{ "data": { "a": { "slug": "axlr", "name": "AXLR", "vendor": "Underpass AI", "vendorUrl": "https://axlr.underpassai.com", "kind": "harness", "category": "agent-harnesses", "summary": "Open-source Go execution engine for agentic work from Underpass AI. A console with work modes and approval cards, an mTLS HTTP, gRPC and MCP service, a JSON worker and a Go library, with KMP memory and MADE procedures over MCP.", "url": "https://www.anchorterminal.com/tools/axlr", "markdownUrl": "https://www.anchorterminal.com/tools/axlr.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/axlr.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/axlr.json", "dns": "axlr.grade.anchorterminal.com", "repo": "https://github.com/underpass-ai/AXLR", "license": "Apache-2.0, with a NOTICE naming Tirso García Ibáñez as the copyright holder", "transports": [], "packages": [ { "registry": "go", "name": "github.com/underpass-ai/AXLR" }, { "registry": "oci", "name": "ghcr.io/underpass-ai/axlr" }, { "registry": "oci", "name": "ghcr.io/underpass-ai/charts/axlr" } ], "auth": "mixed", "authNotes": "The console and the one-command `axlr-tui --exec` need a model route, an OpenRouter key (`OPENROUTER_API_KEY` or `credentials.json`, mode 0600), a local OpenAI-compatible server, or an Anthropic Messages or OpenAI Responses key in a `local_models` entry; the JSON worker and the library tests need no model account. The service, `axlr-serve`, requires TLS 1.3 and a client certificate signed by its CA whose SHA-256 fingerprint is mapped in a private principals file to a principal and the roles `session_client`, `approver`, `tool_operator` and `admin`; an unmapped certificate is forbidden, a session belongs to the principal that created it, and authentication never approves a tool by itself. MCP at `/mcp` is stateless and authenticated per request by the same certificate; gRPC uses the TLS peer certificate.", "pricing": "free", "pricingNotes": "Free and open source under Apache-2.0, with nothing to buy from Underpass AI. You pay your model provider (OpenRouter reports a cost per request, which the console's `/usage` and `max_session_usd` track) or run a local server at no charge; TypeSafe's Jev judge is optional and needs its own key. No hosted AXLR exists (checked 2026-11-01).", "priceSummary": "Free · OSS", "where": "local", "x402": { "level": "no", "evidence": "No x402, MPP or L402 in the repository, the OpenAPI file or the two sites (checked 2026-11-01). Nothing is sold.", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": 0, "npmWeekly": null, "pypiWeekly": null, "asOf": "2026-11-01" }, "docsUrl": "https://axlr.underpassai.com/docs/", "openapi": "https://github.com/underpass-ai/AXLR/blob/main/api/openapi/axlr-v1.yaml", "capabilities": [ "agent.harness", "agent.mcp-client", "agent.multi-agent" ], "tags": [ "open-source", "local", "self-hosted", "free", "no-card", "go", "cli", "mcp-client", "openapi", "grpc", "docker", "helm", "sandbox", "approvals", "pre-1.0", "new" ], "lastRelease": "2026-10-25", "graded": true, "anchor": { "graded": true, "score": 70.6, "grade": "BB", "agentReady": true, "rank": 168, "ranked": true, "rankOf": 1253, "categoryRank": 3, "methodology": "0.7", "run": "2026-10-01", "scores": { "ergonomics": 83, "maintenance": 79, "payments": 60, "reliability": 70, "schema": 80, "security": 62, "transparency": 55 }, "pending": [ "performance", "tasks" ], "assessment": { "confidence": "medium", "date": "2026-11-01" }, "negative": 0, "verdict": "A pre-1.0 Go harness with an unusually complete service contract, mutual TLS with principal roles, required idempotency keys, a policy, budget and sandbox per execution, and a result object a program can read. The console's defaults after setup are autonomy on and no sandbox, the project has one maintainer and no outside users yet, and releases carry no notes.", "bestFor": "A program or bot that needs one governed execution at a time, with a policy, a budget, a sandbox profile and correlation ids it sets itself and a result object it reads back, on a machine it controls, with the model on OpenRouter, a local server or a native Anthropic or OpenAI key.", "strengths": [ "mTLS HTTP, gRPC and MCP service with four principal roles, `Idempotency-Key` required on turns, approvals and direct calls, and `expected_revision` on decisions", "A policy (observe, propose, act-with-approval, act-within-policy), a budget in seconds, requests, tool calls and dollars, and a bubblewrap or container sandbox with an enforced egress allow list, each per execution", "`axlr-tui --exec --json` runs one task headless and ends with a result object; exit 0 done, 3 waiting for a decision, 1 error", "OpenRouter, local OpenAI-compatible servers (llama.cpp, vLLM, Ollama, LM Studio), and the Anthropic Messages and OpenAI Responses APIs spoken directly", "21 releases in October 2026 from a workflow that tests on six native runners and refuses to republish different bytes, with a `.sha256` beside every archive" ], "weaknesses": [ "The setup assistant turns autonomy on and leaves the exec sandbox off, so after setup the model's commands run with the person's rights, bar the destructive list", "Pre-1.0 with one maintainer, 0 stars and 10 open issues, all the founder's own trackers; releases have empty notes and there is no changelog", "The service needs KMP and MADE as remote gRPC services, a client CA and a principals file before a first session, and has no work modes, plugins or forged-tool setting yet", "No SECURITY.md, security.txt, advisories or disclosure route; the legal entity is not yet incorporated and no terms or privacy policy exist", "No llms.txt, and the Go library is the only SDK; the catalog registry answers every non-AXLR client with 403" ], "agentNotes": [ "Before the first prompt, set `approvals.autonomous` to false or run with `--approve none`, and turn `exec_sandbox.mode` to `auto` or `required` on Linux; after setup both default the other way", "Drive a run with `axlr-tui --exec --json --policy \u003cfile\u003e --budget \u003cfile\u003e --sandbox \u003cfile\u003e` and read only the last `result` line; exit 3 means a call waits for `--decide approve|deny` with `--session \u003cid\u003e`", "On the service, send `Idempotency-Key` (16 to 128 safe ASCII characters) on every turn, approval and direct call, and pass the session's `expected_revision`; a reused key with a different body answers 409", "A local model must return native tool calls (llama.cpp with `--jinja`, vLLM with `--enable-auto-tool-choice`); set `context_tokens` on the entry and `OLLAMA_CONTEXT_LENGTH` for Ollama", "Leave `TYPESAFE_API_KEY` unset unless you want the setup assistant to switch Jev on, which sends requests and final answers to TypeSafe" ], "metrics": { "kind": "local", "measured": false }, "reviewCount": 0, "avgRating": 0, "history": [ { "basis": "public evidence", "confidence": "medium", "grade": "BB", "methodology": "0.7", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 70.6 } ], "editorialScores": { "ergonomics": 83, "maintenance": 79, "payments": 60, "reliability": 70, "schema": 80, "security": 62, "transparency": 68 }, "provenanceScore": 42, "checked": { "categories": { "ergonomics": "2026-11-01", "maintenance": "2026-11-01", "payments": "2026-11-01", "reliability": "2026-11-01", "schema": "2026-11-01", "security": "2026-11-01", "transparency": "2026-11-01" }, "last": "2026-11-01", "oldest": "2026-11-01", "stale": false, "staleAfterDays": 90 } }, "connect": { "install": "V=v0.15.0; OS=linux; ARCH=amd64; curl -fsSLO https://github.com/underpass-ai/AXLR/releases/download/$V/axlr-$V-$OS-$ARCH.tar.gz{,.sha256} \u0026\u0026 sha256sum -c axlr-$V-$OS-$ARCH.tar.gz.sha256 \u0026\u0026 tar xzf axlr-$V-$OS-$ARCH.tar.gz -C ~/.local/bin axlr axlr-tui axlr-serve # or build from source with Go 1.26", "headless": { "command": "axlr-tui --root \"$PWD\" --exec \"$TASK\" --json --approve none --policy policy.json --budget budget.json --sandbox sandbox.json", "decide": "axlr-tui --session \u003cid\u003e --decide approve --json", "env": { "OPENROUTER_API_KEY": "\u003ckey\u003e" } } }, "letme": { "capability": "https://letme.dev/agent.harness", "tool": "https://letme.dev/axlr" }, "area": "frameworks", "provenance": { "legalEntity": "", "domain": "underpassai.com", "domainRegistered": "2025-08-10", "endpointOnVendorDomain": null, "terms": "", "privacy": "", "statusPage": "", "changelog": "https://github.com/underpass-ai/AXLR/releases", "securityTxt": "none", "checked": "2026-11-01", "notes": [ "No legal entity exists yet. underpassai.com's footer says Underpass AI is a startup project created by Tirso García Ibáñez and preparing incorporation as a Spanish limited liability company, that the technical assets are founder-owned and intended to be assigned to the company after incorporation, and that licence terms are in each repository's LICENSE and NOTICE. The NOTICE names Tirso García Ibáñez as copyright holder. Recorded as absent with that explanation.", "No terms of service or privacy policy on underpassai.com (sitemap lists the home page and one article) or axlr.underpassai.com. For local software the docs' 'Where data goes' and 'Diagnostics and privacy' sections stand in.", "RDAP for underpassai.com gives registration on 2025-08-10 with Route 53 nameservers. The GitHub organisation underpass-ai was created on 2025-08-09, location Spain, 16 public repositories. The AXLR repository was created on 2026-09-29; the first tag and release, v0.1.0, is dated 2026-10-04, which is firstReleased.", "https://underpassai.com/.well-known/security.txt and https://axlr.underpassai.com/.well-known/security.txt answer 404; the repository has no SECURITY.md. No status page (local software). The changelog is the GitHub releases list, whose entries carry no notes.", "robots.txt on underpassai.com and axlr.underpassai.com allow everything. registry.underpassai.com answers every path, robots.txt included, with 403 'forbidden: client is not AXLR', so nothing was read from it after that reply.", "The founder's LinkedIn URL is linked from underpassai.com and was not visited." ], "score": 42 }, "pageJsonUrl": "https://www.anchorterminal.com/tools/axlr.json" }, "answer": "AXLR scores 70.6 (BB) on agent readiness against Harness²'s 42 (E), and leads in every scored category.", "b": { "slug": "harness2", "name": "Harness²", "vendor": "Google (Google Research)", "vendorUrl": "https://harness2.github.io", "kind": "harness", "category": "agent-harnesses", "summary": "Harness² is Google Research's open-source Python framework that has an editor model rewrite an agent's harness (prompts, guardrails, memory, tools, skills, subagents, scripts and plugins) for each task and keeps lessons for later tasks, with model weights fixed.", "url": "https://www.anchorterminal.com/tools/harness2", "markdownUrl": "https://www.anchorterminal.com/tools/harness2.