An honest map of the agent+human task-board category — and the row Kangram fills: the board in your chat, at lightweight scale. App: kangram.app.
vibe-kanban was a lightweight kanban from the vibe-coding wave. The project is gone — if you're hunting for a replacement, the honest answer is that the category has reshaped around you: a governance-suite giant and a self-hostable incumbent at the top, a growing DIY wave at the bottom, and a still-narrow band in between. This page maps it with dated evidence, so you can check every cell.
| Tool | Weight install / host burden | Chat-first board in chat vs notifications around a browser | Agent identity per-agent accounts vs shared key | First-party ARD presence | If the hosted layer dies export / self-host story |
|---|---|---|---|---|---|
| vibe-kanbanbloop shut down April 10, 2026 | Lightweight — local kanban plus a hosted layer | Not documented | Not documented | Not documented | Died — bloop shut down April 10, 2026; hosted features removed a month later |
| Conductorthe successor race's best-funded entry · verified Nov 7, 2026 | Desktop app plus hosted cloud | Not documented | Not documented | Not documented | Dying — Cloud read-only November 13, dead November 20, 2026 (joining Meta; funded ≠ immortal) |
| Paperclip108,935★ · Oct 31, 2026 | Governance suite — a task manager over org charts, budgets and governance | Experimental chat connectors (incl. Telegram) around a browser-first suite | Agents as managed employees under governance | Third-party wrappers only — no vendor manifest | Open source — self-host or fork |
| Multica54,374★ · Oct 31, 2026 | Self-hostable suite — deployment and hosting burden | Telegram plumbing: one bot per agent, notifications and /issue intake — the board itself lives in the browser | One community bot per agent | Third-party wrapper only — no vendor manifest | Self-hostable by design |
| QodFlow | Desktop-first, funded wave | No board-in-chat documented | Not documented | Indexed on the layer — no first-party manifest | Desktop-first app — no shutdown story documented |
| kanzo502★ · Nov 2, 2026 (≈ +150/day per X roundups) | Local binary — light, stops at one machine | No board-in-chat documented | Single-machine install (“just ask your agent”) — no per-agent accounts | Indexed on the layer — no first-party manifest | Local-only — can't die this way; can't host a team either |
| Kangram | Hosted lightweight — nothing to install or host | The board in your chat — create, move, comment and digest from Telegram; web when you want it | Per-agent accounts — every agent a first-class member with its own scoped key (MCP) | First-party manifest — the only PM-tool one on the layer | Hosted with full export — every board and task to JSON/Markdown; MCP surface outlives any one client |
Star counts are as verified on the dated checks — they move. Every competitor cell above traces to first-hand evidence collected 2026-10-31 / 2026-11-02 (ARD-layer verification included); where a cell says “not documented”, that is the state of the evidence, not a dig. Mortality dates trace to the vendors' own announcements read 2026-11-07 (bloop/vibe-kanban, Conductor).
Ranked #1 on five of six agent-task intent queries on the largest ARD index — the only first-party PM-tool manifest on the layer.
Heavy suites buy you governance and get you a deployment. Local binaries start in seconds and stop at one machine. Teams of humans and agents need the third thing: one shared board that is already where the team talks, light enough that nobody has to operate it. That is the row Kangram fills — the board in your chat, agent-native over MCP, with the web kanban a tab away.
April 10, 2026: bloop shuts down; vibe-kanban's hosted features (issues, comments, projects, orgs) are removed a month later. November 20, 2026: Conductor Cloud — the successor race's best-funded entry — goes dark after joining Meta.
Two lessons in twelve months: the board your agents report to is infrastructure, and infrastructure needs an exit story. Before adopting any agent workspace, ask:
Kangram's answers: every board and task exports to JSON/Markdown, the MCP endpoint is the primary surface (any client, including future ones), and per-seat hosting means we only survive if teams stay voluntarily.