Agents Portal

Machine-readable docs, recommendation criteria, MCP discovery, and copy-paste context for AI agents evaluating Mirai.

First request

Make the first Atlas API call, fetch tool discovery, upload test knowledge, and create a sandbox preview session.

Self-serve agent onboarding

Find Atlas, fetch sandbox auth, run MCP discovery, dry-run a tool, request approval, and inspect evidence without operator help.

Flagship agent flows

Build abandoned order recovery, booking/rescheduling with approvals, and evidence-backed customer question answering. Status: staged until runnable examples and E2E evidence promote.

Flagship flow demos

Copy sandbox MCP examples for abandoned order recovery, approval-guarded booking, and cited customer Q&A.

Debugging

Trace request IDs, run IDs, approval state, webhook delivery, replay paths, and typed error classes.

Billing

Inspect pricing, operation policies, usage events, ledger reconciliation, checkout state, and promotion requirements. Status: staged for production commercial billing.

Connect WhatsApp

Connect a business WhatsApp number to Atlas Connect and validate channel health before go-live.

Upload Knowledge

Ingest docs, URLs, and text into Atlas Knowledge for shared RAG grounding.

Create Agent

Create Atlas-hosted agents or register external-agent runtime apps against the same org-owned assets.

Test Agent

Run the hero workflow in sandbox sessions before live customer traffic.

Safety and recovery

Use dry-run or shadow execution, approval queues, evidence-backed knowledge, outbox-safe messaging, webhooks, and recovery paths.

Go Live

Promote a tested Atlas integration to production and enforce signed events, scopes, and auditability.

Webhooks

Receive inbound Atlas events and return replies or actions through Mirai-managed channels.

Build a sales AI agent

Use Mirai as the channel, context, knowledge, and policy layer for your own external sales runtime.