VibeHard Documentation.

How to install the VibeHard Agent Edition 2.0.7, connect your agent, and run a build.

1. Install (Mac and Linux)

curl -fsSL https://vibehard.ai/download/install-agent.sh | bash

The installer reads the release manifest, downloads the current release, checks its SHA-256 checksum before installing and stops on any mismatch. It installs ~/.local/bin/vibehard-agent, starts the background service, and prints the MCP config entry for your agent. Downloads: Mac (Apple Silicon), Linux x86_64 and Linux ARM64. Download page.

The Community Edition is free for personal use only. Using it for a business, a client or a paid product needs the commercial-use license, which comes with the Software Factory plan ($199/mo).

2. Connect your agent

VibeHard works with any agent that supports MCP, the standard way AI agents connect to tools. Add this entry to your agent’s MCP config (the installer can do it for you):

{
  "mcpServers": {
    "vibehard-agent": {
      "command": "/Users/you/.local/bin/vibehard-agent",
      "args": ["mcp"]
    }
  }
}

Use the absolute path the installer prints (for example /Users/you/.local/bin/vibehard-agent on a Mac or /home/you/.local/bin/vibehard-agent on Linux); MCP clients do not expand ~. The installer writes this same entry, named vibehard-agent, when you agree at the prompt or pass --mcp-register.

The factory makes no calls to AI model providers itself. Your agent does the reasoning, so your code goes only to the model provider your agent already uses.

3. Tools your agent gets (16)

  • — vibehard_session_open / vibehard_session_close: Open and close a session with the factory.
  • — vibehard_build_start: Submit a software request. The factory runs the pipeline and hands the reasoning steps to your agent as work orders.
  • — vibehard_workflow_brief: Prepare a workflow brief or evidence bundle locally on your machine. Hosted workflows are coming soon.
  • — vibehard_task_next / vibehard_task_submit: Fetch the next work order and return your agent’s result; the pipeline resumes.
  • — vibehard_review_next / vibehard_review_respond: See the next open human checkpoint and record the answer.
  • — vibehard_status / vibehard_progress: Check a job’s status, or get a plain-English progress report.
  • — vibehard_gate: Run the verification chain on a folder as a background job.
  • — vibehard_deploy: Run the chain and, only if it passes, write the sealed pass marker.
  • — vibehard_snapshot / vibehard_rollback: Take a labeled snapshot, or roll the workspace back to one. A safety snapshot is taken first, and untracked files are set aside, not deleted. Rollback needs a target and a reason, is limited to two per job, and may use one repair attempt.
  • — vibehard_handoff: Generate the handoff package for a finished job: the as-built report and a maintainer runbook.
  • — vibehard_install_info: Get install instructions and download details for your edition and platform.

4. How a build runs

Recorded stages: questions, reference analysis, buy before build, planning documents, plan review, your approval, plan seal, slice-by-slice build, repair loop, the 44-gate chain, and the build audit and handoff. The full pipeline.

How a build works

You describe a program in plain English; your agent hands the request to VibeHard and does the thinking and the typing. VibeHard decides what happens next and checks everything.

  • Questions. If your request is ambiguous, VibeHard asks (for example, how you will use the program, or how to round money). You answer; the answers go into the plan. On the strictest setting, sensitive questions must be answered by a person, not by the agent.
  • Plan. VibeHard writes the planning documents and three reviewers audit them. You then approve the plan, ask for changes, or stop. Nothing is built before you approve, and the approved plan is sealed with a digest.
  • Build. Your agent writes the code in slices from work orders; the factory checks each result.
  • Gates. A 44-gate chain runs on the result. Gates that do not apply report n/a. Failures go back to your agent to repair, up to ten attempts.
  • Verified means the chain passed with nothing blocking and the pass marker matches the current files. It is not a guarantee that the program is bug-free.
  • Handoff. You get the program, an as-built report and a maintainer runbook, and every build is snapshotted so you can roll back.

Try it: Your first build · What you can build.

5. The gates

Every full build runs the same 44-gate chain (48 with a paid license). Gates that do not apply report n/a. The secret, code-pattern and dependency scanners run in containers on your machine; if your machine cannot run containers, those gates cannot run and the chain cannot pass. You can add your own gates. Every gate.

6. Updates and rollback

Installed copies check for a signed update about every 6 hours, verify the signature before installing, and refuse downgrades. Run vibehard-agent update to update now. Every build is snapshotted before it starts and after each successful repair.

7. What is not available yet

  • — Hosted factory runs, hosted agentic workflows and hosted app connections: coming soon.
  • — VibeHard Cloud, which runs the security scanners for machines that cannot: closed testing.
  • — Custom gates are checks, not certifications. VibeHard is not SOC 2 audited; do not send health records or card numbers to any VibeHard service.