Open source Apache 2.0 · one static Go binary · no runtime

Your codebase's journey, narrated by agents

Orchestrate AI agents across issues, profiles, and machines. Full visibility from dispatch to PR.

brew install vnovick/tap/itervox

or the desktop app for macOS, Windows and Linux →

The Itervox dashboard on a phone: status bar, what needs you, status chips The Itervox dashboard on a phone: status bar, what needs you, status chips
The Itervox dashboard: what needs you, live status chips, capacity and the fleet overview The Itervox dashboard: what needs you, live status chips, capacity and the fleet overview
Works with LinearGitHub IssuesMarkdown issues in your repo Claude CodeCodexOpenCodeCursor CLI
01 — HOW IT WORKS

You already have the agents. Now orchestrate them.

Running coding agents by hand, one issue at a time, doesn't scale. Itervox turns them into a fleet.

1Tracker

Issues appear in your tracker

Itervox polls Linear, GitHub Issues, or Markdown files under .itervox/issues/, and claims the work that is ready.

ENG-148 Paginate the audit log
ENG-149 Audit log pages in admin
2Agents

Agents spawn in parallel, one per issue

Each agent runs in its own git worktree, with retries, stall detection and live logs. Locally, or on SSH worker hosts.

ENG-148 claude Running
ENG-145 codex @ build-1 Running
3PRs

PRs opened, issues transitioned

When the work is done the PR link lands on the issue and its state moves on. You review and merge — that's it.

ENG-142 pull/87 In Review
ENG-136 merged Done
# in your repository
$ itervox quickstart
  detects tracker + agent CLI, writes WORKFLOW.md,
  runs itervox doctor, starts the daemon
$ itervox demo   # or try a throwaway board first
09:41:35 INFO worker: starting ENG-142 backend=claude
09:41:36 INFO dispatch complete dispatched=2 running=3
09:44:12 INFO worker: PR link commented pr_url=pull/87
09:44:12 INFO ENG-142 transitioned to="Done"
02 — ORCHESTRATION

Built for autonomous agents at scale

Autonomous, not unsupervised. Agents do the work; you stay in control at every checkpoint.

Concurrent agents

Run up to N agents in parallel with per-state concurrency limits. Scale from 1 to 50+ without config changes.

Retry queue

Failed runs retry with exponential backoff — 10s, 20s, 40s, capped at 5 minutes. Stalled workers are killed and retried.

Input required

Agents can ask you a question. It is posted to the tracker; reply from Linear, GitHub or the dashboard and the run resumes.

Pause & resume

Free a slot now; resume later and the agent continues the same session from exactly where it stopped.

Automations

Trigger helper runs on cron, new comments, state changes, failed runs, opened or merged PRs, rate limits and resolved blockers.

Dependency-aware dispatch

Tracker blockers gate dispatch, inferred edges soft-gate it, and a dependent can stack on its blocker's branch while that one is in review.

Scoped daemon actions

Profiles opt into comment, review, create issue, move state or provide input — a short-lived per-run grant, not blanket access.

Configure without restarts

Concurrency, profiles, tracker states and more are live-editable from the dashboard and persist to your settings files.

Scale across machines

Route agents to SSH worker hosts with live load bars in the host pool. The same wrapper runs every backend remotely.

03 — DASHBOARD

Full visibility into every agent

A real-time board, issue inspector and dependency graph — and itervox watch for the same view in a terminal.

Kanban board with Backlog, Todo, In Progress, In Review and Done columns; each card shows its profile, backend and run state Kanban board with Backlog, Todo, In Progress, In Review and Done columns; each card shows its profile, backend and run state
Issue detail of a running issue: the current step, changed files, steering box and status history Issue detail of a running issue: the current step, changed files, steering box and status history
Steer a run while it works See the current step and changed files, send the agent a message, review the diff.
Dependency graph with tracker and inferred blocker edges between issues Dependency graph with tracker and inferred blocker edges between issues
See what blocks what Tracker relations and inferred edges in one graph, with a per-issue override.
04 — IN CONTROL

Review, steer and see inside every run

Send a whole review to the agent, watch what each run does to your machine, and reach everything from one palette.

Changes tab with a line comment and a draft review ready to send to the agent Changes tab with a line comment and a draft review ready to send to the agent
Review and continue Comment on diff lines, collect a draft review, send it to the agent in one action.
Issue view with the run’s resource use and its activity audit Issue view with the run’s resource use and its activity audit
System visibility CPU, memory, processes and disk for every run, plus an audit of commands, files and connections.
Command palette searching for “pa”: Pause dispatch, pause a running issue, matching settings Command palette searching for “pa”: Pause dispatch, pause a running issue, matching settings
Command palette Ctrl/⌘+K finds issues, actions, settings, profiles and automations.
Automations page with three rules and starter templates Automations page with three rules and starter templates
Automations Cron and event-driven helper runs on your own profiles, from starter templates.
New task dialog: title, instructions, profile, backend and base branch New task dialog: title, instructions, profile, backend and base branch
New task A one-off task with its own instructions, no tracker issue needed.
Hub page with three projects and what each needs Hub page with three projects and what each needs
Every project, one hub The hub lists every project on the machine and what needs you.
05 — BACKENDS & TRACKERS

Bring your own agent. Keep your tracker.

The backend comes from the command's program name. Route issue types to different profiles, models and machines.

Agent backends

5 built in
claudeClaude Code — the defaultActive
codexOpenAI Codex CLIActive
opencodeOpenCodeActive
cursor-agentCursor CLIActive
customAny command that speaks the itervox agent protocolJSON lines
How each backend is launched →
# .itervox/settings.jsonc
"agent": {
  "command": "claude",
  "profiles": {
    "coder-opencode": {
      "command": "opencode --model …"
    },
    "coder-cursor": {
      "command": "cursor-agent --model …"
    }
  }
}
06 — DESKTOP APP

Your fleet, one click from the menu bar

Lists every project's daemon, starts, stops and restarts them, tells you when an agent needs input or a run fails, and opens each dashboard in its own window. No terminal needed: the daemon ships inside the app.

macOS menu barLinux tray · AppImage, .debWindows tray · MSIUpdates itself
v0.3.3
Download for macOS Universal .dmg · Apple silicon and Intel Unsigned for now

The app’s Homebrew cask comes with the first signed release; until then, use the .dmg.

Command line only (the daemon, no app)
brew install vnovick/tap/itervox
All downloads and checksums →
The issue view in a desktop app window, with Board, Sessions and New task in its bar The issue view in a desktop app window, with Board, Sessions and New task in its bar
07 — START

Ready in two commands

Try a throwaway board with a scripted agent — no tracker account, API keys or agent CLI needed. Then point it at your repository.