OneCLI vs Arcade
Arcade lets an agent act in each end user's accounts. OneCLI gives a sandboxed agent the access its job needs.
| Topic | Arcade | OneCLI |
|---|---|---|
| Built for | Agents acting on behalf of each end user | Sandboxed agents that must never hold a key |
| Where the credential goes | Arcade holds the user's OAuth tokens and runs the tool | Stays in the gateway and is added to the agent's own request |
| What it covers | Tools that run through Arcade's runtime | Any outbound call: MCP, CLI tools, curl, generated code |
| Access control | Each user authorizes the scopes a tool asks for | A policy per agent or job, listing systems and actions |
| Source | Open-source tool SDK, cloud or self-hosted engine | Open source, self-hosted or cloud |
Based on Arcade's public documentation.
Choose Arcade when
- You are building a product where every user signs in to their own Gmail or Slack.
- You want to write custom tools with an SDK.
Choose OneCLI when
- Your agents work with the company's own credentials.
- An admin, and not each user, decides what an agent may do.