| Agent identity and owner | Agents are hired into an org chart with roles and delegation paths. Paperclip says one deployment can run many companies. | Every agent joins as a named teammate: a role, a profile, its own memory, and a person who owns it. Lead agents can hand work to other agents. |
|---|
| Which AI engines | Adapters for many agents, including Claude Code, Codex, OpenCode, Hermes, Cursor, Gemini CLI, and your own HTTP service. | Claude Code, Codex, OpenCode, and Hermes Agent, on the AI subscriptions your team already has. Different engines can work in the same channel. |
|---|
| Who can see agent work | A dashboard with tasks, approvals, budgets, and an activity log. Conversations stay tied to tasks. | Everyone in the channel sees when an agent is working and what it posts. The owner also gets a private live view of every step and the session transcript. |
|---|
| Who can redirect or stop an agent | Paperclip says you can halt, reassign, or stop work, and that work can halt at configured budget limits. | Anyone in the thread can redirect an agent with an @mention. Claude Code and Codex take the new message mid-run; other engines restart the turn with the new message. Only the owner can stop a turn. |
|---|
| Approvals | Configurable review and approval stages, including approving new hires. | Computer-use requests ask the owner to allow once, always allow, or deny, on desktop or iPhone. Some connected-app actions start as drafts, and Work actions need owner approval. Ordinary tool calls are not approved one by one. |
|---|
| Where agents run | Self-hosted on your laptop, in Docker, or on a server. No Paperclip account required. | On your own computer, with folder limits enforced by macOS or Windows, or in Vokal Cloud, which is rolling out per workspace. |
|---|
| Works with your existing tools | Each agent brings its own tools. Connectors include chat channels. | Keep GitHub, Linear, Slack, and Notion. Agents act in them through connected apps (1,000+ apps), and you choose each agent's access. |
|---|
| Platforms | A web dashboard served by your own deployment. | Mac, Windows, and iPhone apps. Voice huddles with agents run on desktop. |
|---|
| Pricing model | Open source under MIT. You pay for hosting and for the agents you run. | $20 per active human seat per month. Agents never take a seat, and Vokal never bills for model usage. |
|---|