AI in the Absidion platform

AI experts who know your processes.

Policies, clauses, templates, precedent — we make sure the Absidion platform hands every agent the domain knowledge of your organisation.

No data to third parties. No training on customer data. Frontier models in our Swiss data centre.

Request a live demo
  • Data stays local
  • No training
  • Frontier models
  • Platform-wide
  • With your know-how
Absidion · AI assistance running as m.keller@scholarships.zh.ch
Write a reply… Send ↵
The four prerequisites

How AI carries administrative processes.

Four prerequisites for AI to actually take effect in an administrative procedure — embedded rather than bolted on the side.

  1. 01

    Data protection

    Personal data must not reach third parties. Publicly hosted language models are not an option in an administrative context.

  2. 02

    Consistency

    Only embedded, problem-specific tooling produces consistent inputs and outputs.

  3. 03

    Case context

    All relevant data of the current case must be aggregated, prepared and provided to the AI — in the right cut.

  4. 04

    Domain expertise

    Generic models know language, not procedures. For AI to work reliably in your organisation it must know your policies, clauses and precedent.

Our approach

We meet all four prerequisites, in one system.

The AI sits where the process runs. It knows the application, the case and the policy — and the entire processing happens on our infrastructure.

  • Sovereign
    Data does not leave our Swiss data centres, no transfer to third countries.
  • No training
    Customer data does not flow into model training — neither ours nor anyone else's.
  • Embedded
    AI agents are integrated directly into the specialised processes, not placed beside them.
  • Frontier
    Best available models, operated on our own infrastructure, not at a hyperscaler.
Domain knowledge

The AI supports you inside your specialised processes.

Frontier language models are universal but not in charge. Absidion hands every agent exactly the legal basis, policies and precedents that apply to the current case.

Curated by your subject-matter experts, maintained inside the platform, fed in in real time — no separate training, no model hosting at third parties.

Source-bound Every answer references a policy, a clause or a record
Knowledge base · live 6 sources · active case
Legal basis
CO · FADP · IVöB · Subsidies Act · cantonal law
ON
Organisational policies
BVW §12 · Procurement directive v2026.1 · 47 internal directives
ON
Clause & template library
184 standard clauses · 27 application templates
ON
Case history
51 closed cases in the same class, anonymised
ON
Master data & glossary
Persons, organisations, domain terminology DE / FR
ON
Validation rules
612 domain rules, curated by experts
ON
AI agent · active case
Replies with your know-how, in the context of this file
Security at process level

Same permissions, same frame — for agents too.

Sovereign infrastructure is one half. The other happens in every single action — whether a user invokes the AI from within a case or an agent runs in the background: the same permission and context rules apply.

01 · Permission fidelity

Identity-bound access

Every action runs under a known identity: for interactive requests, the signed-in user; for background agents, an explicit service account with narrowly defined rights. In both cases the same permission checks apply as for a click in the front end. What the identity may not open, the AI cannot read, cite or summarise either.

Request as · m.keller@scholarships.zh.ch ACL filter
Application A-2026-0481
Scholarships · own responsibility
ACCESS
Application A-2026-0312
Research · different department
DENIED
File 2024-0177
HR file · not authorised
DENIED
Receipt booking-conf.pdf
in application A-2026-0481
ACCESS
No bypass, no service accounts with full access, no "admin view" for the AI — not even for autonomous agents.
02 · Context scoping

Task-bound context

The context an agent sees is cut to its current task — the application, the case, the meeting, or, for background routines, the explicitly assigned data set. Data from parallel procedures, other tenants or unrelated files does not enter, even if the identity would in principle be allowed to open it.

CONTEXT · APPLICATION A-2026-0481
Application data
24 fields · structured
Attachments
3 documents · proposal, CV, reference
History
7 comments · 4 status changes
OUT OF SCOPE
Application A-2026-0479
same tenant
Case F-2025-1102
own responsibility
File 2024-0177
archived
Need-to-know at record level, not just at role level.
Capabilities

Three layers of impact, one system.

  • Background agents
    Data analysis and validation run in the background through AI agents, flagging anomalies before they become problems.
  • Embedded agents
    Interactive expert agents for evaluating and editing data, directly in the form a user is currently working in.
  • Platform-wide
    AI capabilities are available across every functional area — one architecture, one security model, one interaction paradigm.
How we work with you

From use case to productive agent — with you.

Rolling out AI is not a project you tackle alone. Five phases, clear roles, verifiable outputs.

  1. 01

    Use-case workshop

    Together with your experts we identify where AI has impact — and where it doesn't.

    Who
    • Your domain lead
    • our AI specialist
    Duration
    ½ day
  2. 02

    Knowledge ingestion

    We curate policies, templates and clauses into the knowledge base.

