Tavora

About

We're building the substrate for code-reasoning agents.

Tavora is a pre-MVP product that bets agents will reason as code, not as function calls — and that the application layer above the agent will get a lot thinner. This page is the long version: our founding thesis, the founder, and how we got here.

Founding thesis

Four claims we build against.

These four propositions are the why under every Tavora product decision. They're interdependent: removing any one collapses the others.

  1. 01

    Agents are first-class consumers of data — not visitors to applications designed for humans.

    REST endpoints built for human forms, paginated list views, click-to-confirm flows are the wrong shape for an agent. The data layer should be designed with agents as a primary audience, not retrofitted from a UI.

  2. 02

    The right interface between agent and data is a typed, queryable API the agent writes against in a real programming language — not a fixed set of function calls.

    A real language lets the agent compose, branch, transform, and join in one reasoning step instead of stitching together pre-declared functions. Tavora's think primitive runs JavaScript in a sandbox; the program is the plan.

  3. 03

    An agent's reasoning is a multi-step process with persistent state — not a chain of single-turn function calls where the LLM has to re-derive its working set from chat history each turn.

    Inside a Tavora think block and across them, JS variables, fetched data, and intermediate results persist. The LLM composes against state, instead of reconstructing it.

  4. 04

    The application layer becomes thinner as the substrate gets richer.

    Many "applications" reduce to a memory schema, a few agents, and a delivery mechanism. The orchestration, judging, project isolation, and audit that used to be the app team's problem move into the substrate.

Founder

Jürgen Ryannel

AI consultant, software architect, founder

Software engineer turned AI consultant. Co-founder of Pelagicore (the platform behind Mercedes-Benz MBUX, acquired by Luxoft) and ApiGear (acquired by Epic Games). Most recently Senior Engineer at Epic Games, building AI agent platforms.

Currently building two products in parallel: Valiro for construction and engineering, and Tavora for production AI agents. The thread running through both is the same conviction — that real adoption comes from evals, not demos, and that good platform architecture is the prerequisite to either.

Based in the DACH region. GDPR is a starting condition, not an afterthought.

juergenryannel.com → LinkedIn → GitHub →

Why we're building this

The shape of an AI product team in 2026 is different.

Teams shipping AI into their product today burn a quarter of an engineering year on the same four things: a runtime for the agent loop, an eval harness that doesn't lie, per-project isolation, and an audit story that survives the first compliance question. None of that is the customer-facing feature. None of it differentiates.

Tavora is the layer that absorbs those four problems so the app team can spend the quarter on the agent itself — its persona, its skills, the workflows that actually move the customer's needle. Code-first authoring (the tavora/ folder) is what makes that real: the agent definition lives in your repo, alongside the rest of your code, where Cursor and Claude Code already know how to edit it.

We're early — pre-MVP, no paying customers yet. That's why this page exists: if any of the four thesis claims above resonate, we'd like to talk before the product gets pinned down.

See a real agent — Trello-style demo on GitHub →

Drop an agent into your SaaS, in 5 minutes.

Early access — limited spots while we onboard the first wave. Tell us about your app or grab a 30-minute slot.

Get early access Speak to us →