Kirha Monthly Recap — December 2025

December closed out the year with some of our most impactful work yet. We shipped proof that our approach works, opened new data verticals, made progress on spreadsheet integrations, and watched our community take shape. Here’s what happened.

What We Shipped

Benchmark: Kirha vs Web Search

We finally published the numbers. Our benchmark puts Kirha head-to-head against Web Search for agentic use cases, measuring accuracy, latency, and structured output quality across real-world queries. The results validate what we’ve been building toward: deterministic routing to premium data providers outperforms Web Search for domain specific queries.

New Vertical: News & Prediction Markets

We opened a News vertical combining three sources:

• Polymarket MCP for prediction market data
• Exa for news retrieval
• X MCP for real-time social signals

This gives agents access to current events alongside market sentiment. It’s useful for research workflows, trend analysis, and anything that needs to understand what’s happening right now.

We are excited to expand this news vertical by partnering with the most reliable premium sources.

Spreadsheet Integration: FormulaJS

Software engineering was the first profession to see its daily workflow completely reshaped by AI. We spend less time writing code and more time reviewing what AI generates. The first disruption came from Cursor, who forked the open-source VSCode editor to embed AI features directly into the development environment. Then Claude Code invaded our terminals. The pattern is clear: AI meets you where you already work.

2026 will be the year other knowledge professions get the same treatment: auditors, analysts, finance teams. And where do these people work? In spreadsheets. Excel is closed source, so nobody can fork it. But open-source alternatives exist: Fileverse dSheets, HyperFormula, and others. We’re enabling any spreadsheet builder to add Kirha Search and power their users with premium real-time data from the comfort of their cells. The goal: let users query Kirha directly from a formula. They can use =KIRHA() the same way they’d use =SUM() or =VLOOKUP(). Spreadsheet builders can integrate via our open-source FormulaJS implementation. See the commit.

More announcements on this front coming soon.

Semantic Tool Filtering

Last month, we migrated our MCP metadata to Neo4j. This month, we built on that foundation. We now have a compiler-inspired semantic layer that translates natural language into tool intent. It auto-generates an IR grammar abstracting tool capabilities into searchable semantics. What this means in practice: Kirha doesn’t slow down as our tool catalogue grows. We pre-filter tools before the planner ever runs. The semantic layer narrows the search space, so latency stays flat even as we add hundreds of new tools.

Business & Ecosystem

Trading Community Partnerships

We partnered with two agentic trading communities this month: Swing Vault and Hdge.

These collaborations bring concrete use cases for our Crypto and Financial Analysis verticals. Real traders using real data in production. It’s the kind of feedback loop that sharpens the product.

Odyssey Integration

We launched a partnership with Odyssey, powered by Umi Network.

Kirha is now featured directly on their platform through interactive learning quests. Users can discover Kirha while learning, and we get to experiment with social farming mechanics focused on AI education.

Team & Community

Discord & Telegram Launch

December marked the official launch of our Discord and Telegram communities.

In just a few weeks, several hundred members joined. The early group is small but engaged. People sharing workflows, asking questions, and building alongside us.

This is becoming a meeting point for AI builders who want to experiment with real-world data access.

Starting in January, we’re introducing a structured reward system. Members who contribute, engage, and build will earn recognition and incentives. The objective is simple: reward the community for the value they bring.

Closing Thoughts

2025 was the year AI agents transformed software engineering by deeply integrating with the profession’s tools: code editors, terminals, and developer platforms. It was also the birth year of Kirha. In six months, a team of three with a small pre-seed check went from concept to production-grade search infrastructure. From early demos to enterprise deployments. From zero to a real community. This transformation of the field enabled us to ship MCPs at the speed of light. We developed our agentic coding stack to deploy a new server from an OpenAPI spec with a single Claude command.

I believe there are three reasons why AI came for software engineering first:

• AI is built by software engineers
• Software engineering tools are open source (VS Code versus Excel)
• Software engineering is easy to verify, which makes reinforcement learning (RL) effective

In 2026, AI will disrupt other professions. The most verifiable ones will come first, because you cannot reward what you cannot verify.

Kirha positions itself as Context as a Service (CaaS) for AI agents and is committed to meeting them where they are. From spreadsheets to code editors, from n8n workflows to custom integrations.

As LLMs continue to evolve toward a general form of intelligence, what will matter most is the quality of the context they are able to fetch. When it comes to external context, it will come from premium data providers who have already done the hard work: domain aggregation and real-time synchronization.

We unlock this context for your AI agents on a usage-based model, laying the foundations of the agentic economy.

Thanks for building with us. See you next year 🎉