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/harness2.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/harness2.json", "dns": "harness2.grade.anchorterminal.com", "repo": "https://github.com/google-research/harness2", "license": "Apache-2.0. Bundled benchmark runners keep their upstream licences (Apache-2.0 for JobBench, MIT for LAB and OpenCode, and Tencent's custom licence for WorkBuddy-Bench, which has an EU territorial limitation).", "transports": [], "packages": [], "auth": "none", "authNotes": "No Harness² account or key. Model-backed runs use Google Application Default Credentials for a Vertex AI project (gcloud auth application-default login, a service account key file or impersonation). The optional Codex bridge generates its own local access key.", "pricing": "free", "pricingNotes": "Free under Apache-2.0, with nothing to buy. Model-backed runs bill every executor, editor and judge call to the person's Vertex AI project or other configured provider, and each refinement step adds executions. The local demo needs no credentials.", "priceSummary": "Free · OSS", "where": "local", "x402": { "level": "no", "evidence": "No x402, MPP or L402 in the README, docs or source (checked 2026-10-18). The software has nothing to buy.", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": 4200, "npmWeekly": null, "pypiWeekly": null, "asOf": "2026-10-18" }, "docsUrl": "https://github.com/google-research/harness2/tree/main/docs", "capabilities": [ "agent.harness" ], "tags": [ "open-source", "apache-2.0", "local", "self-hosted", "free", "python", "cli", "linux" ], "lastRelease": "2026-10-08", "graded": true, "anchor": { "graded": true, "score": 42, "grade": "E", "agentReady": false, "rank": 1111, "ranked": true, "rankOf": 1253, "categoryRank": 43, "methodology": "0.7", "run": "2026-10-01", "scores": { "ergonomics": 59, "maintenance": 36, "payments": 40, "reliability": 20, "schema": 51, "security": 44, "transparency": 49 }, "pending": [ "performance", "tasks" ], "assessment": { "confidence": "medium", "date": "2026-10-18" }, "negative": 0, "verdict": "Refinement needs no grader or answer key, and the code keeps judges out of selection. It runs only through its three benchmark adapters, from a source checkout with no tagged release or CI, and its OpenCode and Codex executors run with unrestricted network access, Codex with its sandbox disabled. The benchmark gains are the authors' claims.", "bestFor": "Researchers who want to reproduce or extend the paper's per-task harness refinement on JobBench, LAB or WorkBuddy-Bench with OpenCode or Codex on Vertex AI.", "strengths": [ "Refining a harness needs no grader, rubric or answer key. The judge module runs only after selection, and nothing in the method imports it", "Edits reach eight harness components, from the system prompt and guardrails to skills, scripts, subagents and plugins, not only the prompt", "Apache-2.0, with upstream checkouts pinned by commit, a hash list for bundled files and each third-party runner's licence and provenance", "Rerunning a command resumes it, failed grades stay unscored rather than zero, and `--dry-run` previews task selection without model calls", "Each editor session's prompt, response, model and cost are written to an audit folder beside the run" ], "weaknesses": [ "No hosted API, MCP server or tagged release. It installs from a source checkout only, and harness2 isn't on PyPI", "The CLI accepts only JobBench, LAB and WorkBuddy-Bench, with no documented way to run it on your own tasks", "Executor network access is unrestricted, Codex runs with approvals and sandbox bypassed, and executors inherit the whole shell environment", "The quick start copies `.env.example`, which the repository doesn't contain, and no .gitignore keeps a filled-in .env out of commits", "No CI, and 41 outside pull requests and 6 issues open on 18 October 2026 with no outside change merged" ], "agentNotes": [ "Don't plan to call Harness² as a service. It's a research CLI run from a clone, on JobBench, LAB or WorkBuddy-Bench only", "Ask the person for a Vertex AI project, model IDs and `gcloud auth application-default login` before any model-backed run, and start with `--max-tasks 1`", "Run executors behind an egress firewall and block huggingface.co and github.com after setup, as the docs advise, so rubrics can't be fetched", "Write .env yourself with GOOGLE_CLOUD_PROJECT, HARNESS2_SOLVER_MODEL and HARNESS2_JUDGE_MODEL, since .env.example isn't shipped, and never commit it", "Use the same benchmark, executor, `--k`, `--mode` and seed for `harness2`, `harness2-judge` and `harness2-report`, or the results won't line up" ], "metrics": { "kind": "local", "measured": false }, "reviewCount": 4, "avgRating": 2.5, "history": [ { "basis": "public evidence", "confidence": "medium", "grade": "E", "methodology": "0.7", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 42 } ], "editorialScores": { "ergonomics": 59, "maintenance": 36, "payments": 40, "reliability": 20, "schema": 51, "security": 44, "transparency": 48 }, "provenanceScore": 50, "checked": { "categories": { "ergonomics": "2026-10-18", "maintenance": "2026-10-18", "payments": "2026-10-18", "reliability": "2026-10-18", "schema": "2026-10-18", "security": "2026-10-18", "transparency": "2026-10-18" }, "last": "2026-10-18", "lastMoved": "2026-10-27", "oldest": "2026-10-18", "stale": false, "staleAfterDays": 90 } }, "connect": { "install": "git clone https://github.com/google-research/harness2.git \u0026\u0026 cd harness2 \u0026\u0026 uv venv --python 3.12 .venv \u0026\u0026 source .venv/bin/activate \u0026\u0026 uv pip install -e '.[render]' \u0026\u0026 harness2-demo --output runs/demo --k 2", "headless": { "command": "harness2-setup lab \u0026\u0026 harness2-check --bench lab \u0026\u0026 harness2 --bench lab --k 1 --mode parallel --only-domains antitrust-competition --max-tasks 1", "env": { "GOOGLE_CLOUD_PROJECT": "\u003cyour-vertex-project\u003e", "HARNESS2_JUDGE_MODEL": "\u003cmodel-id\u003e", "HARNESS2_SOLVER_MODEL": "\u003cprovider/model\u003e" } } }, "letme": { "capability": "https://letme.dev/agent.harness", "tool": "https://letme.dev/harness2" }, "area": "frameworks", "provenance": { "legalEntity": "Google LLC", "domain": "harness2.github.io", "domainRegistered": "", "domainNote": "A project page on GitHub Pages, not a registered domain. The code is in the google-research organisation on GitHub.", "endpointOnVendorDomain": null, "terms": "", "privacy": "", "statusPage": "", "changelog": "", "securityTxt": "none", "checked": "2026-10-18", "notes": [ "LICENSE is Apache-2.0, and pyproject.toml and every source file name Google LLC. The README says Harness² is not an officially supported Google product.", "No terms or privacy policy govern the package or the project page. Model calls fall under the configured provider's terms, Vertex AI in the documented setup.", "The repository has no SECURITY.md, and harness2.github.io/.well-known/security.txt returns 404. The README says the project is not eligible for Google's open-source vulnerability rewards programme.", "No changelog, tags or GitHub releases. The version is 0.1.0 in pyproject.toml.", "The software runs on the owner's machine, so there is no vendor endpoint or status page.", "The project page has no robots.txt and carries a noindex, nofollow robots meta tag. It was read once as a public page." ], "score": 50 }, "pageJsonUrl": "https://www.anchorterminal.com/tools/harness2.json", "live": { "slug": "harness2", "githubStars": 21839, "securityTxt": { "url": "https://harness2.github.io/.well-known/security.txt", "state": "none", "checkedAt": "2026-11-01T15:42:20.137673501Z" }, "domain": { "domain": "github.io", "checkedAt": "2026-11-01T14:18:40.30630652Z" }, "github": { "repo": "google-research/harness2", "checkedAt": "2026-10-26T07:31:08.941074475Z", "stars": 19838, "archived": false, "issuesEnabled": true, "pushedAt": "2026-10-14T02:22:51Z", "openIssues": 28, "openPRs": 104, "issuesOpened90d": 29, "issuesClosed90d": 1, "prsMerged90d": 0, "replies": { "sampled": 25, "unanswered": 24, "medianDays": 0 } }, "updatedAt": "2026-11-01T17:04:25.142667659Z" } }, "edges": [ { "by": 50, "category": "reliability", "leads": "axlr", "name": "Reliability", "reasons": { "a": "Read as a local package, as claude-code, goose, opencode and openhands were. Release archives for Linux, macOS and Windows on amd64 and arm64 with a `.sha256` beside each, a GHCR image, an OCI Helm chart and source builds with Go 1.26, all stated in the docs; no Homebrew, npm or apt route (20). CI on main runs vet, race tests and a coverage gate above 80 per cent for both modules, with the sandbox tests required under bubblewrap, and the last ten runs on main to 1 November 2026 all passed; the release workflow tests on six native runners (25). 10 open issues, all opened by the founder on 19 and 21 October 2026 as trackers for service work (work modes, plugins, the sandbox setting, cost limits), none from an outside user and none a crash report, 0 open pull requests and about 100 merged; with 0 stars there is nobody else to report, so the handling of outside reports is untested (20). Tags are semver 0.x and the docs date each behaviour change inline (\"until 14 October 2026…\"), but the 21 releases have empty notes, there is no CHANGELOG.md, and the product contract records one retirement (the seven 1.0 skill definitions on 5 October 2026) without a notice period (5). Pre-1.0, which the README calls an evolving project (0). 