    Who
    • Your subject experts
    • our solution engineer
    Duration
    2 – 4 weeks
  3. 03

    Pilot

    The first agent goes into production within a clearly bounded process.

    Who
    • Your pilot users
    • our engineering team
    Duration
    4 – 8 weeks
  4. 04

    Training & handover

    Your experts learn to maintain the knowledge base on their own.

    Who
    • Your power users
    • our AI trainer
    Duration
    2 days
  5. 05

    Scale-up & operations

    More procedures, continuous improvement, SLA-grade operations.

    Who
    • Your domain & IT lead
    • our customer success team
    Duration
    ongoing
Example · case management

AI agents for all stakeholders.

Applicants are guided through the process; case workers are relieved — on the same application, with a consistent context.

FOR APPLICANTS
Application with guidance
  1. 1
    Application capture
    The AI guides applicants through the process in a conversational way, picks the right application type and collects all relevant data.
  2. 2
    Fill-in assistance
    Applicants supply texts and documents in free form; the AI populates the structured fields.
  3. 3
    AI-driven pre-check
    The application is checked against the policy. The pre-check points out issues before submission.
FOR CASE WORKERS
Processing with depth
  1. 1
    Application agent
    Ask the application questions via chat and trigger actions, e.g. "Draft a correction request".
  2. 2
    Policy check
    Compare against the policy, flag shortcomings, provide a draft correction request.
  3. 3
    Document alignment
    Check the files in the application for content and consistency with the structured data.
  4. 4
    Translation
    Applications in other languages are translated — bidirectionally, for both application and correspondence.
APPLICATION · CONVERSATIONAL Application for research funding
Write a reply… Send ↵
Quellen Research funding policy 2026 · §3 para. 2 · Application template F-417
CASE WORK · APPLICATION #A-2026-0481 Check travel expense claim
Write a reply… Send ↵
Kontext Application A-2026-0481 · 12 documents · role: scholarship case worker
Quellen Travel expense policy 2026 · BVW §8 · Booking receipt booking-confirmation.pdf
Example · contract management

Every clause, checked against your contract directive.

The expert agent knows the legal basis, organisational policies and contract type, checks every clause, identifies issues and suggests concrete improvements.

AI CONTEXT
What the agent knows about this contract
Contract type
Services agreement
  • Template SVC · v3.2
  • Language EN · primary
  • Term 24 months, with renewal
Legal basis
CO · FADP · public procurement
  • CO Art. 394 ff. Mandate law
  • FADP (revFADP) Art. 9, 19, 21
  • IVöB Thresholds
Organisational policies
Procurement & contract directive
  • Liability Cap at order volume
  • Data processing CH hosting, DPA required
  • Termination max. 6 months' notice
Services agreement · V-2026-0143
Draft v0.3 · 14 clauses · AI review active
1 Critical 2 Notice 11 OK
§ 4
Fees
The contractor invoices monthly on a time-and-materials basis at CHF 185.– per hour (excl. VAT). Expenses are itemised separately.
§ 7
Liability Critical
The contractor is liable without limit for any damage arising in connection with the performance of the services, including lost profits and consequential damage.
§ 9
Data processing Notice
Personal data is processed by the contractor for the performance of the contract and may be passed on within the group for improvement purposes.
§ 12
Termination Notice
The contract may be terminated by either party for the end of a calendar year with twelve (12) months' notice.
AI analysis · § 7 · Liability Critical
ISSUE

Unlimited liability contradicts the organisation's procurement and contract directive (liability cap at order volume) and is not required under mandate law.

PROPOSAL · NEW WORDING § 7
The contractor's aggregate liability for damages out of or in connection with this contract is limited to the fees effectively paid in the twelve (12) months prior to the damaging event. Liability for lost profits and indirect consequential damage is excluded. Liability for intent and gross negligence remains reserved.
Audit trail is being recorded
Example · form designer

Build forms in dialogue.

The expert agent knows your platform's field definitions, validation rules and translation glossaries and edits forms via chat: add fields, reorder, validate, translate. Domain experts shape their procedures themselves, without a detour through IT.

  • Add or reorder fields by description
  • Express validation rules in plain language
  • Generate multilingual versions automatically
FORM DESIGNER · CHAT EDIT Housing allowance application
Write a reply… Send ↵
First conversation

Let's talk about your first use case.

30 minutes, no sales pitch. We listen to what you have in mind and tell you honestly whether — and where — AI can help.

Schedule a use-case workshop
  • Swiss location
  • No sales pressure
  • NDA on request
Use-case workshop 30 min · online or on site
  1. 01
    You describe process & bottleneck
    In your own words, no preparation needed.
  2. 02
    We mirror suitable use cases
    Drawn from our experience in administrative and funding projects.
  3. 03
    You receive an assessment
    What is worth doing, what isn't, and what first.