No real pipeline.
No real visibility.

A real pipeline. Not a patch over spreadsheets.

Most sales and support teams run their leads on spreadsheets, generic CRMs or disconnected tools. There are no clear stages, no visual Kanban, no way to know where each contact stands.

With ATLAS, the pipeline is native. A single pipeline with draggable stages that all campaigns and teams share. Each campaign and each team generates its own Kanban experience on top of that same pipeline, seeing only its contacts and its context.

What the ATLAS pipeline includes

• ✓ Draggable stages with AI/Human/Shared type: define the phases and control who runs each one. AI qualifies, the human closes.
• ✓ No-code visual triggers: 4 events (ON_ENTER, ON_EXIT, ON_TIMER, ON_DEMAND) × dozens of actions (move stage, create task, send message, switch agent, escalate, webhook, n8n...).
• ✓ Kanban board with in-message search: filters by tag, full-text search with highlighting, new-message LEDs and full chat from the card.
• ✓ Per-contact matrices: structured tables that AI fills during the conversation. Mortgage qualification, document checklist, product catalog.
• ✓ Calendar integrated with Google Calendar: AI books appointments respecting business hours and holidays. Briefing email to the assigned operator.
• ✓ Tasks with 5 types: call, email, meeting, follow-up and generic task. Structured close with outcome and automatic follow-up.
• ✓ Campaigns with T0–T3 tiers: interest classification (no interest / low / medium-high / appointment or task) with close and sign-off configurable per tier.

⚙️ Triggers, in depth

Every stage can have its own behavior. No code, no integrators.

Four events per stage (ON_ENTER, ON_EXIT, ON_TIMER, ON_DEMAND) and dozens of actions — notify, assign team, reassign AI agent, launch a message matrix, escalate with a return stack, call a webhook, run an n8n workflow. All with conditions, time windows and audit log on every change. Two real examples running today:

ON_TIMER · 5 min → Notify supervisor

The customer asked to be escalated. If nobody picks up in 5 minutes, the supervisor is notified.

When a conversation enters the "Awaiting escalation" stage, ATLAS starts a timer. If no human on the assigned team has taken the conversation within 5 minutes, it fires a notification to the supervisor — with the contact's name, the originating channel and the reason on record. The supervisor reassigns or takes it themselves. Nobody is left waiting because someone missed it.

Per-stage configuration: time window, recipients (team supervisor, specific role, named user), message template with variables ({contact.firstName}, {stage.name}, {wait.minutes}) and delivery channel. Every firing is recorded in the audit log.

ON_ENTER → Escalate to team + swap AI agent

Entering "Technical handling" swaps both the team and the AI agent — without the customer losing the thread.

The conversation was running with the "Sales specialist" agent (LLM A, commercial knowledge base). On entering the "Technical support" stage, the Escalate to Technical Support trigger does two things at once: it assigns the conversation to the human technical team and replaces the AI agent with "IMHOIT — Technical Support" (a different LLM, possibly more specialized, with its own technical knowledge base — manuals, troubleshooting, resolved cases).

The full conversation context travels with it. The customer repeats nothing. And when the technical issue is resolved, the stage returns to "Sales", the escalation stack restores the original commercial agent — which picks up the conversation with the full memory of everything that happened in between.

The customer perceives a single conversation. Underneath, ATLAS swapped stage, swapped human team, swapped AI agent, swapped LLM model and swapped knowledge base — three times. Memory is shared, the customer never repeats, the supervisor sees the full trace.

Multi-agent AI in a single session, with shared memory and a native escalation stack.
The customer doesn't notice when the stage changed, when the team changed, when the AI agent changed or when the LLM model changed. They only notice that they were answered with authority — at every moment — and that nobody made them repeat themselves.

No real visibility: you can't improve what you don't measure.

Who handled that customer? When? With what outcome? How long did they take to reply? In most tools, audit logging and advanced analytics are paid add-ons reserved for enterprise plans. In ATLAS, they're included from day one.

What ATLAS measures and logs

• ✓ Live KPIs without ETL: active campaigns, conversations by state, message volume, operator productivity — all in real time on top of PostgreSQL.
• ✓ Audit with dozens of actions: who did what, when, from which IP, with before/after values. Charts, bulk archival and anomaly filtering.
• ✓ Multi-level escalation across teams: escalation stack with return-to-origin. The case goes up, comes down or returns without losing traceability.
• ✓ 4 RBAC roles with real-time presence: Admin, Campaign Manager, Supervisor, Operator. Granular permissions, auto-away and workload dashboard.
• ✓ Dynamic segments: define audiences by tags, fields, states and dates. Segments are recomputed on every query — always up to date.
• ✓ Global health monitor: permanent icon with 7 categories (DB, WebSocket, channels, AI, queue, storage, memory). Audible alarm on red and anomalies audited.