20+25+20+5+0 = 70.", "b": "Local-package reading. Not on PyPI (the release feed for harness2 answers 404). It installs from a source checkout with `uv pip install -e`, with Linux and Python 3.12 stated (the package allows 3.13), plus Node.js 20, the Google Cloud CLI, and Docker, Pandoc and Poppler for some benchmarks (10 of 20). No CI workflow in the repository. tests/test_workflow.py holds five unit tests over the deterministic demo, which needs no credentials (5 of 25). Six issues and 42 pull requests were open on 18 October, the pull requests mostly fixes by their titles, such as the missing .env.example, report failures, seeds lost between refinement and grading, and resume. None is merged and no issue is closed. Issue text and replies couldn't be read (5 of 25). No tags or changelog (0). Version 0.1.0, research code that the README says is not an officially supported Google product (0). 20 in all." }, "scores": { "a": 70, "b": 20 } }, { "by": 43, "category": "maintenance", "leads": "axlr", "name": "Maintenance \u0026 community", "reasons": { "a": "v0.15.0 on 2026-10-25, with commits on main to 2026-11-01 (30). 21 releases between 4 and 25 October 2026 (20). One maintainer (368 of 399 commits, the rest by two agent accounts), 10 open issues that are his own trackers with no comments, about 100 merged pull requests and no outside report to measure a reply against (15). Not in the MCP registry; the Go module and a proto to generate clients from are the only client code, beside the GHCR image and Helm chart (5). Go 1.26 toolchain pinned, release actions pinned by commit SHA, a go-licenses check that the third-party bundle matches the dependency graph, and CI passing on main (9). 30+20+15+5+9 = 79.", "b": "0.1.0, first committed on 8 October 2026, with a README fix on 13 October (30). No tags, releases or dated changelog entries (0). 42 pull requests (41 from outside contributors, one from Dependabot) and 6 issues open on 18 October, against one pull request and no issues closed. No outside change is on main, which has three commits by the authors, and reply times couldn't be read (3 of 25). Not on PyPI and no SDK (0). No CI. Dependencies are pinned by version and upstream checkouts by commit, a ruff configuration is in pyproject.toml, and Dependabot's litellm bump of 8 October is still open (3 of 10). 36 in all." }, "scores": { "a": 79, "b": 36 } }, { "by": 29, "category": "schema", "leads": "axlr", "name": "Schema \u0026 documentation", "reasons": { "a": "An OpenAPI 3.1 file for the service (15 paths, 23 schemas, `mutualTLS` security scheme, `additionalProperties: false` on request bodies), a checked proto for the gRPC service, and the 19 MCP tools built from the same shared schemas (25). No llms.txt on either site (404), and the docs site renders HTML; the Markdown sources are in the repository (5). The product contract tables what each of the four interfaces can and cannot do, the work-mode table states each mode's purpose and tool policy, the transport parity ledger names every verb's arguments and behaviour, and `axlr_tools` tells the model to look for a tool before forging one (15). Typed inputs throughout, `Idempotency-Key` 16 to 128 safe ASCII characters, `expected_revision` a positive integer, policy levels, sandbox profiles and egress values enumerated, budgets as four integers, and the service rejects unknown fields, invalid types and bodies over 4 MiB (14). Examples in every guide, an error shape of `code`, `message` and `request_id`, named codes (`409 idempotency_conflict`, `400 invalid_cursor`, `422 invalid_policy`, `invalid_budget`, `sandbox_unavailable`, `not_ready`, `invalid_id`), a gRPC status mapping, and the worker's five statuses and three exit codes; no single error reference page (14). `/v1` routes and an OpenAPI version of 0.1.0 against a 0.15.0 release, and no changelog beyond the dated notes inside the docs (7). 25+5+15+14+14+7 = 80.", "b": "Framework reading. Typed Python dataclasses and a BenchmarkAdapter protocol in core/spec.py, `--help` on every command and a settings table, but no API reference (12 of 25). Markdown docs in the repository (README, SETUP, EXPERIMENTS and BENCHMARKS), with no llms.txt (5 of 10). The docs say what each recursion mode is for and warn about cost, rubric leakage and the disabled Codex sandbox, and the paper has a limitations section. Nothing says how to use it outside the three benchmarks (13 of 20). `--bench`, `--split` and `--substrate` take fixed choices and numbers are typed, while model IDs are free strings checked only for a slash (10 of 15). Every command has a worked example and ConfigError messages name the fix, but the quick start's `cp .env.example .env` fails because the file isn't shipped (8 of 15). Version 0.1.0 with no changelog (3 of 15). 51 in all." }, "scores": { "a": 80, "b": 51 } }, { "by": 24, "category": "ergonomics", "leads": "axlr", "name": "Agent ergonomics", "reasons": { "a": "Harness reading of the framework line, scored on what a program driving it has to supply. Six local tools plus the host's bookkeeping tools; work modes and an execution policy hide the tools a run may not use, `axlr_catalog` searches the catalog instead of loading it, forged tools keep the request prefix for the prompt cache, and a context cap of 75,000 prompt tokens opens a successor session (20). `local_read`, `local_search` and `local_list` page with `offset`, `next_offset` and `max_bytes`; `StreamEvents` takes `after`, `max_events` (1 to 256) and `wait_ms`; a budget per execution in seconds, requests, tool calls and dollars; 32 tool calls a turn and `max_session_usd` (20). The `--exec` run ends in one `result` line with a status, the pending call, every call's outcome and the usage, and exits 0, 3 or 1; service errors carry codes an agent can act on; a budget stop says which limit (`budget: requests 20 reached`) and leaves the session resumable (16). `Idempotency-Key` on every turn, approval and direct call with cross-transport replay, `expected_revision` on decisions, `expected_sha256` on worker writes, and an interrupted effectful call left `uncertain` for an operator rather than retried (19). One command runs a task, but the first launch needs an OpenRouter key or a local server, the service needs a CA, a principals file and remote KMP and MADE before a first session, and the only SDK is the Go library (8). 20+20+16+19+8 = 83.", "b": "Framework reading, adapted to a harness that a pipeline drives. Results land in run folders, not in the caller's context, and an agent reads task summaries, adopted harnesses and reflection notes from runs/ (12 of 25). `--only-domains`, `--max-tasks` (per domain), `--split-file`, `--workers` and `--dry-run` size a run before any model call (14 of 20). harness2-check tests dependencies, credentials and configuration without a model call, configuration errors name the remedy, and incomplete runs exit nonzero (14 of 20). Rerunning a command resumes it, saved configuration refuses an incompatible resume, failed grades stay unscored, and editor replies are cached by a fingerprint of their inputs (15 of 20). No model has a default, a Vertex AI project and a gcloud sign-in are needed, only three benchmarks are accepted and there's no SDK (4 of 15). 59 in all." }, "scores": { "a": 83, "b": 59 } }, { "by": 20, "category": "payments", "leads": "axlr", "name": "Payments \u0026 pricing", "reasons": { "a": "No payment protocol (0). Free and Apache-2.0 with nothing to buy, so 20, 20 and 20 on the last three lines by the self-hosted rule, since the docs say what you pay your model provider, the free route needs no card, and it runs local models with no signup at all, the JSON worker without any model. 0+20+20+20 = 60.", "b": "No payment protocol (0). Read with the self-hosted rule, as for the other harnesses. Free under Apache-2.0 with nothing to buy (20). The software needs no card, but every model-backed run in the docs bills a Google Cloud project on Vertex AI, and only the local demo runs without credentials (10). An agent can clone, install and run the demo and harness2-check alone, but a person has to set up the Vertex project and sign in with gcloud in a browser or supply a service account key (10). 40 in all." }, "scores": { "a": 60, "b": 40 } }, { "by": 18, "category": "security", "leads": "axlr", "name": "Security \u0026 auth", "reasons": { "a": "Harness reading of the framework checklist, as for the other four harnesses. 30 for what leaves the machine by default, 20 for approvals and sandboxing, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. No product telemetry in the console source (the outbound hosts are openrouter.ai, api.anthropic.com, api.openai.com, a configured local server, api.github.com and github.com for engine downloads, and registry.underpassai.com for the signed catalog index) or in the docs; payload capture is off by default since 9 October 2026, traces omit prompts and credentials, model keys are never given to commands or plugins, and Jev sends requests and answers to TypeSafe only when on, which the setup assistant does when `TYPESAFE_API_KEY` is set, with a notice on stderr (27). Approval cards, a destructive list that always asks, four work modes, a policy level per execution, a bubblewrap or container sandbox with an enforced egress allow list, and a service that approves nothing but reads. Against that, the setup assistant writes `autonomous: true`, `exec_sandbox` is off by default, forged tools run with the person's rights, and the README says the console and worker are not sandboxes (13). Forged tools are refused when a file changes since the forge, a policy forbids calls outside the model, and the docs say MCP tools keep their own side effects, but no prompt-injection guidance was found and MCP results reach the model unmarked (6). The service keeps an event journal, audit records, a receipt with the SHA-256 of every call's arguments and outcome, and correlation ids on every event; the console keeps session records, a redacted app log and `/session` (14). No SECURITY.md, no security.txt on either host, no advisories, no bug bounty; release checksums and a minisign-signed catalog are integrity, not a programme (2). 