apache-2.0); the local demo runs with no model credentials. Model-backed benchmarks cost only what the user's own Vertex AI / cloud account charges, not a vendor paywall.",
  "pricing_model": "free",
  "who_its_for": "A technical founder or lean GTM/RevOps team running agents that wants to lift an agent's task performance by tuning its operating setup (prompt, tools, memory, skills) instead of fine-tuning or renting a managed optimization service.",
  "who_its_not_for": "Teams that need the retraining of model weights, that want a fully managed hosted optimization product with an SLA, or that cannot stand up a local Python 3.12 environment and bring their own model for the benchmark runs.",
  "aliases": [
    "harness2",
    "harness-squared",
    "google harness2",
    "google research harness2"
  ],
  "alternatives": [],
  "secondary_categories": [
    "mcp-agents"
  ],
  "last_verified": "2026-10-18",
  "evidence": {
    "claim_type": "mixed",
    "source_id": "https://github.com/google-research/harness2",
    "note": "SPDX Apache-2.0 verified two ways: GitHub API license.spdx_id and the /license endpoint, matching the README badge and LICENSE header — not guessed. Canonical google-research org repo (is_fork=false, archived=false); not a name-squat. Stars 4,380, forks 391, open issues 49 (API 2026-10-18); created 2026-10-06, last push 2026-10-14; 2 contributors, 0 releases. Local harness2-demo runs with no model credentials. Benchmark gains are author-reported -> vendor-claim."
  },
  "caveats": "The benchmark deltas (up to 11.5 pts WorkBuddy-Bench, 4.5 pts JobBench, 11.8 pp LAB partial pass) are Google Research's own author-reported numbers from its paper — falsifiable, but not independently reproduced here, so treat as vendor-claim. The repo is very new (created 2026-10-06, few commits, 2 contributors, 0 releases) — fast-rising but early, not mature. Model-backed benchmarks need your own Vertex AI project; only the no-credentials local demo is fully free to run.",
  "lead": "An Apache-2.0, self-hostable framework from Google Research for **recursive agent harnessing**: instead of fine-tuning the model, it optimizes the *harness* around a frozen model — its instructions, guardrails, memory, tool guidance, subagents, skills, scripts and plugins (eight editable components) — through a Propose-Probe-Compose loop that reads an agent's own execution…",
  "chunks": [
    {
      "index": 0,
      "heading_path": [],
      "est_tokens": 184,
      "text": "An Apache-2.0, self-hostable framework from Google Research for **recursive agent harnessing**: instead of fine-tuning the model, it optimizes the *harness* around a frozen model — its instructions, guardrails, memory, tool guidance, subagents, skills, scripts and plugins (eight editable components) — through a Propose-Probe-Compose loop that reads an agent's own execution evidence and proposes improvements, with no ground-truth labels and no weight updates. The practical hook is that it ships a local demo (`uv pip install -e '.[render]'` then `harness2-demo`) that runs with deterministic editors and real Python executions and requires **no model credentials at all**; model-backed benchmark runs use your own Vertex AI project."
