Trende

Inspiration AI-driven market research shouldn't require blind trust. We built Trende because media signals already move markets — yet the AI systems interpreting those signals are opaque by design. We changed that.

What it does Trende is an intelligence-to-asset pipeline that transforms social signals into conviction-ready research — with cryptographic proof. Users submit a research query; multi-platform agents gather signals across the web; AI models reach consensus; and a Trusted Execution Environment attests the entire process from input to output.

How we built it The backend is a LangGraph agentic pipeline (Python/FastAPI) running a structured multi-stage workflow: Planner → Researcher → Validator → Consensus → Architect. It pulls from 8+ data sources including [Twitter, TikTok, LinkedIn - coming soon], NewsAPI, Tabstack Api, and the TinyFish AI agent, with cryptographic attestation handled via EigenCompute's TEE infrastructure. The frontend is built in Next.js, with kinetic typography animations that make the research process feel as dynamic as the data behind it.

Challenges we ran into The hardest problem was treating the TinyFish autonomous agent as a first-class research source rather than just another API — it required rethinking how we handle non-deterministic inputs in a pipeline designed for consistency. Getting cryptographic attestation to work reliably inside the TEE took more iteration than expected. We also had to strike a careful balance between research depth and execution speed: comprehensive analysis is worthless if the UX feels like it's thinking forever.

Accomplishments that we're proud of Building an AI research agent with TEE-attested consensus — a meaningful step toward verifiable AI. We shipped production-ready ACP integration for agent-to-agent commerce, and built a working multi-model consensus layer across Venice, AIsa, and OpenRouter that genuinely catches blind spots no single model would surface alone.

What we learned Verifiability matters more to users than we anticipated — people don't just want good answers, they want to know the answer can be trusted. Multi-model consensus proved its value quickly: models disagree in ways that reveal real uncertainty. And on the product side, progressive disclosure made a measurable difference in how new users found their footing.

What's next for Trende On the infrastructure side, we're focused on scaling Agent Commerce Protocol (ACP by Virtuals) operations to handle concurrent agent jobs, and expanding our data sources to include GDELT, Wikimedia, and StackExchange for deeper, more defensible research coverage.

Built With

• eigen
• kilocode
• langchain
• nextjs
• python