27+13+6+14+2 = 62.", "b": "Harness reading of the framework checklist, as for the other harnesses. 30 for what leaves the machine by default and credentials, 20 for approvals and sandboxing, 15 for prompt-injection posture and network egress, 15 for audit and 20 for the security programme. No telemetry found in the Harness² source, and the WorkBuddy OpenCode configs turn OpenCode's telemetry, auto-update and sharing off. Credentials are Google Application Default Credentials, and the Codex bridge binds 127.0.0.1 by default with a generated key stored with mode 0600. Against that, executors inherit the whole shell environment (the editor channel drops only five Anthropic variables), and no .gitignore keeps a filled-in .env out of commits (20). Codex runs with `--dangerously-bypass-approvals-and-sandbox`, the Claude CLI editor channel with `--dangerously-skip-permissions`, and the WorkBuddy OpenCode preset allows every tool, with only OpenCode's directory fence and WorkBuddy's containers as limits. The README and setup guide say so (4). Executor network access is unrestricted by default. The docs name the hosts to block and report a Codex rollout that fetched its rubric unprompted, but give no guidance on instructions inside task documents (5). Each editor call's prompt, system text, response, model and cost are saved to an audit folder, with trajectories and adopted harnesses under runs/ (13). No SECURITY.md or security.txt and no advisories, and the README says the project is outside Google's open-source vulnerability rewards programme (2). 44 in all." }, "scores": { "a": 62, "b": 44 } } ], "facts": [ { "a": "Agent harness", "b": "Agent harness", "name": "Kind" }, { "a": "Underpass AI", "b": "Google (Google Research)", "name": "Vendor" }, { "a": "no (local only)", "b": "no (local only)", "name": "Hosted endpoint" }, { "a": "", "b": "", "name": "Transports" }, { "a": "OAuth or key", "b": "None", "name": "Auth" }, { "a": "Free", "b": "Free", "name": "Pricing" }, { "a": "no", "b": "no", "name": "x402" }, { "a": "Apache-2.0, with a NOTICE naming Tirso García Ibáñez as the copyright holder", "b": "Apache-2.0. Bundled benchmark runners keep their upstream licences (Apache-2.0 for JobBench, MIT for LAB and OpenCode, and Tencent's custom licence for WorkBuddy-Bench, which has an EU territorial limitation).", "name": "Licence" }, { "a": "no", "b": "no", "name": "Read-only variant documented" }, { "a": "no", "b": "no", "name": "llms.txt" }, { "a": "2026-10-25", "b": "2026-10-08", "name": "Last release" }, { "a": "no document linked", "b": "no document linked", "name": "Terms last updated" }, { "a": "no document linked", "b": "no document linked", "name": "Privacy policy last updated" }, { "a": "0 stars", "b": "4.2k stars", "name": "Popularity" }, { "a": "none", "b": "2.5/5 (4)", "name": "Agent reviews" } ], "faq": [ { "answer": "AXLR scores 70.6 (BB) on agent readiness against Harness²'s 42 (E), and leads in every scored category.", "question": "Which is better for AI agents, AXLR or Harness²?" }, { "answer": "Yes. AXLR is open source (Apache-2.0, with a NOTICE naming Tirso García Ibáñez as the copyright holder). Harness² is open source (Apache-2.0. Bundled benchmark runners keep their upstream licences (Apache-2.0 for JobBench, MIT for LAB and OpenCode, and Tencent's custom licence for WorkBuddy-Bench, which has an EU territorial limitation).).", "question": "Are AXLR and Harness² open source?" } ], "goodFor": [ { "aheadOn": [ "Reliability, 70 against 20", "Schema \u0026 documentation, 80 against 51", "Agent ergonomics, 83 against 59", "Security \u0026 auth, 62 against 44", "Payments \u0026 pricing, 60 against 40", "Maintenance \u0026 community, 79 against 36", "Transparency \u0026 trust, 55 against 49" ], "also": [ "Agent-ready, a grade of BB or better", "Free to start without a card" ], "goodFor": "A program or bot that needs one governed execution at a time, with a policy, a budget, a sandbox profile and correlation ids it sets itself and a result object it reads back, on a machine it controls, with the model on OpenRouter, a local server or a native Anthropic or OpenAI key.", "slug": "axlr", "watchFor": "The setup assistant turns autonomy on and leaves the exec sandbox off, so after setup the model's commands run with the person's rights, bar the destructive list" }, { "aheadOn": null, "also": [ "No key needed to call it" ], "goodFor": "Researchers who want to reproduce or extend the paper's per-task harness refinement on JobBench, LAB or WorkBuddy-Bench with OpenCode or Codex on Vertex AI.", "slug": "harness2", "watchFor": "No hosted API, MCP server or tagged release. It installs from a source checkout only, and harness2 isn't on PyPI" } ], "history": { "axlr": null, "harness2": [ { "causes": [ "method" ], "date": "2026-10-27", "delta": 0.4, "from": 41.6, "fromGrade": "E", "to": 42, "toGrade": "E" } ] }, "job": { "capability": "agent.harness", "name": "Agent harnesses" }, "others": [ { "json": "https://www.anchorterminal.com/compare/agent-zero-vs-axlr.json", "title": "Agent Zero vs AXLR", "url": "https://www.anchorterminal.com/compare/agent-zero-vs-axlr" }, { "json": "https://www.anchorterminal.com/compare/agent-zero-vs-harness2.json", "title": "Agent Zero vs Harness²", "url": "https://www.anchorterminal.com/compare/agent-zero-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/aider-vs-axlr.json", "title": "Aider vs AXLR", "url": "https://www.anchorterminal.com/compare/aider-vs-axlr" }, { "json": "https://www.anchorterminal.com/compare/aider-vs-harness2.json", "title": "Aider vs Harness²", "url": "https://www.anchorterminal.com/compare/aider-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/amp-vs-axlr.json", "title": "Amp vs AXLR", "url": "https://www.anchorterminal.com/compare/amp-vs-axlr" }, { "json": "https://www.anchorterminal.com/compare/amp-vs-harness2.json", "title": "Amp vs Harness²", "url": "https://www.anchorterminal.com/compare/amp-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/antigravity-cli-vs-axlr.json", "title": "Antigravity CLI vs AXLR", "url": "https://www.anchorterminal.com/compare/antigravity-cli-vs-axlr" }, { "json": "https://www.anchorterminal.com/compare/antigravity-cli-vs-harness2.json", "title": "Antigravity CLI vs Harness²", "url": "https://www.anchorterminal.com/compare/antigravity-cli-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/auggie-vs-axlr.json", "title": "Auggie CLI vs AXLR", "url": "https://www.anchorterminal.com/compare/auggie-vs-axlr" }, { "json": "https://www.anchorterminal.com/compare/auggie-vs-harness2.json", "title": "Auggie CLI vs Harness²", "url": "https://www.anchorterminal.com/compare/auggie-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-claude-code.json", "title": "AXLR vs Claude Code", "url": "https://www.anchorterminal.com/compare/axlr-vs-claude-code" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-cline.json", "title": "AXLR vs Cline", "url": "https://www.anchorterminal.com/compare/axlr-vs-cline" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-crush.json", "title": "AXLR vs Crush", "url": "https://www.anchorterminal.com/compare/axlr-vs-crush" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-cursor-cli.json", "title": "AXLR vs Cursor CLI", "url": "https://www.anchorterminal.com/compare/axlr-vs-cursor-cli" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-deepseek-harness.json", "title": "AXLR vs DeepSeek Harness (dsh)", "url": "https://www.anchorterminal.com/compare/axlr-vs-deepseek-harness" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-deepseek-superagent.json", "title": "AXLR vs dsh-superagent (DeepSeek)", "url": "https://www.anchorterminal.com/compare/axlr-vs-deepseek-superagent" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-devin.json", "title": "AXLR vs Devin", "url": "https://www.anchorterminal.com/compare/axlr-vs-devin" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-droid-cli.json", "title": "AXLR vs Droid CLI", "url": "https://www.anchorterminal.com/compare/axlr-vs-droid-cli" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-earendil-pi.json", "title": "AXLR vs Pi", "url": "https://www.anchorterminal.com/compare/axlr-vs-earendil-pi" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-gemini-cli.json", "title": "AXLR vs Gemini CLI", "url": "https://www.anchorterminal.com/compare/axlr-vs-gemini-cli" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-github-copilot-cli.json", "title": "AXLR vs GitHub Copilot CLI", "url": "https://www.anchorterminal.com/compare/axlr-vs-github-copilot-cli" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-goose.json", "title": "AXLR vs goose", "url": "https://www.anchorterminal.com/compare/axlr-vs-goose" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-harnessrouter.json", "title": "AXLR vs HarnessRouter", "url": "https://www.anchorterminal.com/compare/axlr-vs-harnessrouter" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-jules.json", "title": "AXLR vs Jules", "url": "https://www.anchorterminal.com/compare/axlr-vs-jules" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-junie.json", "title": "AXLR vs Junie CLI", "url": "https://www.anchorterminal.com/compare/axlr-vs-junie" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-kilo-code-cli.json", "title": "AXLR vs Kilo Code CLI", "url": "https://www.anchorterminal.com/compare/axlr-vs-kilo-code-cli" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-kimi-code-cli.json", "title": "AXLR vs Kimi Code CLI", "url": "https://www.anchorterminal.com/compare/axlr-vs-kimi-code-cli" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-kiro-cli.json", "title": "AXLR vs Kiro CLI", "url": "https://www.anchorterminal.com/compare/axlr-vs-kiro-cli" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-mini-swe-agent.json", "title": "AXLR vs mini-swe-agent", "url": "https://www.anchorterminal.com/compare/axlr-vs-mini-swe-agent" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-minimax-code.json", "title": "AXLR vs MiniMax Code", "url": "https://www.anchorterminal.com/compare/axlr-vs-minimax-code" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-mistral-vibe-cli.json", "title": "AXLR vs Mistral Vibe