Skip to content

FactFlow AI (Newtral)

PaidInstitutional

Vendor: newtral.es | Type: Non-detector | Cost: Paid

TL;DR

Telegram-focused AI tool from Newtral / Animal Político that extracts and maps misinformation patterns across text, audio, and video Telegram messages, with documented processing of 10 million Telegram messages by Animal Político that compressed detection time from 45 minutes to one. The Telegram-specialist entry in 2C.1 for institutional partners with deep Telegram monitoring requirements.

What it does

FactFlow AI ingests Telegram channel and group data and produces pattern-extraction output across text, audio, and video, surfacing recurring claim formulations, amplification clusters, and content copy across channels. The underlying model layer wraps Qwen-class multilingual capability for content recognition and extraction, and the operational headline finding is the Animal Político deployment that processed 10 million Telegram messages and reduced detection time on misinformation patterns from approximately 45 minutes to approximately one minute.

The tool fits 2C.1 because in several focus countries Telegram is a distinct disinformation surface that does not appear cleanly on the broader cross-platform tools. Where Information Tracer covers multiple platforms with breadth, FactFlow AI covers Telegram with depth. For institutional partners running election-cycle monitoring or sustained influence-operation investigations on Telegram-heavy information environments, the depth advantage is the operational case for the tool.

When to use it

  • An institutional partner is monitoring Telegram channel networks during an election cycle and needs the depth of pattern-extraction across text, audio, and video that broader cross-platform tools do not match for Telegram specifically.
  • A regional civil-society lab is building a sustained misinformation-pattern dataset on Telegram-mediated narratives and needs an automated pipeline rather than manual channel-by- channel review.
  • A research partner is replicating the Animal Político 10M-message workflow on a regional dataset and needs an open-source extension built specifically for Telegram-volume processing.
  • A frontline newsroom is not the target audience; Telegram is a niche surface in most of the toolkit's focus countries, and for breadth-first monitoring Information Tracer or Media Cloud is the right entry.

Limitations

Limitations

  • Telegram-specific. The tool's design focus is Telegram channel and group data; for Facebook, X, TikTok, or LINE-mediated narratives this is not the right entry point. Cross-link to Information Tracer for cross-platform breadth, to Media Cloud for news-side coverage, and to Sinar Project iMAP for SEA-specific network monitoring.
  • English-language interface; the model layer is multilingual (Qwen-based) so SEA-language Telegram content can be processed, but no SEA-specific deployment case is on record; the documented Animal Político deployment is Spanish-language Mexican content.
  • Telegram's role as a primary disinformation surface is uneven across the six focus countries. Indonesia, Philippines, Thailand: secondary surface. Malaysia, Sri Lanka, Laos: lower penetration. The cost-benefit for adopting FactFlow AI shifts with that distribution; for a frontline newsroom outside Telegram-heavy work, the tool is a niche fit.
  • Pattern-extraction output is behavioural signal, not a verdict on claim truth. Per Anchor 2 the pattern signal combines with a claim-extraction or fact-check signal from 2B before any publishable assertion about the operation.

Privacy and threat model

FactFlow AI ingests Telegram channel and group data; the privacy concentration depends on how the analyst obtained the data. Public channel data is open by Telegram's design; private group data requires either manual joining of the group or access through Telegram's data export, with all the source-protection considerations that pattern carries (identification of the analyst inside the group; risk to any source already inside the group whose behaviour shifts on observation). For institutional partners the standard data-handling regime should govern Telegram-data collection and downstream processing; the tool itself sits at the processing layer and does not change the upstream collection threat model.

For surveillance-environment work in Sri Lanka or Laos, an analyst joining a Telegram group from a personal device or attributable account can itself be a flagged action. Operational tradecraft (separate device, dedicated account, documented purpose) belongs upstream of the FactFlow AI processing step, not at the tool layer.

