Goodwill Project · Open Career Record
person.work — maintain your CV in conversation, not in a form
person.work is built for experienced freelancers and IT consultants who already use Claude, ChatGPT, Codex, or a local agent and have an existing CV. Instead of maintaining the same resume across documents, forms, and PDFs, they connect their own agent through MCP. The agent structures a change, person.work makes it reviewable, and only the owner decides what counts. The confirmed Career Record produces the web profile, PDF, and open data from one source of facts.
What was built
Three deliberately separate layers: interpretation by the chosen agent, control with the owner, and reliable outputs from one record:
The owner's agent. Claude, ChatGPT, Codex, or a local agent reads the existing CV, asks about ambiguities, and sends only structured facts through MCP. During MCP sync the raw CV stays with the selected agent; person.work does not run a second model pipeline that reinterprets the content and may reduce its quality.
The control boundary — person.work. OAuth 2.1 limits what an agent may read, initially sync, or propose as a change. New records stay private. Ongoing changes appear as reviewable diffs; an agent can neither approve nor publish its own proposals.
The shared source — the Open Career Record. Confirmed roles, projects, and dates are stored once and rendered as a web profile, PDF, and JSON. The underlying format is documented on openworkid.org as an experimental, MIT-licensed draft, so the record remains portable and does not depend on person.work.
The hard decisions
The quality of interpretation should not depend on a small model embedded in the product. It therefore happens in the agent the user has already chosen and supplied with context. person.work then validates the structure deterministically instead of rewriting the content with another model.
Agents may maintain a record, but they may not decide what counts. Separate OAuth permissions, private initial syncs, and mandatory owner approval for ongoing proposals turn human control into a system property rather than a disclaimer.
person.work is deliberately a non-commercial goodwill and public-commons project by Entwicklerherz GmbH: no plans, no paid visibility, and no sale of profile data. The cold-start phase tests whether people voluntarily use this Career Record, not how early it can become a business model.
Why this is here
person.work demonstrates the Entwicklerherz approach in a production system of our own: use models where interpretation creates value; keep identity, permissions, validation, and approval deterministic; build open interfaces in from the start. That is how an AI feature becomes an accountable agent system in real operation.
Agent-maintained · owner-controlled · portable
See it yourself
Want to integrate agents without handing control and domain logic to the model? Let's talk.
Send an inquiry