Resources · Docs, Academy, Hub & Blog
Everything you need to build with FlowX.
Product documentation, hands-on training, and the latest thinking on shipping mission-critical AI agents in regulated enterprises.
3 resource hubs24 articles15 topics● updated weekly
Documentation, training, and guides.
Start with the reference, go deep in the Academy, and browse the Knowledge Hub by topic.
Documentation
Guides, API references, and platform concepts — everything to design, build, integrate, and run agents on FlowX.
docs.flowx.ai trainingAcademy
Structured courses and certifications that take teams from first agent to production, at their own pace.
academy.flowx.ai hubKnowledge Hub
Product deep-dives and answers to your questions, filterable by topic across every resource type.
/hubLatest articles and insights.
Field notes on agentic AI, governance, and integration from the team building it.
Speed Without Sacrifice: Agentic AI in Insurance Claims
Faster claims processing does not require loosening regulatory controls. Four production workflows — FNOL triage, adjudication, subrogation, and compliance reporting — with 80% faster turnaround, 75% fewer errors, and €0.7M in projected annual savings.
Read the articleOur Voice Agent: Turn-Taking
We put an Agent Builder agent on the phone for live calls in several languages. The answers were rarely the problem; knowing when the caller had finished was. A reply cut off after 160 ms and five seconds of silence: what broke, and what we changed.
Jev: Good, Bad, Ugly
We benchmarked a typed classifier against a chat model on 22 labeled states and 106 judgments. Jev tied on detection at a tenth of the cost, but fell behind on rubric scoring. Repeating 21 questions ten times each revealed why stable verdicts and high confidence are no guarantee of correctness.
Computer Use Without the Pixels
An LLM agent reads IBM 3270 screens without images. Six single-step tasks, a text-versus-vision comparison, and lessons from building a TN3270 harness.
Browser Agents Fail on Consistency
We tested nine browser-agent models across 405 runs. Five passed all 45 of their runs, but other agents reported success after failures or extra actions. The findings show why completion needs independent verification. The browser node ships with FlowX.AI 6, in preview now and generally available this fall.
Sovereign AI, Open Guardrails
Sovereign AI is every check on the way in and out of a model running where you run it, on weights you can read, with evidence you keep. When LLM Guard went read-only on July 9, 2026, a year after the category’s three best-known detection stacks were acquired into closed platforms, we published border: two functions, 30 detectors, 26 languages scored one at a time, CPU-only, offline after the first download, an evidence record for every scan. Apache-2.0, open weights, pip install flowx-border.
Intelligence + CL = Expertise
Every vendor buys the same frontier models by the token, so raw intelligence is a commodity that resets to zero on every request. What compounds is expertise, and it only comes from systems that learn from their own production traffic. The thesis, the two FlowX.AI papers behind it (SIFT and ORNA), and the research those papers stand on.
FlowX.AI on Gemini Enterprise
FlowX.AI is among the first partners bringing specialized industry agents to Gemini Enterprise, available through Google Cloud Marketplace and featured in the Gemini Enterprise for Financial Services launch.
Override by Design
Human-in-the-loop is not a default setting, it is a structural commitment. Genuine oversight in regulated agentic AI takes four integrated mechanisms, deliberate decision-surface design, and escalation logic built into the workflow engine itself.
Compliance Built the Case
Compliance complexity isn't a reason to delay agentic AI in banking — it's the condition that makes the case for it, proven in production across KYC/AML, lending, and regulatory reporting.
We Built an ROI Calculator
Why measuring impact matters — and a way to project the return before you build.
FlowX.AI 6 Release Summary
The questions every enterprise is asking about their agentic AI roadmap — answered in one release: real use cases, measurable ROI, zero-hallucination guardrails, and recurring self-improvement at scale.
The Day Onboarding Stopped Being a Loop
Turning commercial onboarding from an endless back-and-forth into a single agentic flow.
The Five Tests of Mission-Critical AI
Pragmatic standards for judging whether an AI system is ready for real, regulated work.
NTT DATA and FlowX.AI partner to move agentic AI into mission-critical operations
Enterprise AI has a production problem. A platform built for production meets a transformation partner built for scale.
The Control Deficit: Why AI Copilots Fail the Moment They Touch Real Work
The gap isn't intelligence. It's controllability: evidence, identity, oversight, and reliability under stress.
Copilots Raise Productivity. Operating Models Create Outcomes
Copilots alone are incomplete — outcomes need an operating model built around them.
How a large European bank increased their daily underwriting cases throughput by 600%
Corporate & SME underwriting turnaround cut from 2–6 weeks to under 7 days — throughput up from under 5 to 35+ cases a day, error rates down to under 3%.
We already have Copilot — and why that’s not the same as mission-critical AI
Copilots raise productivity. Mission-critical AI is an operating model — five tests that separate AI you can demo from AI you can deploy.
Your AI Strategy Is Only as Strong as Your Integration Layer
Where your AI gets its data decides how far your strategy can actually go.
Engines vs. Railroads: Why AI Agents Stall in Regulated Value Streams
The barrier isn't model capability — it's the rails the agents have to run on.
Turn AI Potential into Production Reality with the Integration Designer
Cutting the timeline from API spec to a live, production integration.
Bridging the Operational Divide: What Banking Can Learn From Process-Heavy Industries
Risk and compliance lessons banking can borrow from industries built on process.
Your mainframe isn’t legacy — the way you are using it is
Nick Donofrio (ex-IBM EVP of Innovation & Technology) on why the mainframe isn’t the bottleneck — the access pattern is.