Every model output in rinkats is a read for a person, never a decision. Then the same board is open to your own agents, so the work you would hand a junior can be handed to a tool.
Each grounded in the candidate's own resume and your requirement, each shown with its reasoning.
Headline, summary points, skills with months, the timeline of roles. It waits until you accept it, section by section, or keep your own.
A one-line read, strengths with evidence, the career arc, watch-outs, a stability read and verification flags. Flags, never verdicts.
A score with the reasoning shown. Loud on purpose, because it is the one thing that should pull your eye. It never moves a card.
Questions from the candidate's history and the role, a place to record answers, and an evaluation of the recorded responses when you ask.
The lines are drawn in the product, not in a policy document.
No candidate is dropped by a model. A terminal move is a person's decision, with a reason, and it is recorded.
The model's words stay inside your workspace. Invites go out only when a person, or an agent acting as one, schedules a meeting.
Candidates, resumes and AI spend belong to one organisation. Nothing learned about your candidates helps anyone else.
How screening works, as a picture: the model sweeps every seat on the board against the requirement and lights the one worth a call. It does not clear the others.
A fit score, a summary, a set of questions. Each one shows its reasoning and each one can be ignored. Nothing a model writes moves a card.
Bring your own provider keys. A recruiter can carry a personal key, the organisation can hold shared ones, and usage is visible per key.
Any AI feature can be switched off for the whole organisation. The board, the candidates, the interviews and the fees keep working without it.
Your assistant already reaches your calendar, your mail and your documents. Over MCP it reaches the board too, and one request can span all of them. Every change it makes carries the name of the person whose key it holds.
Reads and writes, as you
Account, then API keys. A key is scoped to you and your organisation; revoke it any time.
One block of configuration, the same for every MCP client.
The agent discovers the tools on its own. A move it makes is a move on the board, recorded as yours.
Scripts read the same API the product uses. Other agents find the workspace through its agent card.
Jobs, applications, candidates, requirements, pipelines, interviews, placements and invoices, documented as OpenAPI 3.1 with a live console, with API keys or OAuth 2.1.
API referenceThe capability surface as tools, one server for each side, with the same permissions as the person whose key is used.
Connect an agentA discoverable card so other agents can find the workspace and talk to it agent to agent.
Agent protocolsEvery statistic on this page links to its source. Read at the source on 15 Sept 2026; tell us if one has moved.
Tell us about your desk and we will set your workspace up with you, with your own roles on the board from the first day.