Tuturama is an AI-native software factory. We build products, embed with security teams inside large institutions, and contribute to the code the world runs on.
We work embedded, forward-deployed in the client's own team, with a mandate on security and artificial intelligence. The advice comes with the engineering that follows it: threat models for AI adoption, architecture reviews, tooling the team keeps after we leave.We also install the operating model we run on inside a team: roles as skills, gates before shipping, memory that outlives the session. Two to four weeks. Your people keep it.
Right now that seat is inside the CISO Office of a European retail group. Over twenty years the same seat has been inside Pearson, Société Générale, Walmart, Banco do Brasil and Kraken, always where new technology meets old structure.
Radio from the whole planet, in your menu bar and in your pocket.
A world radio player with a map, favourites that survive catalogue changes, and a curated station catalogue with programme schedules. Its cognitive layer turns listening into learning: vocabulary, sessions and goals sit next to the stations. The name comes from skywave propagation, the ionospheric bounce that lets a signal from Tokyo reach Tallinn. macOS and Android are live. iOS is in review.
Audio intelligence for radio.
Continuous listening to the broadcast, turned into quota compliance reports, proof of airplay and market intelligence. The cognitive layer is the listening itself: recognising speech, music and language on the live signal, hour after hour. Built for regulators, rights societies and stations, starting with regional languages that mainstream speech recognition ignores. Piloting in Europe.
A legal companion that lives inside your coding agent.
While the project is being born, Ulpiano reads the repository and flags trademark, privacy, licence and tax issues, records each risk next to the code, and calls a real lawyer when it is time. Reading code the way a lawyer reads a file is its cognitive layer. It suggests. It never issues a legal opinion.
The factory's own output. It is how we know the method works. More on the products
Six years, three threads. Each one starts in the studio years and ends in something shipping now.
From explaining by ear to machines that listen.
2023
Tuturama Studios
Story-driven audio courses on markets and money: Wonderville, Dojima.
2024
Icarus
An audio series on the Internet Computer, released as a podcast on Spotify.
2026
World radio on macOS and Android. Listening turned into learning.
2026
Continuous listening to the broadcast, turned into compliance reports and proof of airplay. Piloting.
Prove it without revealing it.
2024
Tuturama Labs
The studio's technical arm: tokenization and zero-knowledge development behind the courses.
2025
Zero-knowledge proof demos and zkvox, a ZK voting prototype, public on GitHub.
2025
A zero-knowledge canister on the Internet Computer.
2026
Zero-knowledge gated x402 access for autonomous agents on Stellar.
Agents that pay and browse, and the guards around them.
Apr 2026
Solana payment scheme merged into the Rust implementation.
Aug 2026
For an MCP server that lets an agent pay over x402 on Stellar. Six primitives, in phases.
Aug 2026
Payment guard
Policy before the signature and a signed audit log, proven end to end on Stellar testnet.
Sep 2026
Four audits as skills for coding agents, with a runnable verifier. MIT.
Multiple commits in mainline (staging, IIO drivers), reviewed by maintainers at Intel and Huawei. More in the IIO queue.
Fix for SkipWhitespace returning an early EOF on whitespace-only chunks.
Merged PRs: automated nix flake updates, deprecation warnings on hidden CLI commands, unwrap to expect in non-test code.
Capture ResponseFailed errors in OpenAI stream mode. Released in rig-core 0.34.1.
Trace option consolidation, authored commit merged through the maintainers' PR.
Solana exact payment scheme for the Rust implementation, with integration tests and examples.
Every item links to the maintainers' review. Nothing here is self-published. The work is continuous; the live state is on GitHub. github.com/gabrielrondon · Full list, including what is in review
Agents now browse, pay and read private data on our behalf. The security around them lags behind what they can do. This is where the factory spends its own time in 2026: threat models for agentic spend, gates in agent tooling, guards for autonomous payments.
Findings go to the maintainers first. Methods go public as skills.
Research and essays by the founder, on traceability, verification, and the engineering hidden inside institutions, at gabrielrondon.com.