The platform in practice
Organizational knowledge
Sources, knowledge mining, versions, approval and access — the foundation the Slimaco assistant and agents work on.
A source is the logical place knowledge comes from. A document belongs to a source and can have further versions. Uploading a file does not by itself mean its content becomes approved knowledge available in answers.
Where knowledge comes from
Knowledge enters Slimaco in three ways. The first is sources and documents added by the organization: manually or from OneDrive and Google Drive. The second is knowledge mining, which analyzes day-to-day communication (email, Slack, Microsoft Teams and the team's conversations with the Slimaco assistant) and data from department systems, and picks up agreements, decisions and changes in how work is done. Captured fragments carry their source and go to review as knowledge candidates. They join shared knowledge only after the area owner approves them.
The third is hard data from department systems: CRM, ERP, helpdesk and invoicing. Figures and statuses, such as the number of a customer's open tickets or ongoing work with them, flow into Slimaco continuously, without review. A salesperson preparing a quote sees that the customer is waiting on service, and service knows the same customer is negotiating a contract.
Approved knowledge is the basis for both the assistant's answers and agents' work. Changing an approved procedure immediately changes how the automations that use it behave.
Document states
| State | Meaning |
|---|---|
| Uploaded | The file has been received and saved, but doesn't have to be approved or indexed yet. |
| Normalized | The content has been cleaned up and prepared for further processing. |
| Candidate | The system has created knowledge candidates that need review. |
| Approved | The content has been accepted according to the organization's process. |
| Indexed | Approved content is available in the search index for authorized users. |
| Withdrawn / purged | Withdrawal is reversible; a purge is permanent deletion. |
The approved-knowledge rule
Uploaded doesn't mean approved, and approved doesn't yet mean indexed. Separating these states limits the risk that an accidentally added, outdated or unauthorized document starts influencing answers.
Access to knowledge
Access rules can be applied at the collection, source and document level. The system computes a user's effective access based on role, permissions and resource rules. The interface lets you check who can see a source and why. Access to the admin screen and access to specific content are two separate levels of decision.
Search and evidence
- A query is interpreted in the context of the user, the active assistant and the chosen source scope.
- Retrieval works only on the approved knowledge index available to that user.
- An answer can include citations and information about where the evidence came from.
- The system should say knowledge is missing when sources don't support an answer (a "no-answer").
- Document titles and content are treated as potentially sensitive and are not written to logs as plain metadata.
Knowledge vs. conversation memory
The knowledge base and conversation memory are not the same thing. The knowledge base is a managed organizational resource. The history of a single thread helps maintain conversational context, but another conversation is not automatically treated as a source of truth or evidence for the current answer. Memory across different conversation threads is not currently a product feature.