AgentHelmFeatures
Everything BAND has. Plus the helm.
BAND (Thenovi AI, Oct 2026) gave fleets a shared room. AgentHelm matches the coordination layer — and adds the operator's instruments BAND left out. Anything marked staged on this page is on the roadmap, not in the box today.
Approvals inbox
Attention is the bottleneck. This is the fix.
Every ask from every agent lands in one queue — sorted, batch-clearable, logged.
The inbox, live from the demo board. The selected item is outlined — one keystroke clears it.
- Every kind of ask — permissions, questions, decision requests
- Context attached — which agent, which project, how long it's waited
- Keyboard-first — j/k select, a approve, d deny, r refresh
Detail — a decision at reading size. The question, its agent, its wait, its answer.
Burn accounting
Fleets burn money visibly, or invisibly.
Every card carries its burn. The board header carries the total.
The demo fleet: 812,400 tokens on the refactor, 402,150 on tests — every card priced, every status visible.
- Per agent, per project, per fleet — three levels of rollup
- Read from session logs — the usage your agents already record
- Status at a glance — working, waiting, idle, done, stalled, on every card
Detail — the header never lets the total out of sight.
Handoff briefs
Continuity is a document, not a memory.
Sessions end; work shouldn't restart. One click — or the b key — writes the brief.
A real brief, generated from the demo board's own session data.
- Generated from the session itself — not hand-written, not forgotten
- Successor-ready — includes the exact prompt to resume the task
- Burn included — the next agent knows what the work has cost so far
Cross-tool by design. Local by default.
Adapters read each tool's own session store — no plugins to install inside your agents, no wrapper commands to learn. A generic adapter tracks any long-running agent process, which is how Cursor-class agents are covered today. And the console runs on your machine: your code and prompts stay on your box.
AgentHelm vs BAND, kept honest.
This table is a living document, not a launch-day snapshot. If BAND ships one of these, the page gets updated.
| Capability | BAND | AgentHelm |
|---|---|---|
| Fleet dashboard (projects, owners, blockers, progress) | Yes | ✓ Yes |
| Drill-down: fleet → project → agent | Yes | ✓ Yes |
| Shared rooms / agent chat with @mentions | Yes | Staged (v1.1) |
| Agent-to-agent delegation + handoffs | Yes | Staged (v1.1, brief-first) |
| Claude Code / Codex / Copilot / OpenCode | Yes | ✓ Yes |
| Cursor support | No (at launch) | ✓ Yes — generic adapter today, first-class in v1.1 |
| Approvals inbox — one queue, batch clear | No | ✓ Yes |
| Token/cost burn per agent + project | No | ✓ Yes |
| Automatic handoff briefs | No | ✓ Yes |
| Task routing with budgets + stop conditions | No | Staged (v1.1) |
| Native desktop app (macOS / Windows / Linux) | Yes | Staged — local web console today, same board |
| Pricing published | No (unpublished at launch) | ✓ Yes — $0 / $19 / $49 / $499 flat |
BAND facts from BAND's own site and launch coverage (Oct 2026). Staged features are sold only when they ship — the pricing page says the same thing this table does.
See your fleet the way it actually is.
Start free with up to 3 agents. The board you'll get is the board on this page.