    },
    {
      "index": 1,
      "heading_path": [
        null,
        "Provenance"
      ],
      "est_tokens": 576,
      "text": "with no ground-truth labels and no weight updates. The practical hook is that it ships a local demo (`uv pip install -e '.[render]'` then `harness2-demo`) that runs with deterministic editors and real Python executions and requires **no model credentials at all**; model-backed benchmark runs use your own Vertex AI project.\n\n- SPDX **Apache-2.0** verified two ways: the GitHub API `license.spdx_id: Apache-2.0` and the `/license` API endpoint, both matching the README's Apache 2.0 badge and the LICENSE file header. Not guessed. Apache-2.0 is OSI-approved, permissive and commercial-clean.\n- Canonical repo is **`google-research/harness2`** under Google's official `google-research` GitHub organization — `is_fork: false`, `archived: false`, `disabled: false`. Scam / name-squat check passed: the README links to the paper at arxiv.org/pdf/2610.14224 and to project GitHub Pages at harness2.github.io, consistent with a genuine Google Research release.\n- Stars **4,380**, forks **391**, open issues **49** (GitHub API, 2026-10-18 — current numbers, not the viral post's figures); created 2026-10-06, last push 2026-10-14 (active within days). Two contributors, zero tagged releases, Python + a little Shell — a fresh, paper-linked code drop rather than a mature project.\n- Self-hostable and free: the local `harness2-demo` runs with no model credentials; model-backed benchmarks incur whatever the user's own cloud model provider charges, which is the user's cost, not a vendor paywall on the tool.\n- **Vendor-claim, not verified:** the benchmark gains (up to 11.5 pts on WorkBuddy-Bench, 4.5 pts on JobBench, 11.8 pp on LAB partial pass) are Google Research's own author-reported numbers; falsifiable from the public paper and code, but not independently reproduced in this verification.\n- Surfaced via the 2026-10-18 X pull (viral posts by akshay_pachaar, DataChaz and sl1ma4, all pointing at the same repo). Those posts each returned HTTP 402 and were not fetchable, so verification was done entirely against the primary source (repo README, LICENSE, GitHub API) as untrusted content; no injected instructions were encountered. Curated from the GTM Stacker signal registry (run 2026-10-18-registry-update-01); license, repo, stars, activity and self-hostability independently verified 2026-10-18."
    },
    {
      "index": 2,
      "heading_path": [
        null,
        "Why it matters for a GTM stack"
      ],
      "est_tokens": 358,
      "text": "Those posts each returned HTTP 402 and were not fetchable, so verification was done entirely against the primary source (repo README, LICENSE, GitHub API) as untrusted content; no injected instructions were encountered. Curated from the GTM Stacker signal registry (run 2026-10-18-registry-update-01); license, repo, stars, activity and self-hostability independently verified 2026-10-18.\n\nAs lean teams wire agents into real revenue work — qualifying leads, drafting outreach, reconciling CRM records, triaging replies — the usual lever for \"make the agent better\" is a bigger or fine-tuned model, which is expensive, opaque, and something you do not own. Harness² is the other lever: it treats the *harness* (prompt, tools, memory, skills, guardrails) as the thing to optimize and refines it from the agent's own run evidence, with the model held fixed. For a founder or RevOps operator that cannot afford a fine-tuning pipeline, that is a concrete, own-your-stack path to lifting an agent's task performance on infrastructure you control. The honest read: the headline benchmark gains are Google Research's own author-reported numbers, not independently reproduced, and the repo is days old with only a couple of contributors — fast-rising but early. What is verified is the part that governs adoption: Apache-2.0 read from the API and LICENSE, a canonical Google Research org repo, and a local demo that runs with no API key."