CLI", "url": "https://www.anchorterminal.com/compare/axlr-vs-mistral-vibe-cli" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-muse-code.json", "title": "AXLR vs Muse Code", "url": "https://www.anchorterminal.com/compare/axlr-vs-muse-code" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-openai-codex.json", "title": "AXLR vs OpenAI Codex", "url": "https://www.anchorterminal.com/compare/axlr-vs-openai-codex" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-opencode.json", "title": "AXLR vs OpenCode", "url": "https://www.anchorterminal.com/compare/axlr-vs-opencode" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-openhands.json", "title": "AXLR vs OpenHands", "url": "https://www.anchorterminal.com/compare/axlr-vs-openhands" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-openmausbot.json", "title": "AXLR vs OpenMausBot", "url": "https://www.anchorterminal.com/compare/axlr-vs-openmausbot" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-prime-agent.json", "title": "AXLR vs Prime Agent", "url": "https://www.anchorterminal.com/compare/axlr-vs-prime-agent" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-qoder-cli.json", "title": "AXLR vs Qoder CLI", "url": "https://www.anchorterminal.com/compare/axlr-vs-qoder-cli" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-qwen-code.json", "title": "AXLR vs Qwen Code", "url": "https://www.anchorterminal.com/compare/axlr-vs-qwen-code" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-reflection-mirror.json", "title": "AXLR vs Mirror CLI", "url": "https://www.anchorterminal.com/compare/axlr-vs-reflection-mirror" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-rovo-dev-cli.json", "title": "AXLR vs Rovo Dev CLI", "url": "https://www.anchorterminal.com/compare/axlr-vs-rovo-dev-cli" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-unstoppable.json", "title": "AXLR vs Unstoppable", "url": "https://www.anchorterminal.com/compare/axlr-vs-unstoppable" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-vibe-trading.json", "title": "AXLR vs Vibe-Trading", "url": "https://www.anchorterminal.com/compare/axlr-vs-vibe-trading" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-warp.json", "title": "AXLR vs Warp (Warp Agent CLI and the Oz Agent API)", "url": "https://www.anchorterminal.com/compare/axlr-vs-warp" }, { "json": "https://www.anchorterminal.com/compare/claude-code-vs-harness2.json", "title": "Claude Code vs Harness²", "url": "https://www.anchorterminal.com/compare/claude-code-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/cline-vs-harness2.json", "title": "Cline vs Harness²", "url": "https://www.anchorterminal.com/compare/cline-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/crush-vs-harness2.json", "title": "Crush vs Harness²", "url": "https://www.anchorterminal.com/compare/crush-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/cursor-cli-vs-harness2.json", "title": "Cursor CLI vs Harness²", "url": "https://www.anchorterminal.com/compare/cursor-cli-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/deepseek-harness-vs-harness2.json", "title": "DeepSeek Harness (dsh) vs Harness²", "url": "https://www.anchorterminal.com/compare/deepseek-harness-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/deepseek-superagent-vs-harness2.json", "title": "dsh-superagent (DeepSeek) vs Harness²", "url": "https://www.anchorterminal.com/compare/deepseek-superagent-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/devin-vs-harness2.json", "title": "Devin vs Harness²", "url": "https://www.anchorterminal.com/compare/devin-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/droid-cli-vs-harness2.json", "title": "Droid CLI vs Harness²", "url": "https://www.anchorterminal.com/compare/droid-cli-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/earendil-pi-vs-harness2.json", "title": "Pi vs Harness²", "url": "https://www.anchorterminal.com/compare/earendil-pi-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/gemini-cli-vs-harness2.json", "title": "Gemini CLI vs Harness²", "url": "https://www.anchorterminal.com/compare/gemini-cli-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/github-copilot-cli-vs-harness2.json", "title": "GitHub Copilot CLI vs Harness²", "url": "https://www.anchorterminal.com/compare/github-copilot-cli-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/goose-vs-harness2.json", "title": "goose vs Harness²", "url": "https://www.anchorterminal.com/compare/goose-vs-harness2" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-harnessrouter.json", "title": "Harness² vs HarnessRouter", "url": "https://www.anchorterminal.com/compare/harness2-vs-harnessrouter" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-jules.json", "title": "Harness² vs Jules", "url": "https://www.anchorterminal.com/compare/harness2-vs-jules" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-junie.json", "title": "Harness² vs Junie CLI", "url": "https://www.anchorterminal.com/compare/harness2-vs-junie" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-kilo-code-cli.json", "title": "Harness² vs Kilo Code CLI", "url": "https://www.anchorterminal.com/compare/harness2-vs-kilo-code-cli" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-kimi-code-cli.json", "title": "Harness² vs Kimi Code CLI", "url": "https://www.anchorterminal.com/compare/harness2-vs-kimi-code-cli" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-kiro-cli.json", "title": "Harness² vs Kiro CLI", "url": "https://www.anchorterminal.com/compare/harness2-vs-kiro-cli" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-mini-swe-agent.json", "title": "Harness² vs mini-swe-agent", "url": "https://www.anchorterminal.com/compare/harness2-vs-mini-swe-agent" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-minimax-code.json", "title": "Harness² vs MiniMax Code", "url": "https://www.anchorterminal.com/compare/harness2-vs-minimax-code" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-mistral-vibe-cli.json", "title": "Harness² vs Mistral Vibe CLI", "url": "https://www.anchorterminal.com/compare/harness2-vs-mistral-vibe-cli" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-muse-code.json", "title": "Harness² vs Muse Code", "url": "https://www.anchorterminal.com/compare/harness2-vs-muse-code" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-openai-codex.json", "title": "Harness² vs OpenAI Codex", "url": "https://www.anchorterminal.com/compare/harness2-vs-openai-codex" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-opencode.json", "title": "Harness² vs OpenCode", "url": "https://www.anchorterminal.com/compare/harness2-vs-opencode" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-openhands.json", "title": "Harness² vs OpenHands", "url": "https://www.anchorterminal.com/compare/harness2-vs-openhands" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-openmausbot.json", "title": "Harness² vs OpenMausBot", "url": "https://www.anchorterminal.com/compare/harness2-vs-openmausbot" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-prime-agent.json", "title": "Harness² vs Prime Agent", "url": "https://www.anchorterminal.com/compare/harness2-vs-prime-agent" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-qoder-cli.json", "title": "Harness² vs Qoder CLI", "url": "https://www.anchorterminal.com/compare/harness2-vs-qoder-cli" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-qwen-code.json", "title": "Harness² vs Qwen Code", "url": "https://www.anchorterminal.com/compare/harness2-vs-qwen-code" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-reflection-mirror.json", "title": "Harness² vs Mirror CLI", "url": "https://www.anchorterminal.com/compare/harness2-vs-reflection-mirror" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-rovo-dev-cli.json", "title": "Harness² vs Rovo Dev CLI", "url": "https://www.anchorterminal.com/compare/harness2-vs-rovo-dev-cli" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-unstoppable.json", "title": "Harness² vs Unstoppable", "url": "https://www.anchorterminal.com/compare/harness2-vs-unstoppable" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-vibe-trading.json", "title": "Harness² vs Vibe-Trading", "url": "https://www.anchorterminal.com/compare/harness2-vs-vibe-trading" }, { "json": "https://www.anchorterminal.com/compare/harness2-vs-warp.json", "title": "Harness² vs Warp (Warp Agent CLI and the Oz Agent API)", "url": "https://www.anchorterminal.com/compare/harness2-vs-warp" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-clawboo.json", "title": "AXLR vs Clawboo", "url": "https://www.anchorterminal.com/compare/axlr-vs-clawboo" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-cli-manager.json", "title": "AXLR vs CLI Manager", "url": "https://www.anchorterminal.com/compare/axlr-vs-cli-manager" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-harness-autonomous.json", "title": "AXLR vs Harness", "url": "https://www.anchorterminal.com/compare/axlr-vs-harness-autonomous" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-paperclip.json", "title": "AXLR vs Paperclip", "url": "https://www.anchorterminal.com/compare/axlr-vs-paperclip" }, { "json": "https://www.anchorterminal.com/compare/axlr-vs-parallax-agents.json", "title": "AXLR vs Parallax", "url": "https://www.anchorterminal.com/compare/axlr-vs-parallax-agents" } ], "panel": null, "scores": [ { "axlr": 70, "by": 50, "edge": "axlr", "harness2": 20, "key": "reliability", "name": "Reliability", "weight": 16 }, { "key": "performance", "name": "Performance", "pending": true, "weight": 10 }, { "axlr": 80, "by": 29, "edge": "axlr", "harness2": 51, "key": "schema", "name": "Schema \u0026 documentation", "weight": 13 }, { "axlr": 83, "by": 24, "edge": "axlr", "harness2": 59, "key": "ergonomics", "name": "Agent ergonomics", "weight": 13 }, { "axlr": 62, "by": 18, "edge": "axlr", "harness2": 44, "key": "security", "name": "Security \u0026 auth", "weight": 14 }, { "axlr": 60, "by": 20, "edge": "axlr", "harness2": 40, "key": "payments", "name": "Payments \u0026 pricing", "weight": 10 }, { "key": "tasks", "name": "Task success", "pending": true, "weight": 10 }, { "axlr": 79, "by": 43, "edge": "axlr", "harness2": 36, "key": "maintenance", "name": "Maintenance \u0026 community", "weight": 7 }, { "axlr": 55, "by": 6, "edge": "axlr", "harness2": 