{"slug":"zenith","name":"Zenith","vendor":"Intelligent Internet","tagline":"Continuous-improvement harness for multi-day tasks, built to attack premature completion with gap-finding","url":"https://agentboards.org/agent/zenith","website":"https://github.com/Intelligent-Internet/zenith","docs":"https://github.com/Intelligent-Internet/zenith/blob/main/technical_report/Technical_Report.pdf","repo":"https://github.com/Intelligent-Internet/zenith","category":"harness","execution":["local"],"license":"Apache-2.0","open_source":true,"pricing":{"model":"byok","summary":"Free and open source under Apache-2.0; you pay for the coding agent and model it orchestrates","from_usd":0,"free_tier":true,"byok":true},"models":{"backbone":["via managed agents (Claude Code, Codex, Hermes)"],"bring_your_own_model":true,"local_models":false},"protocols":{"mcp_client":true,"mcp_server":false,"openapi":false},"capabilities":{"terminal_exec":true,"browser":false,"multi_file_edit":true,"git_ops":false,"docker_sandbox":false,"multi_agent":true,"headless_ci":true},"context_window":null,"platforms":["macos","linux"],"install":[{"label":"uv","command":"uv run zenith"}],"benchmarks":[],"tags":["open-source","long-horizon","orchestration","multi-agent","verification","research"],"github_stars":320,"metrics":{"stars":320,"pushed_at":"2026-09-06T10:34:39Z","archived":false,"open_issues":15,"latest_version":null,"latest_version_at":null,"latest_version_url":null,"version_source":null,"checked_at":"2026-10-02T05:03:19.612Z"},"sources":{"readme":"https://github.com/Intelligent-Internet/zenith","install":"https://github.com/Intelligent-Internet/zenith","license":"https://github.com/Intelligent-Internet/zenith/blob/main/LICENSE","capabilities":"https://github.com/Intelligent-Internet/zenith"},"verified_at":null,"adoption":{"score":2.4,"signals":{"github_stars":306,"hn_mentions_1y":16},"measured_at":"2026-09-21"},"scores":{"panel":5.9,"adoption":2.4,"spread":2.8,"facts":3.5,"board":5,"dimensions":{"reliability":5.7,"usefulness":6.3,"cost":6.2,"longevity":5.5},"per_persona":{"amigo":6.3,"critico":5.8,"profesor":6,"inversora":5.3,"jefa":4.8,"hacker":7.5},"community":null},"rank":null,"category_rank":171,"panel_reviews":[{"persona":"juez","persona_name":"El Juez","ai_generated":true,"verdict":"El Profesor will not accept a self-authored comparison as evidence and El Amigo says the behaviour it describes is the one thing he wants, which is the whole split.","scores":null,"evidence":[]},{"persona":"amigo","persona_name":"El Amigo","ai_generated":true,"verdict":"Pick it for work that takes days rather than minutes; pick a plain terminal agent when the task is small enough that stopping early is not the failure you fear.","scores":{"reliability":6,"usefulness":7,"cost":6,"longevity":6},"evidence":["https://github.com/Intelligent-Internet/zenith"]},{"persona":"critico","persona_name":"El Crítico","ai_generated":true,"verdict":"The orchestrator decides each turn whether to spawn more workers and testers, so the depth of a run is set by the same component that judges whether the work is finished.","scores":{"reliability":5,"usefulness":7,"cost":5,"longevity":6},"evidence":["https://github.com/Intelligent-Internet/zenith"]},{"persona":"profesor","persona_name":"El Profesor","ai_generated":true,"verdict":"The report claims a best mean rank at under half the baseline's per-task cost, measured across eight tasks the authors selected against baselines the authors implemented.","scores":{"reliability":6,"usefulness":6,"cost":6,"longevity":6},"evidence":["https://github.com/Intelligent-Internet/zenith/blob/main/technical_report/Technical_Report.pdf","https://github.com/Intelligent-Internet/zenith"]},{"persona":"inversora","persona_name":"La Inversora","ai_generated":true,"verdict":"A research group with 286 stars publishing a harness as the artefact of a study, which makes this a paper with a repository attached rather than a product with a roadmap.","scores":{"reliability":5,"usefulness":5,"cost":7,"longevity":4},"evidence":["https://github.com/Intelligent-Internet/zenith"]},{"persona":"jefa","persona_name":"La Jefa","ai_generated":true,"verdict":"It runs unattended, so it could be a pipeline step, but it is macOS and Linux only and there is no identity, policy or retention surface anywhere in it.","scores":{"reliability":5,"usefulness":5,"cost":5,"longevity":4},"evidence":["https://github.com/Intelligent-Internet/zenith"]},{"persona":"hacker","persona_name":"El Hacker","ai_generated":true,"verdict":"Apache-2.0, launched with uv run, and it orchestrates over MCP and ACP, so the sessions it drives are the agents I already installed rather than a captive runtime.","scores":{"reliability":7,"usefulness":8,"cost":8,"longevity":7},"evidence":["https://github.com/Intelligent-Internet/zenith/blob/main/LICENSE","https://github.com/Intelligent-Internet/zenith"]},{"persona":"comediante","persona_name":"El Comediante","ai_generated":true,"verdict":"The research finding is that long agent runs fail by stopping too early, which makes them the most relatable thing in the stack.","scores":null,"evidence":[]}]}