Skip to main content
These terms get used loosely in conversation, but they mean different things architecturally. This page exists to keep them straight.

​
The short version

​
Tools

Tools are the actual actions SYMI can take: reading and writing files, querying a connected service, creating a document, running code. What tools are available in a given session is determined by that session’s configuration — not by anything in a prompt or a file. A workspace file can tell SYMI how to approach a task, but it can’t grant SYMI a new tool it wasn’t already given.

​
Skills

Skills are specialized instructions for a particular kind of task — for example, how to properly build a spreadsheet, or how to work with a particular file format. Rather than loading every skill’s full instructions into every conversation, SYMI keeps a lightweight catalog of what skills are available and what they’re for, and only reads the full instructions for a skill when it’s actually relevant to what you’ve asked. This is why SYMI can draw on a large and growing library of specialized know-how without every single conversation carrying the weight of all of it.
You don’t need to invoke a skill by name. SYMI selects a relevant skill based on what you’re asking it to do.

​
Sub-Agents

Sub-agents are specialized agents you deploy within SYMI for a specific, ongoing purpose. Each sub-agent has:
  • A label — a name identifying its purpose (e.g. “RFP Responder,” “SOC Triage Agent”)
  • A task prompt — instructions defining what the sub-agent does and how it should behave
  • A model routing assignment — which model the sub-agent queries
Sub-agents are the right place to define role-specific context — for example, “You are a cybersecurity analyst working in a SOC. When given an alert, identify the affected system, classify the severity, and suggest a remediation step.” This is different from a Persona, which controls tone only, not role context. To create a sub-agent:
  1. Navigate to Sub-Agents.
  2. Select + New Sub-Agent.
  3. Enter a label, a task prompt, and the model to route to.
  4. Save and activate.
Sub-agents run with a narrower scope than your main conversation — they’re built to focus on the task they’re configured for, rather than carrying the full breadth of general capabilities.

​
How they relate to the Scheduler

Scheduled (cron-style) jobs run as their own session type, distinct from your regular conversation. A scheduled job can invoke a sub-agent or a direct prompt on a recurring cadence, without needing an inbound trigger. See the Scheduler section for configuring recurring jobs.

​
Next steps

Persistent Workspace

See how sub-agent and task output is stored.

Security & Isolation

Understand what governs tool access, regardless of what a prompt or skill requests.