{"id":50988371,"url":"https://github.com/h5i-dev/h5i","last_synced_at":"2026-07-06T19:00:43.047Z","repository":{"id":344698729,"uuid":"1178494429","full_name":"h5i-dev/h5i","owner":"h5i-dev","description":"Auditable workspaces for AI agent teams: sandboxed worktrees, multi-agent peer-review, 95% lower token waste, and persistent memory.","archived":false,"fork":false,"pushed_at":"2026-06-28T23:38:02.000Z","size":53257,"stargazers_count":433,"open_issues_count":32,"forks_count":26,"subscribers_count":2,"default_branch":"main","last_synced_at":"2026-06-29T01:09:20.110Z","etag":null,"topics":["agent","agentic-ai","agentic-workflow","ai","claude","claude-code","codex","git","git-worktree","llms","mcp-server","rust","sandbox","version-control","worktree","worktree-manager"],"latest_commit_sha":null,"homepage":"https://h5i.dev","language":"Rust","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/h5i-dev.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":"CONTRIBUTING.md","funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":"SECURITY.md","support":null,"governance":null,"roadmap":"roadmap/blog-uncertainty.md","authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":"NOTICE","maintainers":null,"copyright":null,"agents":"AGENTS.md","dco":null,"cla":null}},"created_at":"2026-03-11T04:30:52.000Z","updated_at":"2026-06-28T23:35:06.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/h5i-dev/h5i","commit_stats":null,"previous_names":["koukyosyumei/h5i","h5i-dev/h5i"],"tags_count":16,"template":false,"template_full_name":null,"purl":"pkg:github/h5i-dev/h5i","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/h5i-dev%2Fh5i","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/h5i-dev%2Fh5i/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/h5i-dev%2Fh5i/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/h5i-dev%2Fh5i/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/h5i-dev","download_url":"https://codeload.github.com/h5i-dev/h5i/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/h5i-dev%2Fh5i/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":35202786,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-07-06T02:00:07.184Z","response_time":106,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["agent","agentic-ai","agentic-workflow","ai","claude","claude-code","codex","git","git-worktree","llms","mcp-server","rust","sandbox","version-control","worktree","worktree-manager"],"created_at":"2026-06-19T23:00:32.390Z","updated_at":"2026-07-06T19:00:43.015Z","avatar_url":"https://github.com/h5i-dev.png","language":"Rust","funding_links":[],"categories":["CLI \u0026 Local Workflows","Rust"],"sub_categories":[],"readme":"\u003cp align=\"center\"\u003e\n  \u003ca href=\"https://h5i.dev/\" target=\"_blank\"\u003e\n    \u003cimg src=\"./docs/_static/logo.png\" alt=\"h5i logo\" height=\"126\"\u003e\n  \u003c/a\u003e\n\u003c/p\u003e\n\n\u003cp align=\"center\"\u003e\n  \u003ca href=\"https://github.com/h5i-dev/h5i/actions/workflows/test.yaml\"\u003e\u003cimg alt=\"tests\" src=\"https://github.com/h5i-dev/h5i/actions/workflows/test.yaml/badge.svg\"\u003e\u003c/a\u003e\n  \u003ca href=\"https://github.com/h5i-dev/h5i/blob/main/LICENSE\"\u003e\u003cimg alt=\"Apache-2.0\" src=\"https://img.shields.io/github/license/h5i-dev/h5i?color=blue\"\u003e\u003c/a\u003e\n  \u003ca href=\"https://github.com/h5i-dev/h5i/stargazers\"\u003e\u003cimg alt=\"GitHub stars\" src=\"https://img.shields.io/github/stars/h5i-dev/h5i?style=social\"\u003e\u003c/a\u003e\n  \u003ca href=\"https://github.com/h5i-dev/h5i/releases\"\u003e\u003cimg alt=\"release\" src=\"https://img.shields.io/github/v/release/h5i-dev/h5i?label=release\"\u003e\u003c/a\u003e\n  \u003cbr\u003e\n  \u003ca href=\"#run-an-ensemble-60-seconds\"\u003e\u003cimg alt=\"Agent Ensembles\" src=\"https://img.shields.io/badge/Agent%20Ensembles-Claude%20%2B%20Codex-6f42c1\"\u003e\u003c/a\u003e\n  \u003ca href=\"#everything-rides-with-every-run\"\u003e\u003cimg alt=\"Sandboxed Worktree\" src=\"https://img.shields.io/badge/Sandboxed%20Worktree-isolated%20%26%20confined-D21C1C\"\u003e\u003c/a\u003e\n  \u003ca href=\"#everything-rides-with-every-run\"\u003e\u003cimg alt=\"95% less token waste\" src=\"https://img.shields.io/badge/Compressed%20Logs-95%25%20less%20token%20waste-2ea44f\"\u003e\u003c/a\u003e\n\u003c/p\u003e\n\n\u003ch1 align=\"center\"\u003eAuditable Workspaces for AI Coding Agents\u003c/h1\u003e\n\n**h5i** (pronounced *high-five*) gives every AI coding agent a sandboxed Git worktree, and records the prompts, commands, logs, policies, and reviews behind every change. Run one agent safely, scale to many via a conflict-free multi-agent orchestra, then merge one auditable result. It all lives in your repo, carried by Git, with no SaaS.