Ask the company brain: an AI assistant that answers from your documents and gets things done
Ascend's AI assistant answers employees' questions from the company's own policies, SOPs and handbooks with the source passage attached, tells them about their own training and onboarding, and carries out HR tasks for managers and admins within their permissions and the limits the company sets.
New people learn by asking, and every answer shows where it came from.
Most questions a new hire has are already answered somewhere in a handbook or an SOP. The assistant finds the passage and answers from it. It is told to use only what it retrieved from your documents, never its general knowledge, and to cite which passage backs each claim. The sources sit under the answer, so anyone can open the passage and check.
- Answers come only from passages the person asking is allowed to see. The database filters by department, access role and identity before anything is searched, so a document scoped to Finance never shapes an answer for someone in Support.
- Follow-up questions work. "What about for contractors?" is searched together with the questions before it, so it still finds the leave policy.
- Conversations are saved per person and private to them. Anyone can reopen an earlier session or delete it.
- When the documents do not answer the question, the assistant says so rather than guessing, and the reply is labelled "Not found in the knowledge base".
Ask about your training, your onboarding and where you could go next.
The same box answers questions about the person asking, from live data rather than documents. Each of these is a read: it runs straight away and changes nothing.
- "What training have I been assigned?" lists their assignments and due dates.
- "How is my onboarding going?" reads their own onboarding plan.
- "Where could I go from here?" lists the career paths open from their current job role.
- "What notifications do I have?" reads their notifications, so nothing waits unseen in a bell icon.
The starting prompts on the assistant page are worked out from the tools each person's role can actually use, so nobody is offered a question that would come back refused.
It does the work, within your permissions and the company's limits.
For managers and administrators the assistant is also an AI agent that acts. It has 42 tools, 18 of which change something: assigning and publishing training, setting due dates, creating assignment rules, starting and archiving onboarding plans, creating job roles and skills, and updating an employee's job role or manager.
- It acts as the person asking. Every call carries their own sign-in and goes through the same permission checks as the screens, so it cannot do anything they could not do themselves.
- Each person is offered only the tools their role allows. An employee's assistant has no tool to assign training to anyone.
- Before a change, it shows what will happen in plain words, such as which program goes to which people, and waits for a yes. How often it waits is set per area by an administrator on the autonomy dial.
- At the default setting it only proposes. At assisted, changes that can be undone run on their own and the rest wait. Anything that touches more than 10 people waits for confirmation at every setting.
- It cannot approve, sign off or attest to anything on anyone's behalf. It can find the item, and the person decides.
- Every change it makes is written to the audit log with the accountable person and a mark that an agent carried it out.
Prompts people actually see on the assistant page, and what happens next.
- “What's our policy on parental leave?”Searches the documents the asker may see and answers with the passages it used. If none answer it, the reply says so and the question goes to the escalation queue.
- “What training have I been assigned?”Lists the asker's own assignments and due dates. Reads run straight away.
- “Who is overdue on their training?”For a manager, lists overdue assignments across their own reports; for an admin, across the organization.
- “Which sign-offs are waiting on me?”Lists the onboarding gates waiting on the asker. It does not sign them; the person opens each one and decides.
- “What skills do we track?”Lists the skills already in the catalogue, the check to make before asking it to create one. Creating a skill cannot be undone, so that waits for confirmation unless an admin has set training to autopilot.
Not sure means a person looks, not a confident guess.
Every answer the assistant could not ground in your documents is added to an escalation queue for administrators. The employee still gets the honest reply straight away; the queue is how someone makes sure the question gets a real answer.
- An administrator sees the question, the reply and the conversation around it, and marks it resolved with a note.
- If the answer belongs in the company brain, the administrator can write it up as an FAQ and choose who may see it. The next person who asks something similar gets a sourced answer instead of another escalation.
| Capability | Intranet search | Generic AI chatbot | Ascend's AI assistant |
|---|---|---|---|
| Where answers come from | A list of files matching the keywords. | The model's general training, which knows nothing about your policies. | Passages from your own documents, retrieved for each question. |
| Cites sources | Links to whole documents. | Usually not, or not checkably. | Yes, the passages behind each answer. |
| Respects permissions | Folder permissions, which copies and broad search can step around. | Only sees what someone pastes in, and nothing stops them pasting. | Enforced on every passage by the database, per person. |
| Knows your own work | No. | No. | Your training, onboarding plan, career paths and notifications. |
| Can take actions | No. | No access to your HR records. | Yes, as the person asking and within the autonomy dial. |
| When unsure | Shows loosely related files, or nothing. | Often answers anyway. | Says it could not find an answer and sends the question to administrators. |
A support engineer's third day.
What buyers ask about the AI assistant.
- Does it make things up?
- It is instructed to answer only from passages retrieved from your documents and to cite them, and to say plainly when they do not answer the question. No model is perfect, which is why the sources are shown for checking and why every answer it could not ground goes to administrators.
- Can employees see documents they shouldn't?
- Not through the assistant. Each passage carries the audience set on its document, and the database filters by the asking person before any search runs.
- Can it change things on its own?
- Only as far as your administrators allow. By default it proposes and a person confirms. Administrators can let it run reversible changes, or more, per area; anything touching more than 10 people always waits, and performance reviews, succession and offboarding can never be put on autopilot.
- Is our data used to train models?
- The language model calls behind the assistant go to Anthropic. Our security page sets out where data goes and what is and is not in place today.
What the assistant draws on.
Bring one job role. We'll run the loop on it.
Thirty minutes, your own documents, one job role. You'll see the company brain built, the drafts it produces, the sources behind them, and where the autonomy dial sits before anything publishes.