Skip to content
HeartBeatAgents

Built-In Tools

View as Markdown

Every capability an agent has before you connect a single integration: code execution, live services, memory, skills, delegation, and more.

Every agent has a fixed set of built-in tools available from the moment it's created, independent of any integration you connect. Integrations add more (see Integrations Guide); these are the ones that are already there.

Execution and code

  • code_executor: run Python in an isolated environment with access to your shared folders.
  • shell: run a shell command, with Node.js, npm, yarn, pnpm, Python, pip, git, curl, and full Linux tools available.
  • git_push: push the current workspace to a GitHub repository, creating it first if it doesn't exist.

Files

  • read_file / write_file: read or create/overwrite a file in a shared folder or the agent's own workspace.
  • list_directory: list a directory's contents in shared folders or the workspace.
  • list_shared_folders: list which folders from your machine are shared with this agent, and whether each is read-write or read-only.

Live services and preview

  • start_service / stop_service / list_services: start, stop, and list running dev servers and long-running processes, each with a live preview URL. See Live Preview.
  • update_project: update a project's own metadata: its live URL, tech stack, status, or linked GitHub repository.
  • destroy_project: completely and irreversibly tear a project down: stop its services, remove its files, clean up its resources.
  • present_choices: present 2 to 4 visual options side by side for you to pick between.

Memory

  • remember / recall / forget: explicit control over an agent's own long-term memory, alongside the automatic extraction described in Memory.

Skills

  • create_skill: turn a proven solution into a reusable skill for every future agent.
  • update_skill: repair, enhance, or roll back an existing skill.
  • search_skills: check whether a skill already exists before building one from scratch. See Skills.

Integrations and the web

  • integration_api_request: make an authenticated request to any connected integration, with credentials injected automatically.
  • list_connected_integrations: list what's actually connected and bound to this agent.
  • http_request: a general-purpose HTTP request to any external API or website, connected or not.
  • web_search: search the web for current information.

Monitors

  • create_standing_order / update_standing_order: create or change a scheduled, recurring task.
  • list_standing_orders / manage_standing_order: list, pause, resume, or delete one. See Monitors.

Collaboration and control flow

  • delegate / find_agents: hand a subtask to another agent and find who's available to help. See Delegation.
  • invite_to_call / pass_the_floor: bring a teammate agent onto an active voice call, or hand the current speaking turn to a teammate already on it. See Voice.
  • wait_for: durably pause the current run until a timer fires or a condition is met, without holding a worker the whole time.
  • ask_user: ask you a direct question and wait for your answer before continuing.
  • query_trace: check exactly which tool calls actually happened this run, so a final summary reports what was really done rather than what the agent assumes it did.
  • calculator: evaluate a mathematical expression safely.