Skip to content

Platform

From a source file to work you can sign off.

Marella brings your approved documents into one place, answers questions from them with the source attached, and keeps agent work inside a boundary you set.

Source
Contract register.pdf · cl. 12.1
Answer
Six of 84 supplier contracts deviate from standard terms.
Boundary
Your cloud · not used to train models
Platform atlas

The route from a source file to reviewable work

  1. 01Bring in
    Approved documents

    Upload files, sync Confluence, or connect another source once we have checked it

  2. 02Make legible
    Media and passages

    Supported files are read into passages, and you can see where each one came from

  3. 03Make useful
    Knowledge structure

    The people, companies and topics it finds, in a structure your team can correct

  4. 04Ask
    Cited answers

    Drawn only from the documents in scope, with the source open beside them

  5. 05Repeat
    Workflows with limits

    Agents set up for one job, with the tools they are allowed to use written down

Architecture

Five layers, each with its own failure mode.

A parsing failure and a permissions failure both reach the reader as a bad answer, so test each layer on its own terms.

  1. 01

    Document intelligence

    See what happened to each file: what was processed, which passages were parsed and where each one came from.

    See document processing
  2. 02

    Enterprise search

    Only the documents you approved, and each claim in an answer stays linked to the passage behind it.

    See cited search
  3. 03

    Knowledge structure

    Curate the people, companies and topics Marella has extracted, when the workflow is worth that effort.

    See the knowledge network
  4. 04

    AI agents

    Give repeatable work a bounded agent: its sources, its tools and what it may do without asking.

    See agents with limits
  5. 05

    Control boundary

    Decide who can reach what, where it runs and who signs off, across all four of the other layers.

    See security and trust

The workbench and the control room

The person doing the work and the person accountable need different things.

The workbench

Ask, and check the answer yourself.

Search a scoped set of documents, open the passages behind an answer, and use a task-shaped agent where the workflow calls for one.

  1. QuestionApproved documents
  2. AnswerCitations + validation
  3. KnowledgeEntities + concepts
  4. OutputTemplate + review
The control room

Scope it, then own the outcome.

Administrators set the organisation context, which sources are in scope, which model path runs and what an agent may call. Connector behaviour is then tested against the architecture you proposed.

  1. IdentityOrganisation scope
  2. InferenceConfigured provider
  3. ActionsTool permissions
  4. EvidenceDecision traces

The platform test

Start with one bounded piece of work.

A platform tour only tells you something when it runs a real question over representative documents, with the expected evidence agreed and a named reviewer at the end.

  1. 01Name the job

    For example: answer a policy question, compare a clause or assemble a supported brief.

  2. 02Name the sources

    Include the stale, conflicting and awkward material, not just the clean examples.

  3. 03Name the boundary

    Write down where documents sit, which model path runs and who is responsible for what.

  4. 04Name the decision

    Agree what the system may propose and what an authorised person must review or approve.

Platform questions

Plain answers before a demo.

What is the Marella AI platform?

Marella is a private platform for enterprise knowledge work. It processes your documents, answers questions from them with the source attached, keeps the knowledge your team curates, and runs bounded agents over the same material.

Is Marella only an enterprise search product?

No. Search is one workflow. The same documents and controls also support curated knowledge, configured agents and output templates where those are enabled.

Does Marella replace source systems such as Confluence?

No. Your source systems stay authoritative. Marella works over approved copies, and before a connector goes live we test what it does with permissions and what happens when a page is updated or deleted.

Can Marella be deployed in a private environment?

Yes. Managed, your own cloud and self-hosted are all supported. We write down where your data is stored, which models run over it and who supports it, for the configuration you actually pick.

Test it on one real workflow.

Start with non-confidential characteristics. We agree the evaluation and transfer route before material moves.

Plan an evaluation