# Inspect Swarm > Run a swarm of agents on one sample of an Inspect task. ## Basics - [Getting Started](https://meridianlabs-ai.github.io/inspect_swarm/index.html.md): Install Inspect Swarm, run a swarm as the solver of an existing task or in a task of your own, and see what the log records; read this first. - [Members](https://meridianlabs-ai.github.io/inspect_swarm/members.html.md): Define the swarm's members with `member()`: factory or built agents, names, roles, counts, limits, member builders, tool state, background work, how a member ends, and the agents the swarm refuses; read it to set up who runs in a swarm. - [Controllers](https://meridianlabs-ai.github.io/inspect_swarm/controllers.html.md): The `leaderless` controller, when the swarm stops, which submission is its answer, and how to write a controller of your own; read it to change how members start and stop or how the answer is chosen. ## Communication - [Channels](https://meridianlabs-ai.github.io/inspect_swarm/channels.html.md): How members share information: the default `filesystem` channel, what the log records of files members share, and how to write a channel; read it to choose how members communicate. - [Shared Memory](https://meridianlabs-ai.github.io/inspect_swarm/shared-memory.html.md): The `shared_memory` channel: wiring its tools into members, the tool's commands, notices, `swarm_wait`, subagents, limits, the record of the board, monitoring with Inspect Sentinel, and migrating from `memory(instance=...)`; read it when members need a shared board of files that the log records. ## Resources - [Budgets](https://meridianlabs-ai.github.io/inspect_swarm/budget.html.md): The swarm's cost, token and time caps with `Budget`: defaults from the sample's limits, cost data, pooled or split caps, what members see of the budget, and sweeps; read it to cap or divide what a swarm spends. - [Sandboxes](https://meridianlabs-ai.github.io/inspect_swarm/sandboxes.html.md): Shared or per-member sandboxes: members in the sample's sandbox, a sandbox for each member with shared services or a shared volume, sizing, and scoring; read it when members run code or edit files. ## Results - [Scoring](https://meridianlabs-ai.github.io/inspect_swarm/scoring.html.md): How the task's scorers score a swarm: which submission is the answer, the `inspect_swarm:result` record, scoring what members leave in sandboxes, checks during the run, and comparing a swarm with a single agent; read it to set up scoring or read the answer. - [Analysis](https://meridianlabs-ai.github.io/inspect_swarm/analysis.html.md): The `inspect_swarm:metrics` and `inspect_swarm:ledger` store entries and the swarm columns for Inspect's `samples_df()`; read it to analyse a swarm's usage, cost, communication and waits after a run. - [Restrictions](https://meridianlabs-ai.github.io/inspect_swarm/restrictions.html.md): What the swarm does not support, the limits of bridged members, what the log does not record, and the store keys task code must not write; read it to check whether a setup is supported and what the log holds as evidence. ## Reference - [inspect_swarm.analysis](https://meridianlabs-ai.github.io/inspect_swarm/reference/inspect_swarm.analysis.html.md): Columns for samples_df() and the models of the swarm ledger. - [Configuration](https://meridianlabs-ai.github.io/inspect_swarm/reference/inspect_swarm.configuration.html.md): The member record and the swarm's budget. - [inspect_swarm](https://meridianlabs-ai.github.io/inspect_swarm/reference/inspect_swarm.html.md): Swarms, members, controllers, channels and the budget view.