Parallel Agents
Codebolt lets you run multiple agents at the same time — on the same project, on different tasks, or coordinated in a pipeline. This section covers the patterns for doing that and the tools for monitoring and managing concurrent runs.
Why run agents in parallel
Running agents sequentially — one finishes, then the next starts — works fine for simple linear tasks. Parallel execution becomes valuable when:
- Tasks are independent — writing tests while docs are being generated; refactoring two separate modules simultaneously
- You want to compare approaches — run the same prompt against two different agents or models side by side
- A parent agent delegates subtasks — an orchestrator breaks a large task into pieces and hands each to a specialist agent
- Background work shouldn't block foreground work — a scheduled audit runs while you continue working interactively
Patterns
| Pattern | How it works | When to use |
|---|---|---|
| Tabs | Open multiple chat tabs; each runs its own agent independently | Ad-hoc parallel work, comparing outputs |
| Child agents | An agent spawns sub-agents via codebolt_agent.start; results feed back | Programmatic delegation from agent code |
| Flows | A declared graph of agents connected by data edges | Repeatable pipelines with defined structure |
| Swarms | A dynamic group of cooperating agents sharing memory and tasks | Large, exploratory, or open-ended work |
In this section
- Running Parallel Agents — how to start and manage concurrent runs
- Chat Canvas — visual workspace for monitoring and coordinating parallel agent conversations
- Background Agent — how the Background Agents panel starts, tracks, and surfaces background runs
- Sub Agent — how delegated child runs work and where to follow them