The platform in practice
Assistants and skills
Specialized AI roles with their own knowledge and tools, and the repeatable instructions they can apply.
An assistant is a configured AI role that belongs to the organization — with its own name, instructions, persona, knowledge scope, model choice, tool list and visibility rules. An assistant is not a separate model: it uses the platform, models and tools within the bounds of its configuration.
When an assistant is triggered
| Situation | Behavior |
|---|---|
| The user types @Name | The assistant is explicitly selected. Its knowledge, persona and tools narrow execution. |
| A question without @ | The coordinator can match zero, one or several assistants based on when_to_use and build a plan. |
| No good match | An empty plan is a valid result. The Slimaco assistant answers in standard mode instead of forcing a match. |
| A multi-step task | A plan can route further steps to different assistants and assemble a combined answer. |
| Delegating mid-task | The assistant.delegate tool can hand off a limited subtask to another assistant — it is separately controlled and disabled by default. |
The when_to_use field is a routing hint, not a guarantee of activation. It helps the coordinator tell assistants apart, and the current logic can decline to pick one if no description fits — this protects against choosing the closest-sounding role for a question outside its scope.
What a skill is
A skill is a repeatable instruction or runbook an assistant can apply. It is a configuration, not a program and not a separate agent. It's built from a name, a description of the procedure, a snippet defining when to use it, and a restriction on the available tools. A skill can narrow the tool profile, but it never extends a user's or assistant's permissions.
| Assistant | Skill |
|---|---|
| Defines a role, persona, knowledge and tools | Defines a repeatable way to execute a specific task |
| Can be selected via @ or matched by the coordinator | Can be matched to a task within the active context |
| Can take part in one or several steps of a plan | Provides the instruction for a controlled tool-use loop |
| Has its own visibility and configuration | Can be private, role-based or company-wide depending on permissions |
A safe read of multi-agent behavior
Multi-agent behavior in Slimaco means coordinating specialized assistant configurations and bounded work steps — it does not mean an uncontrolled group of autonomous bots. Every step is still subject to knowledge scope, tools, model, limits and guardrails. An empty plan is a valid answer, and write actions stay behind the consent gate.
Status
| Area | Status | Scope |
|---|---|---|
| Assistants | Available | Creating and managing specialized assistants, their persona, scope and visibility. |
| Skills | Conditional | Creating procedures is available on the relevant plans; company-wide publishing has a separate plan threshold and permission. |