Introduction
WeHarness helps teams build the AI agent inside their application and check that it behaves. You describe what the agent must do. WeHarness turns that into an agent architecture, helps build it in the coding tools you already use or in its own cloud sandboxes, and checks the result with tests made for your project.
WeHarness does not replace your editor, your repository or your coding assistant. It works alongside them.
Who it is for
Section titled “Who it is for”- Developers who add an agent to an application: tool calls, approvals, retries, memory, evaluations and tracing.
- Teams who want a person to approve each plan and each push, and evidence for what the agent does.
How it works
Section titled “How it works”- Describe the project. Say what the agent should do, for whom, and which limits apply. For example: “refund only after a support lead approves”.
- Get a plan. WeHarness proposes how the agent should be built: its tools, the steps that need a person’s approval, the data it may use, and how its behaviour is checked. A person approves the plan in the web app.
- Build it. A WeHarness session builds the plan in a draft, runs your code in an isolated cloud sandbox and pushes a pull request you approve. Or your coding tool builds it over MCP and WeHarness verifies what it sends.
- Check the behaviour. Traces from your running agent, scenario cases, evaluators and lessons show what passed, what failed and why.
Where you work
Section titled “Where you work”| Surface | What you do there |
|---|---|
| The web app, app.weharness.io | Sessions, plans, approvals, pushes, Tracing, Monitoring, Evaluations, Lessons, settings. Every approval happens here. |
| Your coding tool, over MCP | Claude Code, Codex, Cursor, VS Code, Devin or another MCP client uses WeHarness while it works. See MCP overview. |
| Your application | The tracing SDKs or plain OpenTelemetry send your agent’s traces. See Set up tracing. |
Available in the pilot
Section titled “Available in the pilot”- Accounts and workspaces. Access is by invitation. Four roles. See Getting access and Workspaces and roles.
- Sessions. One natural conversation per project: WeHarness answers, reads your code and files, plans, and builds in a draft you review. See Sessions.
- GitHub and uploads. Connect repositories through the WeHarness GitHub App, or upload a folder or zip. See GitHub and Project context.
- Plans and approval. Requirements, acceptance tests and candidate designs; code checks which are eligible, the Elephant-1 decision model compares them, and a person approves. See Plans and approval.
- Cloud work. Commands, clean checks, repairs and previews in isolated sandboxes; approved pushes as pull requests. See Cloud work.
- Build tools. Tracing, Monitoring, Evaluations, Online evaluators, Automations, Lessons, Code search and Improve this agent.
- MCP server. Use WeHarness from your coding tool. See MCP overview.
- Tracing SDKs.
weharness-tracingon PyPI and@weharness/tracingon npm. See SDK reference. - Credits and your data. Work is counted in credits; export a workspace; delete a project with proof. See Credits and usage, Retention and export and Deleting a project.
Not available yet
Section titled “Not available yet”- Public sign-up and payments. During the pilot WeHarness invites people and adds credits; nothing is bought in the app.
- Verified support for each coding tool. The setup steps follow each tool’s own documentation; WeHarness has not yet tested each tool end to end. See MCP overview.
- Claude on the web and ChatGPT as MCP hosts. See Claude web and ChatGPT.
- Work across several repositories in one session.
- Deleting a whole workspace in the app (by request until then).
Get access
Section titled “Get access”Request a demo or email info@weharness.io. See Getting access.