Module · Assistants

Every answer backed. Every task done.

Assistants answer from your data, back every statement with its source and respect your permissions. And they don't stop at answering: they search the Index, generate documents and trigger workflows — in Teams, on the web, via API, even on the phone.

5 minutes
until a new assistant is ready — configured, not coded
24/7
on duty — in Teams, on the web, via API and on the phone
1 request
is enough: research, analyze, generate the document — in one pass
assistant · procurement
Summarize all open POs > €50k delayed by supplier.

14 POs across 6 suppliers. Top 3:

POSupplierValueDays
PO-44218Bosch Rexroth€ 184,20018d
PO-44102Voestalpine€ 92,40012d
PO-43987Festo AG€ 71,9009d

Source: SAP MM · live

Ask anything across your data…
Evidence, not claims

Every statement links to its source — one click opens the original at the right passage.

An assistant you can't trust isn't one. That's why Syntact assistants answer exclusively from your indexed data — and every statement links to the document, page, table or record behind it. If the answer isn't in your data, the assistant says exactly that.

  • Answers only from your data — no guessing from world knowledge
  • Sources down to the page and passage — one click opens the original
  • No match? The assistant says so — instead of hallucinating
  • Structured questions included: sums, filters and analyses over real tables
  • Replies in the user's language — or the one you set
assistant · contracts
What notice period applies to the TechParts framework agreement?

6 months to the end of the quarter. The 2024 amendment shortened it from 12 months.

Framework_TechParts.pdf · p. 14 · §9.2
Amendment_2024-03.pdf · p. 2
Source
Framework_TechParts.pdf — page 14
One click from answer to passage
Assistants that act

From answering to getting it done.

Most chatbots stop where the work begins. Syntact assistants work in multiple steps: they research the Index, query tables, generate Word, Excel and PowerPoint documents and trigger Flows — and you see every step they take along the way.

  • Multi-step requests: research, analyze, create — in one pass
  • Generates finished Office documents: reports, quotes, presentations
  • Triggers workflows in Flows — escalations or data runs, for example
  • Operates systems without an API too — via browser automation
  • Every step visible: what was searched, found and created
assistant · operations
Create a slide deck on all delayed orders and notify purchasing.
Searched index
12 delayed orders found
Queried tables
values & delay days aggregated
Generated report
Delays_Q3.pptx · 8 slides
Triggered flow
Escalation → purchasing team
Delays_Q3.pptxready to download
Built in minutes

One assistant per area of work. Configured, not coded.

One assistant for purchasing, one for sales, one for the project team — each with its own instructions, its own knowledge scope and its own tools. Assembled in the UI, without a line of code.

  • Instructions, knowledge scope, tools and model — set per assistant
  • Enable tools selectively: search, tables, documents, Flows, web search
  • Connect your own tools — internal services become assistant skills
  • Choose any model, including EU-hosted — and switch anytime, no rebuild
  • As many assistants as you need — all on the same data layer
settings · assistants · new
Procurement Assistant
answers in the user's language
active
Instructions
You support the purchasing team. Always cite sources. Escalate unresolved cases…
Knowledge
Index · Procurement
Model
Configurable · EU-hosted
Tools
SearchTable queriesDocumentsFlowsWeb searchCustom tools
Channels

Integrated into the tools you already use.

One assistant, many doors: your employees ask in Teams or the browser, your customers on your website, your systems via API — all talking to the same assistant on the same data.

Web & embedding

In the Syntact portal or as a chat widget on your website — publicly accessible if you want.

Microsoft Teams

As a bot right inside Teams — with single sign-on and answers in the chat thread.

Phone

Call and talk: the assistant understands and answers in real time — with a selectable voice.

REST API

Use assistants programmatically — with streaming responses for your own applications.

MCP

External AI agents and tools access your assistant skills — permissions included.

Slack, WhatsApp & email

More channels on the same foundation — ask us about the current status.

Whatever the channel: same data, same permissions, same assistant.

Governance

Six features that keep your assistants under control.

Who may use which assistant, what was it allowed to see, and what did it do? Assistants has built-in answers to these questions — from the permission model to the conversation log.

The user's permissions

The assistant sees exactly what the asking user is allowed to see — inherited from your identity provider, checked on every query.

Complete audit trail

Every conversation, every search, every tool call — stored, attributed to a user and inspectable. No black box.

Guardrails for tools

Critical parameters of MCP tools — the signed-in user, for example — are pinned from context instead of set by the model. What the model can't set, it can't get wrong.

Roles & approvals

Roles control who may use, build or manage assistants — centrally and per tenant.

Escalation to humans

When the assistant gets stuck, it hands over to your team or your ticketing system — with the full conversation context.

Model sovereignty

You decide which models are available — per tenant, including EU-hosted options. No vendor lock-in.

FAQ

Questions about Assistants.

How do I know the assistant isn't hallucinating?
Assistants answer exclusively from your indexed data, and every statement links to its source — document, page, passage. If the assistant finds no basis, it says so instead of inventing an answer. You can verify any answer against the original with one click.
Can the assistant expose data a user isn't allowed to see?
No. Every query runs with the asking user's permissions — inherited from your identity provider and checked before the answer is formed. What a user can't see, their assistant can't see either.
Which language models are behind it — and where do they run?
That's your call. Assistants is model-agnostic: per assistant, you pick from the models you approve — from leading commercial models to EU-hosted or self-operated ones. Switching models is a setting, not a project.
How long does it take to build an assistant?
Minutes, not weeks: write the instructions, assign the knowledge scope, enable the tools — done. The Index provides the data foundation; new assistants use it immediately.
What happens when the assistant can't answer a question?
It says so — and escalates to your team or your ticketing system if you want, with the full conversation context. Your users never hit a dead end.
Can assistants do more than chat?
Yes. They generate Word, Excel and PowerPoint documents, trigger Flows, run structured queries on tables and operate systems without an API via browser automation. Every step stays visible.

Ask our assistant your hardest question.

30 minutes, your documents, live: you ask, the assistant answers — with sources you can verify on the spot.