Understand the decision boundary.
Security, workflow behavior, integrations, and what Nodsend deliberately does not do.
Security, workflow behavior, integrations, and what Nodsend deliberately does not do.
Nodsend is human approval infrastructure for AI agents. Your server creates a decision request, Nodsend presents it to an authorized person, and your workflow resumes from a signed outcome.
A model-callable tool is useful, but it is not a security boundary because the model may omit it. Put the protected side effect behind an application-enforced Nodsend checkpoint.
No. Email recipients use an opaque, single-use link scoped to one approval. Workspace authentication and human decision authorization remain separate.
Nodsend signs webhook events with HMAC-SHA256. Your server verifies the signature, timestamp, and event ID before resuming the workflow.
The decision transition is atomic. Only the first valid pending-to-terminal transition succeeds; later attempts receive the already-decided state.
Expired requests cannot be approved or rejected. Your workflow can treat expiry as its own terminal outcome and escalate, retry, or stop safely.
The core API works with any stack. The repository also includes optional Python adapters for LangChain/LangGraph, CrewAI, and AutoGen using each framework’s native control-flow patterns.
No. Nodsend records authority; your trusted application owns execution. This keeps credentials and side effects inside your infrastructure.
Email delivery is the supported channel today. We do not advertise Slack as available until its authentication, interaction security, and delivery lifecycle are complete.
Only on your server or in a secrets manager. Never put an API key in frontend code, a public URL, or a model prompt.