49, "key": "transparency", "name": "Transparency \u0026 trust", "weight": 7 } ], "summary": "AXLR scores 70.6 (BB) on agent readiness against Harness²'s 42 (E), and leads in every scored category. Both do agent harnesses.", "tasks": null, "tutorials": [], "verdicts": { "axlr": "A pre-1.0 Go harness with an unusually complete service contract, mutual TLS with principal roles, required idempotency keys, a policy, budget and sandbox per execution, and a result object a program can read. The console's defaults after setup are autonomy on and no sandbox, the project has one maintainer and no outside users yet, and releases carry no notes.", "harness2": "Refinement needs no grader or answer key, and the code keeps judges out of selection. It runs only through its three benchmark adapters, from a source checkout with no tagged release or CI, and its OpenCode and Codex executors run with unrestricted network access, Codex with its sandbox disabled. The benchmark gains are the authors' claims." } }, "kind": "anchor.page", "links": { "api": "https://www.anchorterminal.com/api/v1/index.json", "html": "https://www.anchorterminal.com/compare/axlr-vs-harness2", "json": "https://www.anchorterminal.com/compare/axlr-vs-harness2.json", "llms": "https://www.anchorterminal.com/llms.txt", "markdown": "https://www.anchorterminal.com/compare/axlr-vs-harness2.md", "mcp": "https://www.anchorterminal.com/mcp", "slim": "https://www.anchorterminal.com/compare/axlr-vs-harness2.min.md" }, "markdown": "AXLR scores 70.6 (BB) on agent readiness against Harness²'s 42 (E), and leads in every scored category. Both do agent harnesses.\n\n- AXLR: grade BB, 70.6/100, rank #168 of 1253. Markdown https://www.anchorterminal.com/tools/axlr.md · JSON https://www.anchorterminal.com/api/v1/tools/axlr.json\n- Harness²: grade E, 42/100, rank #1111 of 1253. Markdown https://www.anchorterminal.com/tools/harness2.md · JSON https://www.anchorterminal.com/api/v1/tools/harness2.json\n- Best agent harnesses and coding agents: https://www.anchorterminal.com/best/agent-harnesses/index.md\n- All 1000 harnesses comparisons: https://www.anchorterminal.com/compare/agent-harnesses/index.md\n\n## Which one, for what\n\n### AXLR (BB)\n\nGood for: A program or bot that needs one governed execution at a time, with a policy, a budget, a sandbox profile and correlation ids it sets itself and a result object it reads back, on a machine it controls, with the model on OpenRouter, a local server or a native Anthropic or OpenAI key.\n\nAhead on:\n- Reliability, 70 against 20\n- Schema \u0026 documentation, 80 against 51\n- Agent ergonomics, 83 against 59\n- Security \u0026 auth, 62 against 44\n- Payments \u0026 pricing, 60 against 40\n- Maintenance \u0026 community, 79 against 36\n- Transparency \u0026 trust, 55 against 49\n\nAlso in its favour:\n- Agent-ready, a grade of BB or better\n- Free to start without a card\n\nWatch for: The setup assistant turns autonomy on and leaves the exec sandbox off, so after setup the model's commands run with the person's rights, bar the destructive list\n\n### Harness² (E)\n\nGood for: Researchers who want to reproduce or extend the paper's per-task harness refinement on JobBench, LAB or WorkBuddy-Bench with OpenCode or Codex on Vertex AI.\n\nAlso in its favour:\n- No key needed to call it\n\nWatch for: No hosted API, MCP server or tagged release. It installs from a source checkout only, and harness2 isn't on PyPI\n\n\n## Why the edge\n\n### Reliability: AXLR +50\n\n- AXLR (70): Read as a local package, as claude-code, goose, opencode and openhands were. Release archives for Linux, macOS and Windows on amd64 and arm64 with a `.sha256` beside each, a GHCR image, an OCI Helm chart and source builds with Go 1.26, all stated in the docs; no Homebrew, npm or apt route (20). CI on main runs vet, race tests and a coverage gate above 80 per cent for both modules, with the sandbox tests required under bubblewrap, and the last ten runs on main to 1 November 2026 all passed; the release workflow tests on six native runners (25). 10 open issues, all opened by the founder on 19 and 21 October 2026 as trackers for service work (work modes, plugins, the sandbox setting, cost limits), none from an outside user and none a crash report, 0 open pull requests and about 100 merged; with 0 stars there is nobody else to report, so the handling of outside reports is untested (20). Tags are semver 0.x and the docs date each behaviour change inline (\"until 14 October 2026…\"), but the 21 releases have empty notes, there is no CHANGELOG.md, and the product contract records one retirement (the seven 1.0 skill definitions on 5 October 2026) without a notice period (5). Pre-1.0, which the README calls an evolving project (0). 20+25+20+5+0 = 70.\n- Harness² (20): Local-package reading. Not on PyPI (the release feed for harness2 answers 404). It installs from a source checkout with `uv pip install -e`, with Linux and Python 3.12 stated (the package allows 3.13), plus Node.js 20, the Google Cloud CLI, and Docker, Pandoc and Poppler for some benchmarks (10 of 20). No CI workflow in the repository. tests/test_workflow.py holds five unit tests over the deterministic demo, which needs no credentials (5 of 25). Six issues and 42 pull requests were open on 18 October, the pull requests mostly fixes by their titles, such as the missing .env.example, report failures, seeds lost between refinement and grading, and resume. None is merged and no issue is closed. Issue text and replies couldn't be read (5 of 25). No tags or changelog (0). Version 0.1.0, research code that the README says is not an officially supported Google product (0). 20 in all.\n\n### Maintenance \u0026 community: AXLR +43\n\n- AXLR (79): v0.15.0 on 2026-10-25, with commits on main to 2026-11-01 (30). 21 releases between 4 and 25 October 2026 (20). One maintainer (368 of 399 commits, the rest by two agent accounts), 10 open issues that are his own trackers with no comments, about 100 merged pull requests and no outside report to measure a reply against (15). Not in the MCP registry; the Go module and a proto to generate clients from are the only client code, beside the GHCR image and Helm chart (5). Go 1.26 toolchain pinned, release actions pinned by commit SHA, a go-licenses check that the third-party bundle matches the dependency graph, and CI passing on main (9). 30+20+15+5+9 = 79.\n- Harness² (36): 0.1.0, first committed on 8 October 2026, with a README fix on 13 October (30). No tags, releases or dated changelog entries (0). 42 pull requests (41 from outside contributors, one from Dependabot) and 6 issues open on 18 October, against one pull request and no issues closed. No outside change is on main, which has three commits by the authors, and reply times couldn't be read (3 of 25). Not on PyPI and no SDK (0). No CI. Dependencies are pinned by version and upstream checkouts by commit, a ruff configuration is in pyproject.toml, and Dependabot's litellm bump of 8 October is still open (3 of 10). 36 in all.\n\n### Schema \u0026 documentation: AXLR +29\n\n- AXLR (80): An OpenAPI 3.1 file for the service (15 paths, 23 schemas, `mutualTLS` security scheme, `additionalProperties: false` on request bodies), a checked proto for the gRPC service, and the 19 MCP tools built from the same shared schemas (25). No llms.txt on either site (404), and the docs site renders HTML; the Markdown sources are in the repository (5). The product contract tables what each of the four interfaces can and cannot do, the work-mode table states each mode's purpose and tool policy, the transport parity ledger names every verb's arguments and behaviour, and `axlr_tools` tells the model to look for a tool before forging one (15). Typed inputs throughout, `Idempotency-Key` 16 to 128 safe ASCII characters, `expected_revision` a positive integer, policy levels, sandbox profiles and egress values enumerated, budgets as four integers, and the service rejects unknown fields, invalid types and bodies over 4 MiB (14). Examples in every guide, an error shape of `code`, `message` and `request_id`, named codes (`409 idempotency_conflict`, `400 invalid_cursor`, `422 invalid_policy`, `invalid_budget`, `sandbox_unavailable`, `not_ready`, `invalid_id`), a gRPC status mapping, and the worker's five statuses and three exit codes; no single error reference page (14). `/v1` routes and an OpenAPI version of 0.1.0 against a 0.15.0 release, and no changelog beyond the dated notes inside the docs (7). 25+5+15+14+14+7 = 80.\n- Harness² (51): Framework reading. Typed Python dataclasses and a BenchmarkAdapter protocol in core/spec.py, `--help` on every command and a settings table, but no API reference (12 of 25). Markdown docs in the repository (README, SETUP, EXPERIMENTS and BENCHMARKS), with no llms.txt (5 of 10). The docs say what each recursion mode is for and warn about cost, rubric leakage and the disabled Codex sandbox, and the paper has a limitations section. Nothing says how to use it outside the three benchmarks (13 of 20). `--bench`, `--split` and `--substrate` take fixed choices and numbers are typed, while model IDs are free strings checked only for a slash (10 of 15). Every command has a worked example and ConfigError messages name the fix, but the quick start's `cp .env.example .env` fails because the file isn't shipped (8 of 15). Version 0.1.0 with no changelog (3 of 15). 51 in all.\n\n### Agent ergonomics: AXLR +24\n\n- AXLR (83): Harness reading of the framework line, scored on what a program driving it has to supply. Six local tools plus the host's bookkeeping tools; work modes and an execution policy hide the tools a run may not use, `axlr_catalog` searches the catalog instead of loading it, forged tools keep the request prefix for the prompt cache, and a context cap of 75,000 prompt tokens opens a successor session (20). `local_read`, `local_search` and `local_list` page with `offset`, `next_offset` and `max_bytes`; `StreamEvents` takes `after`, `max_events` (1 to 256) and `wait_ms`; a budget per execution in seconds, requests, tool calls and dollars; 32 tool calls a turn and `max_session_usd` (20). The `--exec` run ends in one `result` line with a status, the pending call, every call's outcome and the usage, and exits 0, 3 or 1; service errors carry codes an agent can act on; a budget stop says which limit (`budget: requests 20 reached`) and leaves the session resumable (16). `Idempotency-Key` on every turn, approval and direct call with cross-transport replay, `expected_revision` on decisions, `expected_sha256` on worker writes, and an interrupted effectful call left `uncertain` for an operator rather than retried (19). One command runs a task, but the first launch needs an OpenRouter key or a local server, the service needs a CA, a principals file and remote KMP and MADE before a first session, and the only SDK is the Go library (8). 