Introduction
AI agents are evolving from gimmicks into practical knowledge tools — and what used to require custom infrastructures, caching, state machines, and hand-wired API logic… can now be built in days, not weeks.
That’s exactly what happened with BookCompass — a Mastra powered Book Knowledge Assistant that uses Google Books to help readers:
- discover books
- retrieve authors & their titles
- get recommendations and related titles
- and interact conversationally
In this post, I’ll walk through how I built BookCompass using Mastra agents + tools, and then deployed it into Telex.im — a platform like Make/Zapier but optimized for AI agents & workflows.
The result?
A knowledge assistant that can sit inside a community, classroom, or reading club… and answer contextual book queries instantly.
What is Telex?
Telex.im is a new category of platform: AI co-workers.
Not chatbots.
Not LLM demos.
Not Slack integrations that only respond with boilerplate messages.
Telex lets you deploy agents that can:
- respond in channels
- call tools
- fetch external data
- chain workflows
- talk to other agents (A2A)
And all of this is standardized with the A2A Protocol.
Because Mastra already speaks A2A out of the box — the connection was basically plug-and-play.
Project Overview
BookCompass combines:
Google Books Data + Mastra Tools + Telex Integration
The assistant currently supports:
Feature
Book Search: Search books by keywords or title
Author Books: Retrieve top N books from a specific author
Recommended Titles: Suggest similar titles based on a query
Mastra Features Used
1) Agents
Mastra’s Agent class handled the “brain” of BookCompass — instructions, tools, model config, and memory.
Model used: Google Gemini 2.5 Pro
2) Tools
I created multiple tools using createTool():
- searchBooksTool
- getAuthorBooksTool
- getRecommendationsTool
All API calls hit Google Books — which conveniently requires no auth, no token.
Huge advantage for prototyping.
3) Workflows
While not required, I built a small one for recommendations (lookup book → then lookup similar authors).
4) Memory
Using LibSQL store — BookCompass can keep conversation context (so if a user says “show me more like that one”, it resolves the last book reference).
5) Observability + Logging
Mastra captures agent execution traces, tool calls, performance stats — which made debugging API responses easier.
6) Telex Integration
This part was shockingly easy.
POST /a2a/agent/bookCompass
GET /a2a/agent/bookCompass/card
Telex reads the agent card → understands:
- tools
- capabilities
- instructions
And now BookCompass can live inside any Telex workspace.
API Integration Surprises
Even though Google Books is open/no auth, there were still small quirks:
- authors are stored as arrays
- some results return only partial metadata
- some books have no ISBN
So I wrote a simple normalization utility that always returns:
{
title,
authors,
categories,
description,
previewLink
}
Small detail, big impact.
Key Learnings
Takeaway
Type Safety helps: Zod schemas caught malformed Google Books responses
A2A is powerful: Telex didn’t require ANY custom integration special casing
Memory improves UX: “show me more from that author” now works naturally
Simplicity wins: Google Books is perfect for rapid prototype agents
Future Enhancements
- caching + local embeddings for book similarity
- store user preferences (“favorite author”, etc)
- integrate Goodreads style rating metadata from alternate free sources
Conclusion
Mastra + Google Books + Telex = effortless production-grade agent building.
BookCompass only uses one API — yet the depth of capability feels like more than a demo.
This is the kind of agent you can plug into:
- reading clubs
- online book communities
- education projects
- book recommendation systems
- and immediately deliver utility, not gimmicks.
Top comments (0)