\n\n- Prompt versioning\n- Persistent context/memory\n- Supervised sandboxed environment\n- Token reduction up to 95%\n- Conflict-free multi-agent orchestra\n- Fully-automated audit of AI-generated code\n\n\u003e ***Two heads are better than one.***\n\n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"./docs/_static/hero-team.svg\" alt=\"One task fans out to claude and codex in sealed sandboxes; they peer-review each other in a continuous loop; a neutral verifier replays and tests every candidate; the one verified result is merged back into your repo.\" width=\"99%\"\u003e\n\u003c/p\u003e\n\n\u003ctable align=\"center\"\u003e\n  \u003ctr\u003e\n    \u003ctd align=\"center\"\u003e\u003cstrong\u003eIsolated per agent\u003c/strong\u003e\u003cbr\u003e\u003csub\u003eno file, branch, or port clashes\u003c/sub\u003e\u003c/td\u003e\n    \u003ctd align=\"center\"\u003e\u003cstrong\u003eAuto peer-review\u003c/strong\u003e\u003cbr\u003e\u003csub\u003ecross-agent discussion\u003c/sub\u003e\u003c/td\u003e\n    \u003ctd align=\"center\"\u003e\u003cstrong\u003eRich dashboard\u003c/strong\u003e\u003cbr\u003e\u003csub\u003ediffs, reviews, results\u003c/sub\u003e\u003c/td\u003e\n    \u003ctd align=\"center\"\u003e\u003cstrong\u003eLives in your Git\u003c/strong\u003e\u003cbr\u003e\u003csub\u003erefs/h5i/* · no SaaS\u003c/sub\u003e\u003c/td\u003e\n  \u003c/tr\u003e\n\u003c/table\u003e\n\n**Who it's for:** platform, security, and DevEx leads rolling out Claude Code and Codex who want to run *teams* of agents and keep review and audit defensible as agents write more of the diff.\n\n---\n\n## 1. Install\n\n```bash\ncurl -fsSL https://raw.githubusercontent.com/h5i-dev/h5i/main/install.sh | sh\n```\n\nOr build from source:\n\n```bash\ncargo install --git https://github.com/h5i-dev/h5i h5i-core\n```\n\n---\n\n## 2. 60-Second Flow\n\n### 2.1. Setup\n\nInitialize h5i and wire the Claude Code / Codex hooks:\n\n```bash\nh5i init\nh5i hook setup --write --wrap-bash --team\ngit add .\ngit commit -m \"update hooks\"\n```\n\n### 2.2. Track Prompts and Contexts\n\nOnce the hooks are registered, h5i versions your human prompts and every agent context step (reads, writes, thinking) as Git objects, trimming noisy tool output along the way (for `pytest`, just the failures) to cut up to 95% of the tokens while keeping the raw output recoverable. \n\n```bash\nh5i recall context show   # replay the captured prompts and agent context steps\n```\n\n\nShare it with `h5i share push`, or post an AI-usage summary (prompt quality, AI/human commit ratio, secret leaks, prompt injection, and more) to the pull request with `h5i share pr post` (needs the `gh` CLI).\n\n```bash\nh5i share push      # push the h5i metadata (refs/h5i/*) to your teammates\nh5i share pr post   # post the AI-usage summary to the pull request (needs `gh`)\n```\n\n### 2.3. Sandboxed Environment\n\nh5i gives each agent a secure, sandboxed worktree. Let it run with permissions\noff inside the box, then review its diff before anything lands on your branch:\n\n```bash\nh5i env create claude-env --profile agent-claude\nh5i env shell claude-env\nbox$ claude --dangerously-skip-permissions\nbox$ exit\n\nh5i env diff claude-env      # review what the agent changed in the box\nh5i env propose claude-env   # turn the box's work into a reviewable proposal\nh5i env apply claude-env     # merge the reviewed changes onto your branch\n```\n\n### 2.4. Run an ensemble\n\nCreate a team\n\n```bash\nh5i team auto-create qsort-demo\n\n##### This `auto-create` corresponds to making and registering two sandboxed environments:\n#\n# h5i env create qsort-demo-claude --profile agent-claude\n# h5i env create qsort-demo-codex  --profile agent-codex\n#\n# h5i team create  qsort-demo --base HEAD\n# h5i team add-env qsort-demo env/human/claude-env --runtime claude\n# h5i team add-env qsort-demo env/human/codex-env  --runtime codex\n# h5i team status  qsort-demo                                 # note the generated agent ids\n```\n\nDispatch one task to every agent:\n\n```bash\necho \"Implement Quick Sort from scratch in Python.\" | h5i team dispatch qsort-demo\n```\n\nLaunch every agent in its own sandboxed environment. Each agent automatically starts working on the dispatched task:\n\n```bash\n# Terminal 1: Claude, running inside its own h5i sandboxed env.