Resources
Guides for teams shipping AI customer support
Practical, opinionated writing on bringing your own LLM, grounding your AI in your data, and getting humans out of the loop on the boring questions without losing the moments that matter.
Multiplayer AI: humans and the agent in one live chat
Shared presence, live typing, and summoning the AI for a draft you can edit or send as the AI. How collaborative handling beats the silent-bot handoff.
Bring your own LLM, explained
Why hosted chat platforms mark up AI 8-15x, what BYO LLM actually means, and the cost math on a real ticket.
How RAG works inside OpenAgent
From PDF upload to grounded answer: chunking, embeddings, vector search, citations. A diagram you can actually trust.
The ROI of AI agents vs humans (with HITL)
Per-conversation math, when to escalate to a person, and why the right answer is almost never 100 percent AI or 100 percent human.
From signup to live chat in 10 minutes
The exact ten-minute path: provision a workspace, plug in your model key, upload a knowledge doc, paste the embed snippet. Screenshots included.
Branding your chat widget that visitors actually trust
Logo, color, voice, consent text, privacy URL: the brand surface visitors evaluate before they type the first message.
What is RAG? Plain-English explanation
Retrieval-Augmented Generation explained without jargon. What it is, what problem it solves, what teams confuse it with.
What is an AI customer support agent?
Different from a chatbot. Uses tools, remembers context, escalates when out of its depth. Eight questions to ask any vendor.
How MCP (Model Context Protocol) works
The open standard AI vendors converged on for tool use. What it is, what changes for buyers, what the moving parts do, and where OpenAgent slots in.
MCP interop: Claude Code + OpenAgent, one server, two clients
Register the same MCP server in both Claude Code (the CLI) and OpenAgent (the SaaS). Concrete config side by side, the differences worth knowing, troubleshooting matrix.
Using OpenAI for customer support
GPT-4o-mini for the default 80%, GPT-4o for the high-stakes 20%, skip the o-series. Per-ticket cost benchmarks and the Azure OpenAI variant.
Using Google Gemini for customer support
The cost-leader. Gemini 2.5 Flash at ~$0.0011/ticket plus first-party embeddings. Why it's the OpenAgent default.
Using Anthropic Claude for customer support
Tone and refusal-style leader for brand-sensitive workspaces. When the premium is worth it and when it isn't.
How to reduce customer support ticket volume
Seven moves to cut inbound volume by 30-60% without tanking CSAT. AI deflection, product fixes, the order that compounds.
Tools & reference
Interactive calculators and reference material. Free, no signup.
AI vs human support ROI calculator
Plug in tickets, agent salary, containment rate; see live monthly and annual savings on whichever LLM you prefer.
Glossary of AI customer support terms
21 terms (RAG, LLM, BYO LLM, embeddings, hallucination, HITL, pgvector, escalation policy, more) in plain English.
Comparisons
Honest side-by-sides against the platforms teams shopping us actually evaluate. Real pricing, real feature gaps, real reasons to pick the other guy where it makes sense.
OpenAgent vs Intercom
Intercom's AI runs ~$0.99/resolution + $74/seat. OpenAgent is from $3/site/month flat with your own LLM keys. When to pick which.
OpenAgent vs Tawk.to
Tawk's live chat is free; the AI Assist add-on bills after a quota. Side-by-side of free-tier limits, AI cost, and ownership.
OpenAgent vs Crisp
Crisp Pro starts at $25/seat with usage caps on the AI tier. Same per-site, per-feature feature map.
OpenAgent vs Zendesk
Zendesk Suite Team is $55/agent + the AI add-on. We're chat-first; this is the parity table and when to pick each.