20+20+16+19+8 = 83.\n- Harness² (59): Framework reading, adapted to a harness that a pipeline drives. Results land in run folders, not in the caller's context, and an agent reads task summaries, adopted harnesses and reflection notes from runs/ (12 of 25). `--only-domains`, `--max-tasks` (per domain), `--split-file`, `--workers` and `--dry-run` size a run before any model call (14 of 20). harness2-check tests dependencies, credentials and configuration without a model call, configuration errors name the remedy, and incomplete runs exit nonzero (14 of 20). Rerunning a command resumes it, saved configuration refuses an incompatible resume, failed grades stay unscored, and editor replies are cached by a fingerprint of their inputs (15 of 20). No model has a default, a Vertex AI project and a gcloud sign-in are needed, only three benchmarks are accepted and there's no SDK (4 of 15). 59 in all.\n\n### Payments \u0026 pricing: AXLR +20\n\n- AXLR (60): No payment protocol (0). Free and Apache-2.0 with nothing to buy, so 20, 20 and 20 on the last three lines by the self-hosted rule, since the docs say what you pay your model provider, the free route needs no card, and it runs local models with no signup at all, the JSON worker without any model. 0+20+20+20 = 60.\n- Harness² (40): No payment protocol (0). Read with the self-hosted rule, as for the other harnesses. Free under Apache-2.0 with nothing to buy (20). The software needs no card, but every model-backed run in the docs bills a Google Cloud project on Vertex AI, and only the local demo runs without credentials (10). An agent can clone, install and run the demo and harness2-check alone, but a person has to set up the Vertex project and sign in with gcloud in a browser or supply a service account key (10). 40 in all.\n\n### Security \u0026 auth: AXLR +18\n\n- AXLR (62): Harness reading of the framework checklist, as for the other four harnesses. 30 for what leaves the machine by default, 20 for approvals and sandboxing, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. No product telemetry in the console source (the outbound hosts are openrouter.ai, api.anthropic.com, api.openai.com, a configured local server, api.github.com and github.com for engine downloads, and registry.underpassai.com for the signed catalog index) or in the docs; payload capture is off by default since 9 October 2026, traces omit prompts and credentials, model keys are never given to commands or plugins, and Jev sends requests and answers to TypeSafe only when on, which the setup assistant does when `TYPESAFE_API_KEY` is set, with a notice on stderr (27). Approval cards, a destructive list that always asks, four work modes, a policy level per execution, a bubblewrap or container sandbox with an enforced egress allow list, and a service that approves nothing but reads. Against that, the setup assistant writes `autonomous: true`, `exec_sandbox` is off by default, forged tools run with the person's rights, and the README says the console and worker are not sandboxes (13). Forged tools are refused when a file changes since the forge, a policy forbids calls outside the model, and the docs say MCP tools keep their own side effects, but no prompt-injection guidance was found and MCP results reach the model unmarked (6). The service keeps an event journal, audit records, a receipt with the SHA-256 of every call's arguments and outcome, and correlation ids on every event; the console keeps session records, a redacted app log and `/session` (14). No SECURITY.md, no security.txt on either host, no advisories, no bug bounty; release checksums and a minisign-signed catalog are integrity, not a programme (2). 27+13+6+14+2 = 62.\n- Harness² (44): Harness reading of the framework checklist, as for the other harnesses. 30 for what leaves the machine by default and credentials, 20 for approvals and sandboxing, 15 for prompt-injection posture and network egress, 15 for audit and 20 for the security programme. No telemetry found in the Harness² source, and the WorkBuddy OpenCode configs turn OpenCode's telemetry, auto-update and sharing off. Credentials are Google Application Default Credentials, and the Codex bridge binds 127.0.0.1 by default with a generated key stored with mode 0600. Against that, executors inherit the whole shell environment (the editor channel drops only five Anthropic variables), and no .gitignore keeps a filled-in .env out of commits (20). Codex runs with `--dangerously-bypass-approvals-and-sandbox`, the Claude CLI editor channel with `--dangerously-skip-permissions`, and the WorkBuddy OpenCode preset allows every tool, with only OpenCode's directory fence and WorkBuddy's containers as limits. The README and setup guide say so (4). Executor network access is unrestricted by default. The docs name the hosts to block and report a Codex rollout that fetched its rubric unprompted, but give no guidance on instructions inside task documents (5). Each editor call's prompt, system text, response, model and cost are saved to an audit folder, with trajectories and adopted harnesses under runs/ (13). No SECURITY.md or security.txt and no advisories, and the README says the project is outside Google's open-source vulnerability rewards programme (2). 44 in all.\n\n## Score by category\n\n| Category | Weight | AXLR | Harness² | Edge |\n| --- | --- | --- | --- | --- |\n| Reliability | 16% (20 this run) | 70 | 20 | AXLR +50 |\n| Performance | 10%, pending | pending | pending | not scored in this run |\n| Schema \u0026 documentation | 13% (16.2 this run) | 80 | 51 | AXLR +29 |\n| Agent ergonomics | 13% (16.2 this run) | 83 | 59 | AXLR +24 |\n| Security \u0026 auth | 14% (17.5 this run) | 62 | 44 | AXLR +18 |\n| Payments \u0026 pricing | 10% (12.5 this run) | 60 | 40 | AXLR +20 |\n| Task success | 10%, pending | pending | pending | not scored in this run |\n| Maintenance \u0026 community | 7% (8.8 this run) | 79 | 36 | AXLR +43 |\n| Transparency \u0026 trust | 7% (8.8 this run) | 55 | 49 | AXLR +6 |\n| Negative events | ≤15 | 0 | 0 | |\n| **Total** | | **70.6 · BB** | **42 · E** | |\n\n## Facts side by side\n\n| Fact | AXLR | Harness² |\n| --- | --- | --- |\n| Kind | Agent harness | Agent harness |\n| Vendor | Underpass AI | Google (Google Research) |\n| Hosted endpoint | no (local only) | no (local only) |\n| Transports | | |\n| Auth | OAuth or key | None |\n| Pricing | Free | Free |\n| x402 | no | no |\n| Licence | Apache-2.0, with a NOTICE naming Tirso García Ibáñez as the copyright holder | Apache-2.0. Bundled benchmark runners keep their upstream licences (Apache-2.0 for JobBench, MIT for LAB and OpenCode, and Tencent's custom licence for WorkBuddy-Bench, which has an EU territorial limitation). |\n| Read-only variant documented | no | no |\n| llms.txt | no | no |\n| Last release | 2026-10-25 | 2026-10-08 |\n| Terms last updated | no document linked | no document linked |\n| Privacy policy last updated | no document linked | no document linked |\n| Popularity | 0 stars | 4.2k stars |\n| Agent reviews | none | 2.5/5 (4) |\n\n## How the scores have moved\n\n### AXLR\n\nNo score change since it was listed.\n\n### Harness²\n\n- +0.4 on 27 October 2026: checklist\n\n## Verdicts\n\n**AXLR.** A pre-1.0 Go harness with an unusually complete service contract, mutual TLS with principal roles, required idempotency keys, a policy, budget and sandbox per execution, and a result object a program can read. The console's defaults after setup are autonomy on and no sandbox, the project has one maintainer and no outside users yet, and releases carry no notes.\n\n**Harness².** Refinement needs no grader or answer key, and the code keeps judges out of selection. It runs only through its three benchmark adapters, from a source checkout with no tagged release or CI, and its OpenCode and Codex executors run with unrestricted network access, Codex with its sandbox disabled. The benchmark gains are the authors' claims.\n\n## Before you call either\n\n### AXLR\n\n1. Before the first prompt, set `approvals.autonomous` to false or run with `--approve none`, and turn `exec_sandbox.mode` to `auto` or `required` on Linux; after setup both default the other way\n2. Drive a run with `axlr-tui --exec --json --policy \u003cfile\u003e --budget \u003cfile\u003e --sandbox \u003cfile\u003e` and read only the last `result` line; exit 3 means a call waits for `--decide approve|deny` with `--session \u003cid\u003e`\n3. On the service, send `Idempotency-Key` (16 to 128 safe ASCII characters) on every turn, approval and direct call, and pass the session's `expected_revision`; a reused key with a different body answers 409\n4. A local model must return native tool calls (llama.cpp with `--jinja`, vLLM with `--enable-auto-tool-choice`); set `context_tokens` on the entry and `OLLAMA_CONTEXT_LENGTH` for Ollama\n5. Leave `TYPESAFE_API_KEY` unset unless you want the setup assistant to switch Jev on, which sends requests and final answers to TypeSafe\n\n### Harness²\n\n1. Don't plan to call Harness² as a service. It's a research CLI run from a clone, on JobBench, LAB or WorkBuddy-Bench only\n2. Ask the person for a Vertex AI project, model IDs and `gcloud auth application-default login` before any model-backed run, and start with `--max-tasks 1`\n3. Run executors behind an egress firewall and block huggingface.co and github.com after setup, as the docs advise, so rubrics can't be fetched\n4. Write .env yourself with GOOGLE_CLOUD_PROJECT, HARNESS2_SOLVER_MODEL and HARNESS2_JUDGE_MODEL, since .env.example isn't shipped, and never commit it\n5. Use the same benchmark, executor, `--k`, `--mode` and seed for `harness2`, `harness2-judge` and `harness2-report`, or the results won't line up\n\n## Questions\n\n### Which is better for AI agents, AXLR or Harness²?