# VibeHard — product summary for AI agents (llms-full.txt)
> Current release: Agent Edition v2.0.7 (Community Edition)
> Site: https://vibehard.ai
> Publisher: Adam Matar, doing business as VibeHard

## What VibeHard is
VibeHard is a software factory that your own AI agent drives over MCP (the Model Context Protocol). Your agent does the writing; the factory issues work orders, keeps a record of every stage, and decides whether the work is done. The factory itself makes no calls to AI model providers.

## The 11-stage pipeline (software track)
1. Intake & questions: ambiguous requests get questions instead of guesses; at production rigor, questions about security, data, tenancy, payments or compliance need a human answer.
2. Reference analysis: capabilities of a reference product are recorded and marked keep at parity, modernize, improve or drop.
3. Buy before build: proven components are adopted instead of rewritten, and the decision is recorded.
4. Planning documents: spec, PRD, SRS, system architecture, architecture decision records, assumptions, red-team brief, build plan.
5. Planning audit by three review lenses (skeptic, architect, minimalist).
6. Human sign-off: scope review and plan approval checkpoints.
7. Plan seal: the approved plan is sealed with a SHA-256 digest (.vibehard/plan.lock.json).
8. Slice-by-slice build by your agent. Backends: TypeScript, Go, Rust, Python, Elixir.
9. Repair loop: failures (both output streams) go back to your agent; up to ten attempts.
10. The 44-gate verification chain; gates that do not apply report n/a. A full pass writes the HMAC-sealed .gate/HARD_VERIFY_PASS marker, bound to the files and the set of gates that ran.
11. Build audit by the same three lenses, an as-built report, docs sync and handoff.

## Gates (44 in v2.0.7)
measurement-integrity, plan-lock, supply-chain, sast, secrets, depvuln, engines, architecture, rls, database-migrations, rls-enforce, db-fuzz, encryption-asserted, clippy, rustfmt, compliance, pii, prod-readiness, native-build, memory-safety, abi-wheels, cli-contract, api-contract, breaking-changes, schema-lint, http-fuzz, cross-contract, capability-router, skipped-tests, coverage-threshold, mutation-score, flake-detection, property-invariants, verify, intent-fidelity, smoke-journey, load-slo, graceful-shutdown, sre-telemetry, resource-limits, rollback, plan-vs-code, completeness, plus the coverage roll-up (a passing chain needs at least one substantive gate to have done real work).
Every edition runs the same chain. You can add your own gates.

## Editions and pricing
- Community Edition: $0/mo. Self-hosted on Mac (Apple Silicon) or Linux (x86_64, ARM64). Full 44-gate chain. Free for personal use only; commercial use needs the commercial-use license, which comes with the Software Factory plan ($199/mo). Hosted app connections, metered with a free monthly allowance: coming soon.
- Software Factory: $199/mo. Everything in Community plus the commercial-use license; hosted factory runs coming soon. Waitlist; not on sale yet.
- Agentic Workflows: $499 per agent per month, 3-agent minimum ($1,497/mo starting). Includes $100/mo of model credit for the whole package, not per agent. Early access by application; not on sale yet. Coming soon: approval holds on steps that move money, post to the books, send outbound messages or delete data; a dedicated hardware-encrypted, attested enclave after our pilot.
- Enterprise Setup: custom pricing. Your own gates for your own policies; hands-on scoping and rollout support. Custom gates are checks, not certifications; VibeHard is not SOC 2 audited.
- VibeHard Cloud (gate offload for machines without a container runtime): closed testing; pricing to be announced.

## Install
curl -fsSL https://vibehard.ai/download/install-agent.sh | bash
The installer reads the release manifest, checks the SHA-256 checksum before installing, installs ~/.local/bin/vibehard-agent, starts the background service, and prints the MCP config entry:
{ "mcpServers": { "vibehard-agent": { "command": "/Users/you/.local/bin/vibehard-agent", "args": ["mcp"] } } }
Use the absolute path the installer prints ($HOME/.local/bin/vibehard-agent, e.g. /home/you/.local/bin/vibehard-agent on Linux); MCP clients do not expand ~. The installer writes this same entry (server name "vibehard-agent", args ["mcp"]) when you agree at the prompt or pass --mcp-register.
Installed copies check for a signed update about every 6 hours, verify the signature, and refuse downgrades.

