Product
How the platform works
The knowledge loop and one consistent execution path — from a question or task to an answer or an action under human control.
Slimaco routes every question and every agent task through the same path: from access control, through retrieving approved company knowledge and using tools, to a result or an action approved by a human wherever that's required.
The knowledge and work loop
Slimaco works as a loop, not a one-off. Every conversation, email and agent result can become knowledge the whole team benefits from — after human approval.
- Sources — documents, mailboxes, calendars and the team's conversations with the Slimaco assistant.
- Knowledge mining — picking up agreements, decisions and changes in how work is done from day-to-day communication.
- Review — the area owner approves, edits or rejects knowledge candidates.
- Approved knowledge — versions, an owner and access rules.
- Assistant and agents — answers with sources and tasks run on demand or on a schedule.
- Actions — work results in the company's apps, which flow back into the loop as new sources.
The question or task path — 7 steps
- Receiving the question or task — text, attachments, an assistant reference and a source scope, or a scheduled agent run.
- Access control — the user's permissions, plan, sources and available tools are checked.
- Choosing the execution mode — the question can be handled in standard mode, by an assistant, by a multi-assistant plan, by a skill, or by a tool.
- Retrieving knowledge — the system searches for evidence within the approved and permitted scope.
- Using tools — if the task requires it, the model can read data from a connected system or propose an action.
- Human consent — a write operation is held pending a human decision: on every run, or according to consent granted for a set period.
- Result — an answer with citations or an agent's work result, recorded and passed to the chosen people or systems.
The same mechanism underlies the product layers described separately in this knowledge base: agents and automations (how work gets done), organizational knowledge (what an answer is based on), the Slimaco assistant (how an answer is built) and integrations and actions (what can happen beyond the conversation).
The fail-closed principle
If the platform cannot confirm a plan, permission, scope, safety check or the availability of a key control mechanism, it should refuse or limit execution rather than default to granting access. This principle applies at every one of the seven steps above.
The building blocks processes are made of
Repeatable procedures and automations in Slimaco are built from the same set of blocks — whether the task is handling a ticket, preparing a document or analyzing data:
| Block category | Role |
|---|---|
| ingest | Receiving and normalizing a document, message or event. |
| classify | Recognizing the type, intent, priority or handling path. |
| retrieve knowledge | Retrieving knowledge from the approved scope. |
| analyze | Analyzing material and producing a structured result. |
| extract facts | Extracting facts into a controlled structure. |
| tool call | Calling an allowed tool through the MCP layer. |
| decision | Branching the process based on a result or policy. |
| review | Pausing the process for a human decision. |
| generate artifact | Creating a draft, report, recommendation or analysis card. |
| notify or handoff | Notification, or a controlled handoff to an external system. |
The Processes module is used to review repeatable ways of working, candidates detected in the organization's material, versions and activation. Approving a candidate creates a process draft but does not run it automatically — activation is a separate decision.