For developers · Remus

Control what your AI agents can do

Remus checks each tool request an agent makes before it runs: allow it, block it, or hold it for a reviewer's approval. Permissions are enforced outside the agent, because prompts alone cannot enforce access rules.

Local developer preview · JavaScript SDK · Node.js 24 · no hosted account required

How Remus works

01
Permissions
Set what each agent may read or change, on which records. Requests outside those limits are blocked.
02
Approvals
Hold sensitive actions for a person. If the request changes, such as a different amount or target, it needs a new approval.
03
Decision history
Record who requested each action, which rule applied and what happened next.

Use cases

Customer operations

Let support agents read the assigned customer's record, and block deletion and access to other customers' data.

Financial actions

Allow small refunds automatically, require review above a threshold and block amounts beyond the agent's limit.

Internal tools

During an incident, let an agent read the affected service's logs and keep unrelated services and credential changes out of reach.

Good to know

Tool credentials stay with the service that checks permissions, so an agent cannot bypass the check with its own credential.
Requests carry both the agent and the user it acts for; an action must fit both sets of permissions and the current task.
The quickstart runs locally: allowed reads, blocked actions and approval-gated refunds against a local ledger.

Also by New Odyssey

More for developers

Next step

Try Remus locally