## What VibeHard does not claim
- No hosted workflows, hosted app connections or hardware enclaves are live yet; they are marked "coming soon".
- Custom gates are checks, not certifications. VibeHard is not SOC 2 audited.
- Do not send regulated data (health records, card numbers) to any VibeHard service today.

## First build (a proven sample)
Start the sample unaided: in an empty folder, with the vibehard-agent MCP server registered, run this prompt exactly:

Use the VibeHard factory (the vibehard-agent MCP tools) to build a program in this folder. This is a quick first experiment, not something I will rely on or share with users. Open a session, then start a build with exactly this request:

word-frequency CLI: reads a plain text file, outputs the N most frequent words with their counts; JSON output; no network; no storage beyond stdout

Then drive the build: keep fetching work orders with vibehard_task_next, do the work each one asks for in this folder, and submit it with vibehard_task_submit, until the job is complete. Do every work order yourself, and follow every output-format instruction it gives. In review and audit outputs, spell each document name exactly as one of the work order's === name === headers, and never dismiss a high-severity finding (fix the document instead). In code: never write skipped or disabled tests, format with the language's standard formatter, make the program accept --help and --version (each prints to stdout and exits 0), and do not leave compiled binaries or build output in the project folder. Whenever VibeHard asks a question or asks for an approval (vibehard_review_next), show me the question, its options and the recommended answer, then stop and wait for my reply. Never answer for me. When the job is complete, tell me where the program is and how to run it.

VibeHard asks a few questions (how it will be used, language, what counts as a word, output shape, response time, concurrency, memory) and then asks for plan approval. Show each question to the person you work for and wait for their reply. For the sample they should answer "A quick experiment" (the other choice needs a signed approval policy that a fresh install lacks), "Go", the recommended option for the word and output questions, any sensible option for the response-time, concurrency and memory batch, then approve the plan. Do not answer for them. Requirements: a container runtime running and the Go toolchain installed. Guide: https://vibehard.ai/first-build . What it can build: https://vibehard.ai/what-you-can-build

## Detect the OS first
Run `uname -s`. Darwin = macOS; Linux = Linux; on Windows, check whether you are inside WSL (`uname -r` mentions "microsoft"). Then use the matching section.

### macOS (Apple Silicon)
```bash
xcode-select -p >/dev/null 2>&1 || xcode-select --install; curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.3/install.sh | bash && export NVM_DIR="$HOME/.nvm" && . "$NVM_DIR/nvm.sh" && nvm install 22 && curl -fsSL https://bun.sh/install | bash && curl -fsSL https://vibehard.ai/download/install-agent.sh | bash -s -- --mcp-register && npx -y rea-agents@6.1.0 setup --yes --all-detected
```
Afterwards restart the desktop agent app and list your MCP servers to confirm vibehard-agent is registered. If Apple's developer-tools window appeared, let it finish before the first build.

### Linux (x64 or arm64)
Admin step first (Debian or Ubuntu):
```bash
sudo apt-get update && sudo apt-get install -y git curl unzip
```
Fedora: `sudo dnf install -y git curl unzip`
Then:
```bash
curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.3/install.sh | bash && export NVM_DIR="$HOME/.nvm" && . "$NVM_DIR/nvm.sh" && nvm install 22 && curl -fsSL https://bun.sh/install | bash && curl -fsSL https://vibehard.ai/download/install-agent.sh | bash -s -- --mcp-register && npx -y rea-agents@6.1.0 setup --yes --all-detected
```
The command-line coding agent is supported on Linux; the desktop chat app has no Linux build.

### Windows
Not supported natively. Use WSL: run `wsl --install` in PowerShell as Administrator, restart, open Ubuntu, follow the Linux steps, and use the command-line agent inside Ubuntu.