How it works

Syto is a system of cooperating agents that gathers and analyses pro-Russian material every day, with no manual step in between.

A multi-agent pipeline

Rather than one monolithic script, Syto runs as a set of specialised agents, each with a narrow job. An orchestrator runs them in sequence every day at 01:00 UTC (04:00 Kyiv time), passing each agent's output to the next.

Ingestion agents — one agent per source type: one reads Telegram, another the X accounts, a third the RSS feeds and outlet websites. They fetch new material from every active source and store it as immutable evidence: the original text, publish time, and reach metrics. The original is kept in full, separate from any analysis.
The analysis agents — language models read the new material, cluster it into narratives, identify the manipulation techniques used and the target audience, and update the narrative cards: trend, intensity, examples, counter-messaging.

Splitting the work across agents makes the system resilient: if one source breaks, it's an error confined to that source's agent — the rest of the pipeline still runs and refreshes the site as usual.

Time windows

The Today view is based on publications from the last 24 hours, updated daily at 01:00 UTC. Weekly and monthly views are built on top of those daily slices. Historical aggregates are never rewritten retroactively: a card reflects what was known as of its day, not later revisions.

Limitations

The analysis is produced by a language model, so it can make mistakes — including misattributing examples or paraphrasing a source loosely. We always keep the original material alongside, so every claim can be checked against the primary source. That is why each example on a card carries a link to an archived copy of the post it paraphrases; where a platform cannot be archived, we print the post’s address as plain text instead of linking it, so that checking our work sends no traffic to the source. Treat conclusions critically and verify anything important yourself.