Case study
Liveline
AI agents handling WhatsApp, Instagram and Facebook from a single inbox.
Context
A business that sells over chat doesn't have just one kind of conversation. Selling isn't support, and support isn't scheduling. One generic bot covering all three answers all three badly.
Liveline is our multi-agent platform for Meta's hub. Instead of a single bot, it runs several distinct AI agents, each with its own knowledge base, and hands the conversation to a person when it needs to.
The challenge
- Three channels — WhatsApp, Instagram and Facebook — with three separate inboxes.
- A single generic bot can't cover sales, support and scheduling at the same quality.
- Invented answers whenever the agent has nothing real to draw the fact from.
- The handoff to a human, which almost always arrives too late.
What we built
- 01
AI agents with their own knowledge bases
Each agent runs on its own knowledge, loaded from the business's menu, catalog or policies, so it answers with data rather than assumptions.
- 02
Meta's hub in one inbox
WhatsApp, Instagram and Facebook converge into La Central, with a New / In Progress / Resolved pipeline.
- 03
Rule-based escalation to a human
The business defines the point where a conversation stops being the agent's and becomes the team's.
- 04
Team collaboration with roles
Shared work on the same inbox, with role-based access.
- 05
Full API and BYOK
Complete API access and the option to bring your own AI key, so Liveline can be wired into other platforms.
- 06
Conversation metrics
Reporting and metrics on conversation volume and resolution.
Results
- All three Meta channels handled from one shared inbox.
- Different agents for different jobs, each on its own knowledge base.
- Live in production, with plans at 3,000, 5,000 and 7,500 MXN per month by conversation volume.
- Full API and BYOK available for custom integrations.
Stack
- PHP
- Docker
- Hetzner
- Vector database
- Job queues
- Meta Graph API