Skip to main content

​
Overview

Apollo Brain is the central AI assistant — a conversational agent powered by Claude that can read and modify your entire workspace. It’s not a simple chatbot; it has 160+ native tools organized into 18 categories with a skill-based routing system.

​
How to Access

Apollo Brain is available from any page via a floating button in the bottom-right corner. Click it to open the Brain drawer with two tabs: Chat and Voice.

​
Chat Interface

The chat tab provides a full agentic experience:
  • Streaming responses — see the AI’s thinking in real-time via SSE (Server-Sent Events)
  • Tool execution — watch as Brain calls tools to query databases, create tasks, analyze code
  • Context panels — side panels show the tasks/projects/resources being referenced
  • Human-in-the-Loop (HITL) — for any write operation (create, update, delete), Brain asks for your confirmation before executing
  • Session persistence — conversations are saved and can be resumed
  • Auto-generated titles — sessions get meaningful titles based on conversation content

​
Agent Architecture

​
Skill-Based Routing

Every message is matched against 4 specialized skills:
SkillDomainExample Queries
project_managementTasks, projects, releases, sprints, epics, team”List all overdue tasks in Project Alpha”
code_devopsGitHub, deployments, infrastructure, CI/CD”Show me recent commits on the main branch”
personalNotes, personal tasks, goals, reminders”Create a reminder for my 1:1 tomorrow at 3pm”
generalistEverything else — analysis, advice, general knowledge”How should I structure this API?”

​
Memory System

Apollo Brain uses a 3-tier memory system:
  1. Session Memory — context within the current conversation
  2. Episodic Memory — past conversation summaries stored in agent_sessions
  3. Procedural Memory — learned patterns and user preferences in agent_procedural_memory

​
Context Preloading

On each message, Brain hydrates a DuckDB warm cache with your workspace data and builds a markdown snapshot of your current state (active projects, recent tasks, team members) that’s injected into the system prompt. This gives Brain instant awareness of your workspace without needing to make queries.

​
Tool Categories

list_tasks, create_task, update_task, get_task_details, search_tasks, create_tasks_bulk, bulk_assign_tasks, bulk_update_task_status, bulk_update_estimated_hours, create_task_dependency, delete_task_dependency, add_task_comment, list_task_comments
list_projects, get_project_status, export_project_data
list_releases, create_release, get_release_details, update_release, assign_tasks_to_release
Notes: search_personal_notes, list_personal_notes, get_personal_note, create_personal_note, update_personal_note, list_personal_foldersTasks: list_personal_tasks, create_personal_task, update_personal_task, delete_personal_taskGoals: list_personal_goals, create_personal_goal, update_personal_goal_progressReminders: list_personal_reminders, create_personal_reminder, complete_personal_reminder, snooze_personal_reminder, update_personal_reminder, dismiss_personal_reminder, delete_personal_reminder
Read: list_connected_repos, get_task_commits, get_task_prs, search_commits, get_repo_activityWrite: analyze_pr_changes, suggest_reviewers, check_merge_readiness, create_pr_review, create_branch, merge_pull_request
ECS: list_ecs_clusters, list_ecs_services, get_ecs_service_details, force_deploy_ecs_service, scale_ecs_service, get_ecs_service_logsLambda: list_lambda_functions, invoke_lambda_function, get_lambda_logsECR: list_ecr_repositories, get_ecr_scan_resultsCI/CD: list_github_workflows, trigger_github_workflowTerraform: list_terraform_states, detect_terraform_driftCosts: get_infrastructure_costs, get_cost_forecast, get_infrastructure_overviewDeployments: list_deployments, get_amplify_build_status, trigger_amplify_deploy
list_conversations, get_conversation_messages, search_chat_messages, send_chat_message, get_unread_summary, list_notifications, mark_notifications_read, get_user_presence, add_message_reaction, mark_conversation_read
search_knowledge_base, get_kb_page, list_kb_pages, list_kb_folders, create_kb_page, update_kb_page, delete_kb_page
Team: list_team_members, get_user_by_nameTime Tracking: list_time_entries, log_time, get_active_timer, start_timer, stop_timer, get_time_summaryWorkload: get_workload_summary, get_time_off, get_weekly_goals, analyze_workload_balanceCode Intelligence: diagnose_error, suggest_fix, get_recent_changes, analyze_repository_structure, read_file_contentsPSE: get_pse_session_status, start_spec_generation, answer_pse_clarification

​
Human-in-the-Loop (HITL)

For safety, all write operations (create, update, delete) require user confirmation before execution. When Brain wants to perform a mutation:
  1. Brain shows you what it wants to do (e.g., “I’d like to create a task with these details…”)
  2. A confirmation panel appears with Approve and Reject buttons
  3. You review the action and decide
  4. Only after approval does the tool execute
Read operations (list, search, get) execute automatically without confirmation.