Country and platform applicability

  • Indonesia: Telegram is a secondary surface in the Indonesian information environment. Where institutional partners are monitoring specific Telegram channel networks, the tool applies; for breadth-first Indonesian monitoring it is not the primary 2C.1 entry.
  • Laos: Telegram penetration is lower than Facebook in Laos; the tool's relevance in Lao operational contexts is correspondingly limited.
  • Malaysia: Telegram is a secondary surface; for Malaysian work the operational pairing is more likely to involve Sinar Project iMAP and Sebenarnya AIFA on the network and tipline layers.
  • Philippines: Telegram is a secondary surface; primary monitoring runs through Facebook and Messenger. The tool applies for institutional partners covering Telegram-specific channel networks.
  • Sri Lanka: Telegram penetration is lower in Sri Lanka; the tool's relevance is limited to specific operational cases.
  • Thailand: Telegram is a secondary surface in Thailand; LINE is the dominant private-channel surface, which this tool does not address. For LINE-specific work Cofact Thailand is the LINE-native intake; for Telegram-specific work FactFlow AI applies where the case warrants.

Platform applicability: Telegram only. Other platforms route through the breadth-first entries elsewhere in 2C.1.

How to access

Open-source extension; access through the Newtral / Animal Político open-source repository. Institutional deployment requires the analyst infrastructure to ingest Telegram data at scale; the Animal Político 10M-message pipeline is the documented reference deployment scale. Smaller-scale use is possible on the same code base; setup is documented through the project's repository.

Cost (current as of 2026-05)

Open-source extension framing . Source code is freely available; the practical cost for institutional deployment sits with the analyst infrastructure and ongoing operational time rather than with the software licence. The toolkit lists the tool at the paid-mid tier in its broader 2C.1 cost banding because Telegram-monitoring at meaningful scale is operationally institutional-tier work even when the underlying software is free; smaller-scale academic use is closer to the free-academic banding of Media Cloud.

Quickstart

  1. Clone the FactFlow AI open-source repository per the project's documentation.
  2. Configure the Telegram data-ingest layer: which channels and groups, with what authentication and access, with what rate-limit handling.
  3. Run the pattern-extraction pipeline against the ingested dataset; expect compute scaling to grow with the data volume per the Animal Político 10M-message reference scale.
  4. Inspect the extracted pattern output; cluster by recurring formulations, amplification cadences, and content-copy relationships across channels.
  5. Combine the FactFlow AI output with a claim-extraction or fact-check signal from 2B before any publishable assertion about the operation.
  6. For institutional reporting, codify the pattern signature under ABCDE (Distribution axis) or DISARM (TTP catalogue) per the operation's institutional reporting target.

In the toolkit's workflow

Behaviour-pillar non-detector tool per Architectural Anchor 1. Sits in 2C.1 Automated claim monitoring as the Telegram-specialist alternative alongside Information Tracer (paid mid-tier cross-platform tracer), Media Cloud (free academic news-source corpus), and Sinar Project iMAP (free open-source SEA-specific network-monitoring). Of the four 2C.1 entries, FactFlow AI is the platform-specialist for Telegram specifically.

Standard combinations:

  • With Information Tracer at 2C.1 for the breadth layer that pairs with FactFlow AI's Telegram depth.
  • With CooRTweet at 1C.1 for IFCN-signatory partners running coordinated-link-sharing analysis on Telegram exports alongside the FactFlow AI pattern-extraction pass.
  • With CIB Mango Tree at 1C.1 for the interactive CIB-detection layer on Telegram networks.
  • With Gephi at 2C.2 for publication-grade visualisation of FactFlow AI's amplification-cluster output.
  • With ABCDE Framework and DISARM at 2C.3 for codifying the pattern signature in respectively public-facing and institutional reporting.

Decision-tree references: T5 escalation routes through FactFlow AI at T5.13 (contextual OSINT) when the operation is Telegram-mediated at scale; the same 2C.1 platform- specialist depth feeds back into T7 tipline routing when a Telegram-heavy case warrants public response.

This card carries no detector signal class: FactFlow AI produces a non-detector pattern-extraction behavioural signal on Telegram data. Per Anchor 2 that signal combines with claim-extraction (2B.2), CIB analysis (1C.1), or source-history (1B.1) to support a publishable institutional finding.

Override notes

Override notes

  • Telegram platform-specific pointer (telegram-platform-specific-pointer): the tool is a Telegram specialist; readers needing breadth-first cross-platform monitoring should reach for Information Tracer instead. The Limitations admonition, Country and platform applicability section, and In the toolkit's workflow section together carry this pointer consistently.

Sources

  • Newtral / Animal Político. FactFlow AI — Telegram monitoring and claim-extraction extension. Newtral, 2026 (open-source). github.com/newtral-tech
  • Animal Político. Animal Político — investigative journalism outlet (documented FactFlow AI deployment). animalpolitico.com.