Release note · knowledge-network
Knowledge network: a map of your own documents
Marella now maintains a curated map of the people, companies, documents and topics inside your corpus, browsable and correctable by your own team.
Release notes
What is new, what has changed, and what still depends on your version and setup.
Release note · knowledge-network
Marella now maintains a curated map of the people, companies, documents and topics inside your corpus, browsable and correctable by your own team.
Release 03
Marella now maintains a curated map of the people, companies, documents and topics inside your corpus, browsable and correctable by your own team.
What shipped
Marella extracts the people, companies and topics from your documents and links them together, so your team can browse the map and correct it.
Entities are the people, organisations, documents and topics Marella finds in your files, each showing the concepts it is linked to. Concepts are the ideas that come up across your documents, linked wherever they appear. The ontology is your taxonomy, curated by you, so the system uses your vocabulary rather than a generic one. You can review possible duplicates and merge them.
None of the extraction is assumed correct. Before anyone relies on it, check on your own documents: are the entities right, do the links hold up, are duplicates handled, and does access stay inside its boundary?
What you get depends on your version, how it is deployed and which connectors are set up, so test the workflow you actually need.
Read the full noteRelease 02
Configure an agent for one task, give it reusable skills and a fixed output format, and set exactly which documents it may read. Managed centrally.
What shipped
Teams ask different questions of different documents. Marella lets you set that up per team.
Agents are assistants configured for one job, such as a contract reviewer for legal or a policy assistant for operations, and each answers only from the document sets it is scoped to. Skills are the capabilities agents share, so improving one improves every agent that uses it. Templates fix the shape of the output for repeatable work: summaries, clause tables, briefs in your house format.
Administrators configure all three centrally, and agent activity lands in the audit trail.
What you get depends on your version, how it is deployed and which connectors are set up, so test the workflow you actually need.
Read the full noteRelease 01
A cited-answer surface, plus media-library controls for inspecting the documents that were indexed and the source passages behind a given answer.
What shipped
Ask questions in plain English and Marella answers from the documents you have put in scope. Answers link to the passages behind them. How precisely a link jumps to the spot depends on the file, and it is still the reviewer who decides whether the passage supports the claim.
The media library shows indexing and processing states for supported document types, including the files that failed, so you can see them.
The loop is: ask, open the source, record whether it holds up.
What you get depends on your version, how it is deployed and which connectors are set up, so test the workflow you actually need.
Read the full noteTry it
Try it on your own documents, with your own permissions, and on the cases you expect to go wrong.
Plan an evaluation