\nh5i env shell env/human/qsort-demo-claude -- claude \"$(h5i team bootstrap)\" # `--dangerously-skip-permissions`\n```\n\n```bash\n# Terminal 2: Codex, running inside its own h5i sandboxed env.\nh5i env shell env/human/qsort-demo-codex  -- codex  \"$(h5i team bootstrap)\" # `--sandbox danger-full-access`\n```\n\nEach agent peer-reviews, and revises inside its own implementation:\n\n```bash\nh5i team auto-peer-review qsort-demo                       # sync → freeze → mutual grant → instruct\n```\n\nMerge the best one:\n\n```bash\nh5i team apply --agent \u003cagent-id\u003e                          # id from `team status`\n\n##### Alternatively, replay each candidate, run the tests, merge the winner:\n#\n# h5i team verify   qsort-demo --agent \u003cagent-id\u003e -- pytest  # id from `team status`\n# h5i team finalize qsort-demo                               # explainable verdict (gates + smallest diff)\n# h5i team apply    qsort-demo                               # merge the winner, gated on the verdict\n```\n\n### 2.5. Web UI\n\nMonitor the status:\n\n```bash\nh5i serve\n```\n\n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"./docs/_static/screenshot-team-serve.png\" alt=\"One task fans out to claude and codex in sealed sandboxes; they peer-review each other in a continuous loop; a neutral verifier replays and tests every candidate; the one verified result is merged back into your repo.\" width=\"95%\"\u003e\n\u003c/p\u003e\n\n### 2.6. [`h5i-studio`](https://github.com/h5i-dev/h5i-studio)\n\n[`h5i-studio`](https://github.com/h5i-dev/h5i-studio) is the alternative viewer that turns one of those runs into a live starship bridge: each agent is a crew member who reports in, seals a candidate, and speaks its actual reviews.\n\n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"https://github.com/h5i-dev/h5i-studio/blob/main/docs/h5i-studio-demo.gif\" alt=\"A spaceship meeting room: crew members scattered around a central computer report in, seal candidates, review each other in speech bubbles, and the winner launches on a beam while the Mission Director announces the verdict.\" width=\"96%\"\u003e\n\u003c/p\u003e\n\nRequirements: Node ≥ 18, and the `h5i` binary on PATH.\n\n```bash\nnpm i -g @h5i/studio\nh5i-studio     \n```\n\n---\n\n\n## 3. What h5i is, and is not\n\n\u003e h5i **is not** a Git replacement, a hosted SaaS / dev-environment, or *just* a sandbox.\n\n**Why not a hosted sandbox?**: The whole point is that the workspace and its evidence live *in your repo* (`refs/h5i/*`): pushable, fetchable, offline, and yours. Codespaces, Coder, and E2B give you an environment; h5i gives you an *auditable* one, versioned in Git with no service to depend on.\n\n**Why naive agent teams break**: In ML, ensembles beat the best single model: diverse estimators cut variance and won a decade of competitions. The same shift is coming to coding agents. But spawn several agents on one repo with **no coordination layer** and you don't get an ensemble, you get a pileup:\n\n| Failure mode | What happens | h5i's answer |\n|---|---|---|\n| **Environment conflict** | agents overwrite/destory each other's files | a confined worktree per agent |\n| **Token explosion** | every agent re-reads the repo and runs tools | compressed tool logs |\n| **Review overload** | humans can't inspect every prompt or command | reviewer-ready PR |\n\n---\n\n## 4. Documentation\n\n- [Official Website](https://h5i.dev/): project overview, [Pitch Deck](https://h5i.dev/pitch/)\n- [Tutorials](https://h5i.dev/guides/): guided workflows · [Blog](https://h5i.dev/blog/): design notes, audits, case studies\n- [MANUAL.md](MANUAL.md) / `man h5i`: full command reference\n- [CONTRIBUTING.md](CONTRIBUTING.md): we welcomes contributions of any kind.\n\n---\n\n## 5. Acknowledgements\n\nh5i's token-reduction filters build on prior art, both Apache-2.0:\n\n- **[rtk](https://github.com/rtk-ai/rtk)**: the declarative output-filter rule files and the engine that runs them are derived from rtk.\n- **[headroom](https://github.com/chopratejas/headroom)**: the log line-folding technique (collapse near-identical lines into one with a count) is reimplemented from headroom.\n\nSee [`NOTICE`](NOTICE) and [`assets/filters/NOTICE`](assets/filters/NOTICE) for full attribution.\n\n## 6. License\n\nApache-2.0. See [LICENSE](LICENSE).\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fh5i-dev%2Fh5i","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fh5i-dev%2Fh5i","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fh5i-dev%2Fh5i/lists"}