Structured Sessions
Structured Sessions are SwarmClaw's bounded collaboration and execution runs. They are designed for work that should start, progress through a defined flow, produce outputs, and conclude cleanly.
They are a core feature, not a separate engine you need to enable.
What They Are
Structured Sessions give you:
- a reusable template or built-in starter
- one bounded run instance
- optional linked context from a chat, chatroom, task, schedule, or project workflow
- durable run state, outputs, and operator actions
- an optional hidden temporary live room you can watch while the run is active
How They Relate to Other Surfaces
| Surface | Best For |
|---|---|
| Direct chat | open-ended help with one agent |
| Chatroom | persistent free-form multi-agent conversation |
| Structured Session | temporary bounded runs with outputs and completion |
| Task | concrete execution leaf work |
| Project | durable operating context and grouped work |
Structured Sessions sit between open-ended conversation and low-level task execution:
- more structured than a normal chat or chatroom
- lighter than pushing every workflow into bespoke task trees
- able to emit summaries, artifacts, and follow-up tasks back into surrounding context
Starting a Structured Session
You can start one from:
- the Sessions page
- a direct chat header
- a chatroom header
- a chatroom
/breakoutcommand - a task
- a schedule configured to launch a structured run
The normal path is contextual: start from the work you are already doing and let SwarmClaw prefill the linked context.
Live Rooms
When transcript creation is enabled, a Structured Session can create a temporary hidden live room.
That room:
- is a real chatroom-backed transcript surface
- stays out of the normal persistent chatroom list
- can be opened from the session detail or parent context
- archives automatically when the run concludes
Templates and Runs
- Templates are reusable definitions for how a run should behave.
- Runs are individual executions of those templates.
Built-in starters cover neutral patterns such as:
- independent collection
- facilitated discussion
- review panel
- decision round
- status roundup
- single-agent structured runs
Custom templates can define step-based flows with branching, repeats, joins, and richer operator controls on the same runtime engine.
Outputs
Structured Sessions can produce:
- summaries
- artifacts
- emitted tasks
- transcript history
- grounded citations when knowledge retrieval is involved
That makes them useful for review, planning, coding, QA, research, decision rounds, and general multi-agent workflows without hard-baking one domain model into the product.