Skip to main content
Second Brain uses a team of five specialized AI agents that collaborate on every task. You never call an individual agent directly — you describe what you need, and the Manager handles the rest, deciding which agents to engage and in what order.

The agent team

Manager

Coordinator and decision-maker. Every task starts here.

Planner

Breaks complex tasks into concrete, grounded steps.

Researcher

Deep-dives on a topic and produces structured findings.

Builder

Writes and edits vault content — notes, folders, and patches.

Critic

Quality gate. Reviews Builder output before the task closes.

Agent roles in detail

Manager

The Manager reads your prompt and decides which agent should act next. It is the only agent that sees every step of the pipeline — it receives results from each agent and uses them to determine the next move. When the task is complete, the Manager signals the pipeline to close. The Manager’s behavior is tuned by the AI mode you select before sending your prompt. Switching modes changes the Manager’s instruction set, which cascades through every agent involved in that task.

Planner

The Planner breaks down complex requests into a concrete, ordered list of steps. Before producing a plan, it reads your current vault structure so its output is grounded in what actually exists — folder names, note paths, existing links — rather than assumptions. For tasks that reference a specific note by name, the Planner also receives that note’s outgoing links, backlinks, and tags so it can plan with full relational context.

Researcher

The Researcher performs a deep investigation on a topic using the context you provide (your prompt, attached files, or pasted text). It produces structured findings that the Builder can turn directly into vault notes. The Researcher is engaged for tasks that require gathering, synthesizing, or verifying information before anything is written.

Builder

The Builder writes and edits your vault. It is the only agent that makes changes to files on disk. It uses a set of structured directives to signal what to create or modify:
DirectiveWhat it does
# vault_file: <path>Creates or overwrites a markdown note
# vault_folder: <path>Creates a folder
# vault_asset: <path>Creates a non-markdown file (CSV, JSON, etc.)
# patch_vault_file: <path>Applies a surgical patch to an existing note
# delete_vault_file: <path>Deletes a note from the vault
Each directive is processed by Second Brain, which saves the output to your vault and updates the index immediately after.

Critic

The Critic reviews the Builder’s output for accuracy, completeness, and formatting before the task closes. If the work meets the standard, the Critic signals CRITIC_APPROVED and the pipeline ends. If it finds issues, it sends detailed feedback back through the Manager to the Builder for revision. The Critic is a quality gate, not a final editor — it does not modify files itself.

Watching the pipeline

The Tasks popover in the chat input bar shows the live status of all five agents during a task — each transitions from idle to working to done as the pipeline progresses. You can follow the full agent conversation in the chat area as each agent completes its step.