Dashboards you can talk to, and Claude in one click

01Dashboards you can talk to#

Everything Hivemind does produces data, and reading it back used to mean opening one screen at a time and holding the comparison in your head. Dashboards put it on boards instead.

Six core boards ship with every account and are hydrated live whenever you open one, so there is nothing to configure and nothing to keep current: overview, pipelines, candidates, interviews, assessments and credits. Change something on a core board and your version forks into a saved board rather than overwriting the built-in, so the original is always there to go back to.

Boards you build yourself sit on a twelve-column grid. Cards drag and resize in both directions and pack together, so a board cannot end up with holes in it. A pipeline picker and a time range sit above every board, and titles follow the range so a board never claims to show thirty days while displaying ninety.

The part worth trying first is Hive, the chat dock beside your boards. Describe the board you want and it builds it. Ask for a change while a board is open and the change lands on that board in place, with nothing to dismiss and no copy to reconcile afterwards. Hive knows which board you are looking at, so follow-ups can be short.

02Connect your AI assistant by signing in#

Connecting Hivemind to Claude used to mean generating an API key, copying it into a config file, and hoping nobody pasted it anywhere it should not go. That key was also company-wide, so revoking it broke every other client using it.

Now you add Hivemind as a connector, sign in, and approve a consent screen that names the app asking for access. There is nothing to copy. The consent screen shows a verified badge for apps Hivemind can vouch for, which is how you tell the real Claude from something that merely registered itself under the name.

Connections are per person and listed under your settings, so you can see what is connected and revoke any of it immediately, without disturbing a colleague or rotating a shared key.

The server itself grew too. It now exposes 38 tools, and they are no longer read-only: alongside reading pipelines, candidates and assessment results, an assistant can draft pipelines, assessments and outreach campaigns. Those drafts are staged, not live, and a separate commit step promotes them, so nothing an assistant produces goes out on its own.

1. 1In your AI assistant, add a custom connector pointing at the Hivemind MCP server.
2. 2Sign in to Hivemind and approve the consent screen.
3. 3Review or revoke connected apps under Settings → Apps & Integrations.

03A behavioural read on every candidate#

Assessments tell you whether someone can do the work. They do not tell you how that person tends to work: how directly they push, how much they lean on other people, how they handle pace and change, how closely they stick to detail. DISC covers that second question.

The candidate answers 28 forced-choice blocks, picking what is most and least like them, and the profile is scored the instant they submit. Scoring is arithmetic over their own answers with no model in the path, so the same answers always produce the same profile. The scoring key never reaches the candidate, so the test cannot be reverse engineered from the page.

You get a style code such as DC with a name that goes with it, scores across the four dimensions, and an intensity reading that tells you whether the style is pronounced or the candidate sits between two. Every profile also carries quality flags, so an incomplete or rushed submission says so instead of being read as a result.

One thing worth knowing before you use the numbers: they are shares of the candidate's own profile and add up to 100 across the four dimensions. They are not percentiles against other people, and two candidates' scores on the same dimension cannot be ranked against each other.

04Interview review in one place#

After an interview the evidence used to scatter: a recording in one system, the interviewer's notes in a doc, and a transcript nobody went back to. Reviewing meant reassembling it.

Now the recording, the notes and the transcript stack under the interview step on the candidate's record, read top to bottom the way an interview actually happened. Click any moment in the transcript and the recording jumps there and plays, which is the quickest way to check a quote you are not sure you read right. Audio-only interviews and phone calls behave the same way.

Take the transcript with you as a PDF for sharing or as plain text for search. Access follows the candidate record: a teammate who cannot open the candidate cannot pull their transcript. On a shared report link, which anyone holding the URL can open, the recording and notes are shown and the verbatim transcript is withheld.

05Search from the job description you already have#

Sourcing starts from a job description somebody already wrote. Turning it into a search used to mean reading it and summarizing it back into a sentence, which is both work and a chance to drop the requirement that mattered.

People search now takes the description itself. Switch the composer to job description mode, paste it in full, and Hivemind reads the requirements out of it while ignoring the parts that are not about the person: the culture blurb, the benefits, the boilerplate.

You still get the same review step before anything runs, so a description that asks for too much is something you can trim before you spend a credit on it. Whichever mode you used last is the one waiting for you next time.

06Assessment results over the API#

Assessment performance lived in the app, which was fine until someone wanted it in a board, a warehouse or a weekly report. Getting it out meant exporting screens by hand.

Six read-only endpoints now cover assessments, questions and their results, under the same base path, key and pagination as the rest of the public API. Assessments reach candidates two ways in Hivemind, sent on their own and run inside a pipeline, and the result endpoints report both through one resource with a field telling you which kind each row came from.

1. 1Generate an API key under Settings → Apps & Integrations.
2. 2Call the assessments, questions and results endpoints under the public API v1 base path.
3. 3Check the source field on a result before joining it to anything else.