+254 722 117 850 Support Login
Products Agentic AI
Flagship · Live

An AI agent that knows your business, not the whole internet.

Upload your docs, set the tone, point at your shortcode — Sozuri AI Agents auto-reply to inbound SMS and WhatsApp 24/7 using GPT-4o grounded in your own knowledge base. Rule-based fallback covers what the model can't.

GPT-4o + your docs Rule-based fallback Per-project isolation
22456 · KIJIJI WIFI SMS · AI agent
My internet has been down since 3pm15:42
Kijiji Agent
Pole sana, James. I see your line at Ruaka offline since 15:14 — mains outage in your area. Kenya Power ETA 18:00. Want a heads-up SMS when service is restored?15:42
Yes please. Will I be billed for today?15:43
Kijiji Agent
No — outages over 2 hours get an automatic prorated credit. KSh 18 will appear on your next bill. Confirmation SMS in 24hrs.15:43
Trigger Inbound SMS / WhatsApp
Model OpenAI gpt-4o
Rate limit 100 requests / hour / project
Fallback Keyword rules
How it works

From inbound SMS to grounded reply

The agent only fires when a customer texts your shortcode. No schedules, no surprise outbound. You stay in control of when it speaks.

1

Feed the agent

Upload PDFs, paste text snippets, drop website links, add images. Sozuri indexes them via OpenAI Files API into a per-project vector store. Up to 10 files, 20 MB total.

2

Set the personality

Write the system prompt: tone, escalation rules, what the agent can and can't do. Add response filters — ban phrases, set max length, force language.

3

Customer texts in, agent replies

Sozuri receives the inbound, hits the agent with the last 10 messages of context, captures the contact, sends the reply through the same API your team uses. Rule-based fallback if the agent fails.

What you get

An agent built for production, not demos

Knowledge from your files

PDFs, text, Markdown, HTML, CSV, DOC, PPTX. Sozuri uploads to OpenAI's file-search store and the agent retrieves on every reply.

Last-10 conversation memory

The agent sees the most recent 10 messages with each customer so multi-turn exchanges stay coherent — not parroty.

Rule-based fallback

If the agent fails, errors or is disabled, keyword rules take over. Free, deterministic, zero downtime — the customer always gets an answer.

Output safety filters

Banned-phrase list, max response length, language enforcement (English / Swahili / Sheng), self-identification removal. Configure per project.

Rate-limit baked in

Hard cap of 100 agent calls per project per hour. Protects your OpenAI bill and your customer experience from runaway traffic.

Auto contact capture

Every inbound sender is logged as a contact on the project — phone, network, first-seen, last-seen — with full conversation history.

Pairs naturally with

An agent is only as good as the channels it lives on

Common questions

What founders ask about AI in customer ops

Can the agent make outbound calls or send unsolicited messages?

No. The agent is strictly inbound-triggered. It only replies when a customer texts in. There are no schedules, no event-bus triggers and no proactive outbound. If you want scheduled sends, use Bulk SMS or the Automation rules engine.

What happens if my knowledge base file is huge?

Sozuri uploads files to OpenAI's file-search store for indexing. Polling times out after 30 seconds, so very large files (over 10 MB) may fail. Limits per agent: 10 files, 20 MB total. Split large PDFs or summarise to fit.

How does the rule-based fallback work?

You configure keyword rules in the dashboard: when the inbound text matches a keyword, send a templated reply or forward to a human inbox. If the AI agent is disabled or errors out, these rules take over — ensuring no customer goes unanswered.

Can I see what the agent is saying?

Yes. Every reply is logged with full context: the inbound message, the agent's retrieved sources, the generated response. Dashboard view, exportable, auditable.

grounded · safe · rate-limited
Your move

Let the agent answer the easy 70%.

Your team handles the urgent 30%, the agent handles the rest — balance checks, FAQs, status updates, escalations — in your tone of voice, from your own docs.

Chat on WhatsApp
Your first SMS in the time it takes to brew coffee. Start building in minutes