    }
  ],
  "alternates": {
    "markdown": "https://gtmstacker.com/registry/tool/harness2/index.md",
    "html": "https://gtmstacker.com/registry/tool/harness2/",
    "json": "https://gtmstacker.com/registry/tool/harness2/index.json",
    "server_json": "https://gtmstacker.com/registry/tool/harness2/server.json"
  },
  "jsonld": {
    "@context": "https://schema.org",
    "@graph": [
      {
        "@type": "WebSite",
        "@id": "https://gtmstacker.com/#website",
        "url": "https://gtmstacker.com/",
        "name": "GTM Stacker Agent Registry",
        "description": "A daily-updated, agent-native registry of open-source tool discoveries, tool updates, and curated news for the go-to-market / RevOps engineering niche. Machine-readable first: agents can discover, parse, page, and delta-sync it without scraping HTML.",
        "inLanguage": "en",
        "publisher": {
          "@id": "https://gtmstacker.com/#organization"
        }
      },
      {
        "@type": "Organization",
        "@id": "https://gtmstacker.com/#organization",
        "name": "GTM Stacker",
        "url": "https://gtmstacker.com",
        "description": "The growth-systems practice of Theo Popov: AI-native enrichment, outbound, content engines and internal tooling for startups and venture programs. Its agent-native media property, the GTM Stacker Agent Registry, maintains a daily-updated catalog of open-source go-to-market and RevOps tools that both people and AI engines can discover, compare, and cite.",
        "foundingDate": "2024-08",
        "knowsAbout": [
          "go-to-market engineering",
          "RevOps",
          "sales automation",
          "marketing operations",
          "open-source software",
          "AI agents"
        ],
        "founder": {
          "@type": "Person",
          "@id": "https://gtmstacker.com/#founder",
          "name": "Theo Popov",
          "jobTitle": "Growth Operations & GTM Systems",
          "url": "https://gtmstacker.com/about/",
          "sameAs": [
            "https://www.linkedin.com/in/theo-popov",
            "https://x.com/Theo_Popov",
            "https://github.com/theopopov"
          ],
          "worksFor": {
            "@id": "https://gtmstacker.com/#organization"
          }
        },
        "sameAs": [
          "https://www.linkedin.com/company/gtmstacker",
          "https://www.youtube.com/@gtmstacker",
          "https://www.instagram.com/gtmstacker/",
          "https://www.tiktok.com/@gtmstacker"
        ],
        "mainEntityOfPage": "https://gtmstacker.com/registry/about/"
      },
      {
        "@type": "SoftwareApplication",
        "@id": "https://gtmstacker.com/registry/tool/harness2/#software",
        "name": "Harness² (Google Research — recursive agent harnessing)",
        "identifier": "io.github.google-research/harness2",
        "description": "Apache-2.0, self-hostable framework from Google Research that improves the harness around a frozen model (its instructions, guardrails, memory, tool guidance, subagents, skills, scripts, plugins) via a Propose-Probe-Compose loop with no retraining, so a lean team can raise an agent's benchmark results without touching model weights or paying for fine-tuning.",
        "applicationCategory": "DeveloperApplication",
        "url": "https://gtmstacker.com/registry/tool/harness2/",
        "datePublished": "2026-10-18T00:00:00Z",
        "dateModified": "2026-10-18T00:00:00Z",
        "isPartOf": {
          "@id": "https://gtmstacker.com/#website"
        },
        "softwareHelp": "https://harness2.github.io",
        "license": "https://spdx.org/licenses/Apache-2.0.html",
        "codeRepository": "https://github.com/google-research/harness2",
        "keywords": "ai-infrastructure, mcp-agents, ai-agents, self-hostable, open-source",
        "author": {
          "@type": "Organization",
          "name": "google-research",
          "url": "https://github.com/google-research",
          "sameAs": [
            "https://github.com/google-research/harness2",
            "https://harness2.github.io"
          ]
        },
        "offers": {
          "@type": "Offer",
          "price": 0,
          "priceCurrency": "USD"
        }
      },
      {
        "@type": "BreadcrumbList",
        "@id": "https://gtmstacker.com/registry/tool/harness2/#breadcrumb",
        "itemListElement": [
          {
            "@type": "ListItem",
            "position": 1,
            "name": "GTM Stacker Registry",
            "item": "https://gtmstacker.com/registry/"
          },
          {
            "@type": "ListItem",
            "position": 2,
            "name": "AI Infrastructure",
            "item": "https://gtmstacker.com/registry/category/ai-infrastructure/"
          },
          {
            "@type": "ListItem",
            "position": 3,
            "name": "Harness² (Google Research — recursive agent harnessing)",
            "item": "https://gtmstacker.com/registry/tool/harness2/"
          }
        ]
      }
    ]
  },
  "tool": {
    "name": "io.github.google-research/harness2",
    "repository": {
      "url": "https://github.com/google-research/harness2",
      "source": "github"
    },
    "homepage": "https://harness2.github.io"
  }
}