Build AI coding workflows you can trust and verify.
HAR runs every coding agent in an isolated, consistent environment, with deterministic validation, verifiable proof, and full observability. Extensible to your own workflow.
Every agent works the same way
Each agent gets its own isolated worktree, ports, and environment, set up the same way every time, one agent or a whole fleet.
Trust every result
Each task runs the same deterministic checks and leaves logs, artifacts, and a validated tree hash your reviewers can inspect.
Zero vendor lock-in
The harness lives in your own repo and reads the same for any coding agent, so you can switch tools freely and adapt it to your workflow.
Everything your agents need to run, verify, and hand off work you can trust.
Isolated slots
Each agent works in its own worktree with dedicated ports and a database, so the main checkout stays clean.
Deterministic checks
The project's real checks run through the same pipeline every time and return a consistent result.
Evidence trail
Every run can leave logs, artifacts, and a validated tree hash tied to the exact code that was checked.
One contract
A machine-readable .har/ folder that Claude Code, Cursor, Codex, and any MCP agent read the same way.
Mission Control
A local dashboard for every repository, worktree, run, validation, and artifact across your projects.
Plugins
Extend the harness with framework-specific stages like Playwright, or build your own.
From an assigned task to a validated branch.
HAR gives every agent its own clean slot, runs the project's real checks on the result, and keeps the evidence, so agents stay focused on the code and reviewers can trust what comes back.
The agent reads one stable interface.
The agent asks HAR what the repository can launch, verify, inspect, reset, and tear down, all through one interface.
MCP tools{
"profile": "default",
"agentSlots": 3,
"stages": [
"launch", "verify", "test"
]
}See what every agent is doing.
Mission Control is a local dashboard that puts every repository, agent slot, run, verification, log, and artifact in one view. Watch one agent or your whole fleet as it works, and open any run to see exactly what happened.
Live slots
Watch every active slot with its worktree, branch, and preview URL, and spot a dirty tree or drift as soon as it appears.
Run history
Open a timeline of every run with its stage, trigger, duration, and status, alongside the logs, artifacts, and cost behind each one.
Traceable results
Every result links back to the validation stages that ran and the exact Git tree they checked.
Web app
For web applications and services. HAR runs your app with per-slot ports and optional shared Docker infrastructure.
CLI / library
For CLIs, libraries, and packages. HAR sets up the worktree and toolchain it needs to build and test.
iOS app
For Xcode and Swift apps. HAR drives xcodebuild and the iOS Simulator, with optional RocketSim flows.
Playwright
Adds a browser-e2e stage with frontend, API health, and accessibility smoke specs.
Semgrep
Adds a sast stage that scans every agent worktree for security findings before code lands.
Your own checks
Custom stages for project-only checks, or publish a plugin from the os-factory/har-plugin template via path, npm, or git.
Bring HAR to your whole organization.
HAR HQ is the hosted team layer for HAR, with governance, analytics, and cost attribution across every agent in your org.
Go to HAR HQ