AI HR platform Hire on evidence.Prove the rest on the job.
Agents screen candidates, onboard new starters, build training on any topic and draft performance reviews — all from your own documents, and a person approves every decision.
- Single sign-on
- Your data isolated from other companies
- Audit log the app can't edit
- A person signs off decisions about people
- You set how far AI agents go
A finished training is not a skill.
People explain it in their own words, and an AI examiner grades the answer against a rubric written from the same sources as the training. A pass writes their skill level and can close an onboarding gate.
- Content and rubric come from the same run and the same source, so the question cannot drift from what was taught.
- Where the skill is mapped to a job role, the new level is compared with the level that role requires — not with a completion percentage.
- A failed attempt gets a re-explanation of what was missed, then a retry. After three attempts the check unlocks and is flagged for follow-up.
First you teach it your company. Then it does the work.
Admins upload the handbook, the policies, the playbooks and the job descriptions, and choose who each document is for. Every AI agent reads from this one brain, scoped to your tenant.
- Answers cite the passages they came from, and only from documents the asker could open themselves. The database enforces it, not the prompt.
- Health checks flag coverage gaps and look-alike documents that could give people two different answers.
- Connectors for SharePoint and Teams are in early access.
AI agents that do real work, inside limits you set.
Each agent has one job and reads from the same company brain. The AI assistant can only do what the asking person could do, and an autonomy dial per area decides which of its actions run without asking.
Meet the AI agentsAI assistant
Answers from the company brain with sources, and acts — assigning programs, starting onboarding — with the asking person's own permissions.
Drafting agents
Draft training, onboarding templates and manager reviews from your documents and records. Nothing publishes until a person approves it.
Context agent
Turns uploaded documents into the company brain, visible only to the people each is for.
The examiner
Grades Understanding Checks and role-play conversations against their rubric. A pass writes the skill level.
Insights agent
Finds people below their job role's required level and overdue work, and points to programs that close the gap.
Assignment engine
Assigns programs by job role, department and hire date, and escalates overdue work.
Every requirement marked evidenced, claimed or absent — never scored.
Load your own pile of CVs and search it in plain language. Each candidate is measured against a real job role's requirements, and every mark carries the sentence from the document that earned it.
- No match percentage. “Five years of Java” is not a measurement, and a number you cannot defend when challenged is worse than three words you can.
- Quotes are checked in code against the document, so a sentence a model invented cannot reach a recruiter as something the candidate wrote.
- Photographs, dates of birth, marital status and state of origin are stripped before any model reads the CV — including CVs that arrive as phone photos.
- There is no endpoint that rejects anyone. Not a setting you could switch on: the route does not exist, so a person makes every call.
A 30/60/90-day plan for every hire, with gates that close only on proof.
An AI agent drafts one onboarding template per job role from your documents, with a reason for every gate. A person reviews and publishes it, and each new hire gets their own copy, dated from their hire date.
- Gates close only on strong evidence: a manager sign-off, a reviewed piece of work, an Understanding Check or a passed certification conversation.
- Importing hires from a CSV starts their plans automatically.
- When someone changes role, the skills they have already proven are skipped, not taught again.
Training generated from your own documents, then checked for understanding.
Point the drafting agent at a policy, a playbook or a product guide and it writes the program in typed blocks — text, slides, audio narration, quizzes, scored scenarios and role-plays — with the Understanding Check from the same run. A critique pass checks every block against its sources before anyone reviews it.
- Blocks are typed, so one regenerates without rebuilding the program, and a person approves each block before the program publishes.
- Assignment rules on job role, department and hire date set due dates and escalate work that goes overdue.
Performance reviews drafted from evidence, held for a human.
Employees nominate their reviewers and their manager approves the list. An AI agent drafts the manager's review from the person's skills and training record — and then stops. Nothing reaches the employee until the manager has approved every section.
- An AI agent flags feedback that may cross your review guidelines. It never rejects anything: the manager reads every response and decides.
- Performance can never be set to autopilot, and the employee can write a response to their published review.
You set how far the agents go.
One dial per area, set by your admin, starting at the most conservative level. It decides which of the AI assistant's actions run without asking. Drafts never publish on their own — a person approves them at every level.
The level every area starts at on a new tenant. Move the dial per area when your own experience says the agents have earned it.
Acts with your permissions
The AI assistant uses the asking person's own token, so it can only do what they could.
Held above 10 people
Assigning or starting onboarding for more than 10 people waits for a person, at every level.
Append-only trail
Each action names the accountable person and any agent involved. The app cannot edit or delete a row.
What a pilot is, and what is not in place yet.
What a pilot includes
- One job role, and the documents that define it.
- The company brain built from them, scoped to your organization.
- Onboarding and training drafts, each block showing its source passages.
- Understanding Checks graded against their rubric.
- The autonomy dial at Proposes only; nothing publishes without a person.
- Scope and hosting agreed before any employee data is loaded.
Not in place yet
- SOC 2 or ISO 27001 certification.
- SAML or SCIM. Single sign-on is OpenID Connect only.
- A fixed hosting region.
What is in place, and how to check it: the security page.
Bring one job role. We'll run the loop on it.
Thirty minutes, your own documents, one job role. You'll see the brain built, the drafts it produces, the sources behind them, and exactly where the dial sits before anything publishes.