Get your agent team running
Install HiveTerm, define your bees in a hive.yml, and start orchestrating. Everything here works on macOS, Windows and Linux.
Quickstart
Install
Download the app for your OS and open it, that's the whole install.
Add a project
Open a project folder. HiveTerm auto-detects your stack and offers a starter hive.yml.
Start
Open the project, every auto-start bee boots and coordinates through the Queen.
Routines
A routine is a task on a schedule, sent into an agent you already have. It lands in that bee's live conversation. If the agent is stopped, HiveTerm starts it. If it's busy, HiveTerm interrupts and injects.
Your first routine
Open Routines
Hit ⌘⇧R, or the + row under Routines in the project tree. You need a project open, and at least one agent in it.
Pick the agent and write the task
Choose a bee from the sidebar, then write what it should do each run. "Summarize what landed on main since yesterday" is enough to start.
Pick when
Daily, weekdays, weekly, or every few hours, in your timezone. Missed slots can run when the app opens, or skip to the next time.
Run now, or wait
Save it and let the schedule fire, or hit Run now. The task shows up in that agent's tab. You can watch, reply, or walk away.
Templates to start from
Morning briefing What changed since yesterday: commits, PRs, issues, and what needs you. Issue triage Read new issues, label them, flag the ones that block you. Dependency audit Check outdated and vulnerable packages, write down what to bump. Weekly changelog Draft release notes from the week's commits and merged PRs. Test health Run the suite, report failures, point at the first file to open. Branch cleanup List stale branches and say which ones are safe to delete. Routines live on the project. Pause, edit, duplicate to another project, or delete from the sidebar. An agent can create and manage them through Hive, the same actions you see in the UI. Free includes 2; Pro is unlimited.
The hive.yml
One file per project defines your agents and processes. Commit it so your whole team shares the same workspace.
name: My App
bees:
claude:
type: agent
command: claude
instructions: "Senior engineer. Ship clean PRs."
api:
command: npm run dev
cwd: ./server
auto_start: true
restart_on_change: ["src/**"]
env:
PORT: 5173| Field | What it does |
|---|---|
name | Project name shown in the sidebar. |
command | The shell command to run for this bee. Required. |
type | "agent" for AI CLIs, "command" for everything else. |
cwd | Working directory for the bee (defaults to the project root). |
env | Environment variables injected when the bee spawns. |
auto_start | Boot this bee automatically when the project opens. |
auto_restart | Restart the bee when it exits with a non-zero code. |
restart_on_change | Restart the bee when files matching these globs change. |
instructions | System-prompt instructions injected into an agent at startup. |
The Queen (MCP server)
A local MCP server on an automatically chosen port. Agents call it to spawn helpers, ask each other, run routines, read output, and notify you. Isolated per project.
create_agent Add an agent to the sidebar; temporary:true for a helper that lives under Sub-agents list_agents See every agent, command and terminal in the project, with status get_agent_status Check health and uptime of one agent start_agent Start a stopped agent; a saved session resumes stop_agent Stop an agent; it stays in the sidebar, ready to start again interrupt_agent Cut the current turn short (Esc) without stopping the agent restart_agent Stop and start again with the same configuration delete_agent Remove an agent from the sidebar and hive.yml for good set_name Give itself a name in the sidebar, or rename a teammate set_avatar Put a face on its row from any image it makes or picks set_instructions Set its own system prompt, or a teammate's, as text or a file ask_agent Ask another agent in its tab and wait for the reply here write_input Type into a running agent or command read_output Read the terminal output of any agent or command send_message Send a structured message to another agent read_inbox Receive messages from other agents, no scraping reply Answer a message, correlated to its thread await_reply Block until another agent answers notify Push native desktop notifications run_command Run a dev server, build or script under Commands, with its own terminal stop_command Stop a running command; it stays listed create_routine Schedule a task on a live agent list_routines See routines in this project update_routine Pause, rename, or change the schedule delete_routine Remove a routine and its run history run_routine Fire it now into that agent's tab How agents talk
ask_agent is the one you see: from CEO, ask CTO, the question appears in CTO's tab, the answer comes back here. Under that, send_message / read_inbox / reply / await_reply still exist for structured hand-offs that never scrape a terminal. Prefer ask_agent when you want the other agent to actually work the question in its session.
Creating agents
create_agent makes a lasting agent in the sidebar, saved to hive.yml; pass temporary:true for a helper that lives under Sub-agents and goes away with its task. Both take two opt-in flags: model pins the agent to a specific model (omit it to let the orchestrator decide), and skip_permissions starts it with its permission prompts bypassed, each CLI gets its own flag injected (Claude, Codex and Antigravity all supported). Dev servers, builds and scripts go through run_command instead, under Commands. Tools that target another agent take agent_id (to_agent_id on send_message). read_output additionally accepts encoding:"screen" to replay the terminal as a rendered VT100 screen instead of raw bytes.
Splits & files
Split groups pair your agents side by side, create one from the sidebar (two ungrouped agents required), drag panes by their grip to reposition or swap, and save layouts with a name. Selecting any member brings the whole split back.
The file explorer opens anything: code gets syntax highlighting with VSCode-style search and in-place editing (⌘S, atomic saves), huge files switch to a virtualized viewer with text search, and images, markdown and binaries each get a proper view.
Need more?
Check the changelog for the latest, or just download and explore, most of HiveTerm is discoverable in the app.