Skip to main content

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

PatternHow it worksWhen to use
TabsOpen multiple chat tabs; each runs its own agent independentlyAd-hoc parallel work, comparing outputs
Child agentsAn agent spawns sub-agents via codebolt_agent.start; results feed backProgrammatic delegation from agent code
FlowsA declared graph of agents connected by data edgesRepeatable pipelines with defined structure
SwarmsA dynamic group of cooperating agents sharing memory and tasksLarge, 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