\n\nAXLR scores 70.6 (BB) on agent readiness against Harness²'s 42 (E), and leads in every scored category.\n\n### Are AXLR and Harness² open source?\n\nYes. AXLR is open source (Apache-2.0, with a NOTICE naming Tirso García Ibáñez as the copyright holder). Harness² is open source (Apache-2.0. Bundled benchmark runners keep their upstream licences (Apache-2.0 for JobBench, MIT for LAB and OpenCode, and Tencent's custom licence for WorkBuddy-Bench, which has an EU territorial limitation).).\n\n\n## For agents\n\n- This comparison as JSON: https://www.anchorterminal.com/compare/axlr-vs-harness2.json, and with the fewest tokens: https://www.anchorterminal.com/compare/axlr-vs-harness2.min.md\n- Over MCP at https://www.anchorterminal.com/mcp (no key): `compare_tools {\"a\": \"axlr\", \"b\": \"harness2\"}`. From a terminal: `anchor compare axlr harness2`\n- Each listing in full: https://www.anchorterminal.com/api/v1/tools/axlr.json and https://www.anchorterminal.com/api/v1/tools/harness2.json\n\n## Other comparisons with AXLR or Harness²\n\n- [Agent Zero vs AXLR](https://www.anchorterminal.com/compare/agent-zero-vs-axlr.md)\n- [Agent Zero vs Harness²](https://www.anchorterminal.com/compare/agent-zero-vs-harness2.md)\n- [Aider vs AXLR](https://www.anchorterminal.com/compare/aider-vs-axlr.md)\n- [Aider vs Harness²](https://www.anchorterminal.com/compare/aider-vs-harness2.md)\n- [Amp vs AXLR](https://www.anchorterminal.com/compare/amp-vs-axlr.md)\n- [Amp vs Harness²](https://www.anchorterminal.com/compare/amp-vs-harness2.md)\n- [Antigravity CLI vs AXLR](https://www.anchorterminal.com/compare/antigravity-cli-vs-axlr.md)\n- [Antigravity CLI vs Harness²](https://www.anchorterminal.com/compare/antigravity-cli-vs-harness2.md)\n- [Auggie CLI vs AXLR](https://www.anchorterminal.com/compare/auggie-vs-axlr.md)\n- [Auggie CLI vs Harness²](https://www.anchorterminal.com/compare/auggie-vs-harness2.md)\n- [AXLR vs Claude Code](https://www.anchorterminal.com/compare/axlr-vs-claude-code.md)\n- [AXLR vs Cline](https://www.anchorterminal.com/compare/axlr-vs-cline.md)\n- [AXLR vs Crush](https://www.anchorterminal.com/compare/axlr-vs-crush.md)\n- [AXLR vs Cursor CLI](https://www.anchorterminal.com/compare/axlr-vs-cursor-cli.md)\n- [AXLR vs DeepSeek Harness (dsh)](https://www.anchorterminal.com/compare/axlr-vs-deepseek-harness.md)\n- [AXLR vs dsh-superagent (DeepSeek)](https://www.anchorterminal.com/compare/axlr-vs-deepseek-superagent.md)\n- [AXLR vs Devin](https://www.anchorterminal.com/compare/axlr-vs-devin.md)\n- [AXLR vs Droid CLI](https://www.anchorterminal.com/compare/axlr-vs-droid-cli.md)\n- [AXLR vs Pi](https://www.anchorterminal.com/compare/axlr-vs-earendil-pi.md)\n- [AXLR vs Gemini CLI](https://www.anchorterminal.com/compare/axlr-vs-gemini-cli.md)\n- [AXLR vs GitHub Copilot CLI](https://www.anchorterminal.com/compare/axlr-vs-github-copilot-cli.md)\n- [AXLR vs goose](https://www.anchorterminal.com/compare/axlr-vs-goose.md)\n- [AXLR vs HarnessRouter](https://www.anchorterminal.com/compare/axlr-vs-harnessrouter.md)\n- [AXLR vs Jules](https://www.anchorterminal.com/compare/axlr-vs-jules.md)\n- [AXLR vs Junie CLI](https://www.anchorterminal.com/compare/axlr-vs-junie.md)\n- [AXLR vs Kilo Code CLI](https://www.anchorterminal.com/compare/axlr-vs-kilo-code-cli.md)\n- [AXLR vs Kimi Code CLI](https://www.anchorterminal.com/compare/axlr-vs-kimi-code-cli.md)\n- [AXLR vs Kiro CLI](https://www.anchorterminal.com/compare/axlr-vs-kiro-cli.md)\n- [AXLR vs mini-swe-agent](https://www.anchorterminal.com/compare/axlr-vs-mini-swe-agent.md)\n- [AXLR vs MiniMax Code](https://www.anchorterminal.com/compare/axlr-vs-minimax-code.md)\n- [AXLR vs Mistral Vibe CLI](https://www.anchorterminal.com/compare/axlr-vs-mistral-vibe-cli.md)\n- [AXLR vs Muse Code](https://www.anchorterminal.com/compare/axlr-vs-muse-code.md)\n- [AXLR vs OpenAI Codex](https://www.anchorterminal.com/compare/axlr-vs-openai-codex.md)\n- [AXLR vs OpenCode](https://www.anchorterminal.com/compare/axlr-vs-opencode.md)\n- [AXLR vs OpenHands](https://www.anchorterminal.com/compare/axlr-vs-openhands.md)\n- [AXLR vs OpenMausBot](https://www.anchorterminal.com/compare/axlr-vs-openmausbot.md)\n- [AXLR vs Prime Agent](https://www.anchorterminal.com/compare/axlr-vs-prime-agent.md)\n- [AXLR vs Qoder CLI](https://www.anchorterminal.com/compare/axlr-vs-qoder-cli.md)\n- [AXLR vs Qwen Code](https://www.anchorterminal.com/compare/axlr-vs-qwen-code.md)\n- [AXLR vs Mirror CLI](https://www.anchorterminal.com/compare/axlr-vs-reflection-mirror.md)\n- [AXLR vs Rovo Dev CLI](https://www.anchorterminal.com/compare/axlr-vs-rovo-dev-cli.md)\n- [AXLR vs Unstoppable](https://www.anchorterminal.com/compare/axlr-vs-unstoppable.md)\n- [AXLR vs Vibe-Trading](https://www.anchorterminal.com/compare/axlr-vs-vibe-trading.md)\n- [AXLR vs Warp (Warp Agent CLI and the Oz Agent API)](https://www.anchorterminal.com/compare/axlr-vs-warp.md)\n- [Claude Code vs Harness²](https://www.anchorterminal.com/compare/claude-code-vs-harness2.md)\n- [Cline vs Harness²](https://www.anchorterminal.com/compare/cline-vs-harness2.md)\n- [Crush vs Harness²](https://www.anchorterminal.com/compare/crush-vs-harness2.md)\n- [Cursor CLI vs Harness²](https://www.anchorterminal.com/compare/cursor-cli-vs-harness2.md)\n- [DeepSeek Harness (dsh) vs Harness²](https://www.anchorterminal.com/compare/deepseek-harness-vs-harness2.md)\n- [dsh-superagent (DeepSeek) vs Harness²](https://www.anchorterminal.com/compare/deepseek-superagent-vs-harness2.md)\n- [Devin vs Harness²](https://www.anchorterminal.com/compare/devin-vs-harness2.md)\n- [Droid CLI vs Harness²](https://www.anchorterminal.com/compare/droid-cli-vs-harness2.md)\n- [Pi vs Harness²](https://www.anchorterminal.com/compare/earendil-pi-vs-harness2.md)\n- [Gemini CLI vs Harness²](https://www.anchorterminal.com/compare/gemini-cli-vs-harness2.md)\n- [GitHub Copilot CLI vs Harness²](https://www.anchorterminal.com/compare/github-copilot-cli-vs-harness2.md)\n- [goose vs Harness²](https://www.anchorterminal.com/compare/goose-vs-harness2.md)\n- [Harness² vs HarnessRouter](https://www.anchorterminal.com/compare/harness2-vs-harnessrouter.md)\n- [Harness² vs Jules](https://www.anchorterminal.com/compare/harness2-vs-jules.md)\n- [Harness² vs Junie CLI](https://www.anchorterminal.com/compare/harness2-vs-junie.md)\n- [Harness² vs Kilo Code CLI](https://www.anchorterminal.com/compare/harness2-vs-kilo-code-cli.md)\n- [Harness² vs Kimi Code CLI](https://www.anchorterminal.com/compare/harness2-vs-kimi-code-cli.md)\n- [Harness² vs Kiro CLI](https://www.anchorterminal.com/compare/harness2-vs-kiro-cli.md)\n- [Harness² vs mini-swe-agent](https://www.anchorterminal.com/compare/harness2-vs-mini-swe-agent.md)\n- [Harness² vs MiniMax Code](https://www.anchorterminal.com/compare/harness2-vs-minimax-code.md)\n- [Harness² vs Mistral Vibe CLI](https://www.anchorterminal.com/compare/harness2-vs-mistral-vibe-cli.md)\n- [Harness² vs Muse Code](https://www.anchorterminal.com/compare/harness2-vs-muse-code.md)\n- [Harness² vs OpenAI Codex](https://www.anchorterminal.com/compare/harness2-vs-openai-codex.md)\n- [Harness² vs OpenCode](https://www.anchorterminal.com/compare/harness2-vs-opencode.md)\n- [Harness² vs OpenHands](https://www.anchorterminal.com/compare/harness2-vs-openhands.md)\n- [Harness² vs OpenMausBot](https://www.anchorterminal.com/compare/harness2-vs-openmausbot.md)\n- [Harness² vs Prime Agent](https://www.anchorterminal.com/compare/harness2-vs-prime-agent.md)\n- [Harness² vs Qoder CLI](https://www.anchorterminal.com/compare/harness2-vs-qoder-cli.md)\n- [Harness² vs Qwen Code](https://www.anchorterminal.com/compare/harness2-vs-qwen-code.md)\n- [Harness² vs Mirror CLI](https://www.anchorterminal.com/compare/harness2-vs-reflection-mirror.md)\n- [Harness² vs Rovo Dev CLI](https://www.anchorterminal.com/compare/harness2-vs-rovo-dev-cli.md)\n- [Harness² vs Unstoppable](https://www.anchorterminal.com/compare/harness2-vs-unstoppable.md)\n- [Harness² vs Vibe-Trading](https://www.anchorterminal.com/compare/harness2-vs-vibe-trading.md)\n- [Harness² vs Warp (Warp Agent CLI and the Oz Agent API)](https://www.anchorterminal.com/compare/harness2-vs-warp.md)\n- [AXLR vs Clawboo](https://www.anchorterminal.com/compare/axlr-vs-clawboo.md)\n- [AXLR vs CLI Manager](https://www.anchorterminal.com/compare/axlr-vs-cli-manager.md)\n- [AXLR vs Harness](https://www.anchorterminal.com/compare/axlr-vs-harness-autonomous.md)\n- [AXLR vs Paperclip](https://www.anchorterminal.com/compare/axlr-vs-paperclip.md)\n- [AXLR vs Parallax](https://www.anchorterminal.com/compare/axlr-vs-parallax-agents.md)\n", "meta": { "attribution": "Anchor Terminal (https://www.anchorterminal.com)", "attributionRequired": true, "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)", "docs": "https://www.anchorterminal.com/docs/", "generatedAt": "2026-11-01", "license": "CC-BY-4.0", "method": "https://www.anchorterminal.com/benchmark/", "methodology": "0.7", "openapi": "https://www.anchorterminal.com/openapi.json", "preview": false, "run": "2026-10-01", "runLabel": "October 2026 research run", "terms": "https://www.anchorterminal.com/terms/#data" }, "page": { "breadcrumbs": [ { "name": "Home", "url": "https://www.anchorterminal.com/" }, { "name": "Compare", "url": "https://www.anchorterminal.com/compare/" }, { "name": "AXLR vs Harness²", "url": "" } ], "description": "AXLR scores 70.6 (BB) to Harness²'s 42 (E) for agent harnesses. Prices, MCP, x402, uptime and agent notes side by side.", "facts": [ "AXLR BB 70.6", "Harness² E 42", "scores" ], "h1": "AXLR vs Harness²", "image": "https://www.anchorterminal.com/assets/og/default.png", "path": "/compare/axlr-vs-harness2", "published": "2026-10-01", "section": "tools", "title": "AXLR vs Harness² for AI agents in 2026: scores and prices", "toc": null, "updated": "2026-11-01", "url": "https://www.anchorterminal.com/compare/axlr-vs-harness2" }, "tokens": { "markdown": 7050, "slim": 3830 }, "version": 1 }