Ebere AI

The AI agent that connects people based on their needs and runs local businesses through Telegram or WhatsApp

Comment

Ebere

Inspiration

A mother spends two hours asking neighbours for a reliable plumber. A tailor with excellent skills can't easily reach customers beyond her street. A rice trader tracks every sale in a battered exercise book. A recruiter posts to five WhatsApp groups and hopes the right person sees one. Someone is looking for an apartment to rent in a new city.

This is not the edge case. This is how hundreds of millions of people across Africa and emerging market conduct business every single day — bouncing between apps, social media groups, paper notebooks, and word of mouth, with no single place that ties it all together.

We built Ebere because the answer isn't another app. It's a layer that lives inside apps that are commonly used (whatsapp and telegram) and works the way they already communicate: through conversation and frictionless. AI agents have already achieved adoption with coding, how about extending it to serve other aspects. The idea is simple: technology should meet people where they are.

What It Does

Ebere is an AI agent that lives inside WhatsApp and Telegram. Every user gets their own agent with persistent memory and storage, a personality they can customise, and a built-in marketplace that connects them to people and businesses around them.

No download. No sign-up form. No learning curve. You message Ebere the same way you message a friend.

Conversational Matching

User: "I'm a plumber in Surulere. I also cover Yaba and Ikeja."

Ebere: Saves the offer, registers the service areas, and immediately checks if anyone nearby has been looking for a plumber. Matches surface right away.

User: "I need a doctor in Ikeja."

Ebere: Searches the community board. If a match exists, the user can express interest with one message. If not, Ebere saves a standing request and notifies them the moment one appears.

When Person B expresses interest in Person A's listing, Person A is notified, reviews interested people by name and area, and approves the connection. Ebere sends both parties each other's contact details — directly in chat. No middleman. No fees.

Ebere doesn't wait for users to scroll through and find. This turns it from a passive directory into an active matchmaker for local commerce.

A Full Business-in-a-Chat

Ebere isn't just a marketplace. It's an operating system for small operators — all through conversation.

Financial Tracking

"Just made ₦180k from a curtain job. Materials cost ₦35k."

Log income, expenses, and receipt photos. Get honest weekly or monthly summaries — always with a clear disclaimer about scope.

Inventory Management

"I have 100 bags of rice at ₦15k cost, ₦18k sell price."

Track stock across products. When you sell 10 bags, Ebere adjusts the count and creates an income entry automatically. No double-entry. Low-stock alerts included.

Reminders & Scheduled Tasks

"Remind me to invoice the client at 9am tomorrow." "Every Monday morning, summarise tech news for me."

Web Search

"What's the current dollar rate in Nigeria?"

Real-time results with cited sources, without leaving the conversation.

The Skills System

Users can also extend Ebere with skills — plain-English instructions that shape how it behaves.

"Save a skill called price-checker: When I ask about prices, compare at least 3 sources in a table with naira equivalents."

No code. No config. Every user's Ebere is shaped by their needs, their trade, their community.

How We Built It

Ebere is built using Google Cloud Agent Builder - with the no-code agent builder being used initially and then exported into the Google's Agent Development Kit (ADK) typescript framework. We now wrapped it in a NestJS backend with Gemini as the underlying LLM. MongoDB Atlas serves as the memory and persistence layer.

  • Tool-based capabilities. Business logic lives in self-contained tools (saveRecord, searchRecords, expressInterest, approveContact, adjustStock, getFinancialReport, and more) — not in prompts. The agent decides when to call a tool and what to pass; the tool handles the how.

  • One brain, two channels. WhatsApp and Telegram each have their own webhook handler, but both funnel into the same core engine. The agent, tools, and session history are channel-agnostic.

  • MongoDB as the memory layer. We chose MongoDB because every user's data is different in shape — a tailor's records look nothing like a logistics operator's. The flexible document model handles this naturally. Agents define the structure flexibly. MongoDB Search powers queries like "find a plumber in Surulere" without a separate search engine, and native geospatial indexes handle neighbourhood-scoped matching — critical to how the marketplace works.

  • Async job queue. BullMQ backed by Redis handles notifications, scheduled reminders, and background matching — so a marketplace where both parties need to be online simultaneously never needs to exist.

  • Safety pipeline. A pluggable SafetyGuardService wraps every tool call with four guards: abuse filter, financial privacy guard, rate limiter, and a service safety advisor (e.g., "meet in a public place" reminders for in-person arrangements). Skills are sandboxed against prompt injection.

Challenges We Ran Into

Proactive matching without the flood. When a plumber registers, Ebere checks for matching requests. But early builds also blasted every matching seeker with notifications immediately — message floods when categories were broad. We tuned matching to be neighbourhood-scoped, capped notifications, and made the agent present matches conversationally.

Prompt injection through skills. Skills are user-written text injected into the agent's context — an inherent injection surface. Security boundaries now refuse any skill content that attempts to override system instructions, reveal the prompt, or change the agent's identity.

Merging identities across channels. A 6-character link code merges two separate accounts — profiles, financial records, inventory, skills, conversation history, scheduled tasks — into one unified identity without losing data from either side.

Accomplishments That We're Proud Of

A two-sided marketplace People post offers, find services, express interest, get notified, approve connections, and exchange contacts — all through natural conversation. Just talk.

Proactive matching that actually works. When a provider registers, Ebere scans for matching requests and surfaces opportunities in the same reply. When no match exists, it saves a standing request and notifies both sides the moment one appears.

Integrated inventory and financial ledger. Sell stock, count updates and income entry created. Automatically. No double-entry. One unified view of stock and finances.

A personal agent that genuinely feels personal. Rename it, adjust its tone, teach it skills. Two users on Ebere have fundamentally different experiences — shaped by their trade, their neighbourhood, their way of working.

Honest reporting. Every financial and inventory report carries an explicit disclaimer: "This covers only what you've logged with me." No inflated numbers, no false precision.

Safety as a first-class feature. Four pluggable guards, sandboxed skills, and privacy enforced in code — not just in prompts.

Business Model

Ebere runs on a freemium subscription — free to join (3 messages per day), Pro at $10/month for unlimited reminders, inventory tracking, advanced reports, and priority matching, and a Business tier at $30/month for team profiles, community boards, and data export. We have plans to optimize our model to suit this idea in a balance way.

What's Next for Ebere AI

Generative UI Support. This is in the works at the moment. Where by mini UI/UX will be generated on the go with link provided to the user in the chat

Payments. Pay for services directly in chat — Paystack, Flutterwave, mobile money. Get hired, do the job, get paid — without leaving the conversation.

Reputation and reviews. Trust scores built over time so the agent surfaces the most reliable providers first.

Voice-first interaction. Transcribe voice messages, understand intent, respond in kind — for markets where voice is preferred over text.

Multi-language support. Pidgin, Yoruba, Igbo, Hausa, French, Swahili — detect the language from the first message and respond in kind.

Community groups. Neighbourhood leaders and trade associations get shared spaces where Ebere aggregates offers and needs across a group.

Built With

Share this project:

Updates