You don't need another chatbot. You need the process to run.
A chat chokes when you ask it to run a business process. AI Flock is an engine of specialized agents that automates your processes end to end — in business language, no coding.
30 min · With our tech team · Technical session
It's not just a chatbot. It's a system of agents and a process engine.
The difference between chatting with an AI and having one of your business processes run on its own.
A chat chokes. A flock does not.
Ask a chat to run a complex business process and it gets lost: endless instructions, copy-paste, manual files. AI Flock splits the work across specialized agents that collaborate, correct each other and share memory.
It works at scale, in production.
Anyone can build a demo RAG. Making it accurate across hundreds of GB of real documentation, without making up data, showing the source consulted in every answer — that took us years to refine. Quality and determinism, not a demo.
No development project.
It's not a months-long custom build you can't touch afterwards. You describe the process in business language and it gets built — in days, not months. Connectors to your ERP, separately.
Flexible architecture that connects with your software, lets you define processes, and is secure with your data
Layer 1 · User interaction
You talk to Flock normally. The Broker understands your intent and routes it to the right flow.
What you can do
Ask
"What margin did the North project have?"
answer with citation (RAG)
Create a process
"Draft the proposal for ACME"
you design it and it runs
Run processes
"Run May's accounting close"
a trigger launches it
Onboarding
"Tell me what your company does"
proposes your agents
Modify agents
"Add VAT to the accounting agent"
adjust them by talking
Agents that work for you
Human in the Loop
Humans stay in control, always
Layer 2 · The AI Flock engine
The Flock of Specialized Agents
input from Layer 1
Routing
Planning
Orchestration & execution
Agent Builder defines each agent
Capabilities
result → Layer 1 (with approval)
Governance & platform
spans every layer
Agent Guardrails
access control · stop
Live Agent View
live observability
Multi-model + fallback
Claude · GPT · Bedrock
Persistent memory
per-thread history
Scheduled execution
daemon · triggers
BPM Engine + R&D with University of Seville & NASA
Layer 3 · Engine connections
The engine connects to your tools and data, and runs on the best AI models.
Your tools & data
Native connectors, no code
AI models
Open API
Any of them on-prem or open source —including the leading open-weight models (Llama, DeepSeek, Qwen)— for maximum privacy.
Universal No-Code Connection
AI Flock natively integrates with the tools you already use (Microsoft 365, HubSpot, Holded…) without any programming required. It speeds up implementation and reduces dependency on technical teams.
Flexible, Tailor-Made Processes
Define and adjust complex workflows using natural language. The platform allows you to adapt automation to how your business actually operates, not the other way around.
Security & Model Freedom
Your data is secure. AI Flock doesn't lock you into a specific AI: you can use local models for maximum privacy, commercial ones (OpenAI, Claude) via API, or within your own cloud environment (Azure, Vertex).
Inside the engine
The subsystems that turn 20 years of research into product. Real components, in production.
Every subsystem comes from in-house R&D — not from wiring up third-party tools. It is built on 20+ years of research by the founders and their group at the University of Seville in multi-agent systems, BPM and process mining.
Agent Builder
Build and improve agents by talking
Describe what you need in natural language and AI Flock builds the agent: connect its tools, define what it can do and refine it by chatting — no code. When your process changes, you adjust it in seconds.
Basis: no-code agent engineering
Name
Quote Agent
Instruction · natural language
“Draft quotes using HubSpot and Holded pricing, and email them after my approval.”
Connected tools
Visual plan manager
Review and approve the plan before it runs
Before anything runs, you see the plan as a canvas of steps: which agents take part, in what order and with which tools. You review, adjust and approve it. It is human control at the plan level, not just each individual action.
Basis: BPM / PPINOT + HITL
Agents
Sales
researches the lead
Data
generates the proposal
sends after approval
Agent process orchestration
The brain that organizes the flock
A router decides which agent steps in, a planner lays out the plan, and an executor walks it step by step — passing each agent’s result to the next and keeping resources used in the right order. Not one agent blindly calling another: process orchestration, the lab’s core research line.
Base: multi-agent systems (thesis) + BPM · PPINOT/RALph
Knowledge engine (RAG)
Your private knowledge base
Indexes your documentation (SharePoint included) and queries it by meaning, not keyword match. Designed to return figures exactly as they appear in the document — dimensions, tolerances, concentrations, legal articles — without rewording them, with the source in view. 100% on European infrastructure.
Base: RAG with traceability · pgvector in production
Agent access control & security
Each agent only does what you authorize
Every agent can only read and run what you allow: tools, data and actions are scoped by role. A central log captures any incident, and you can stop an execution at any time.
Base: Conformance Checking · GDPR/AI Act by design
No-code connectors
Connect your tools without code
connect in 1 click, no code
Connect AI Flock to your tools without writing code: authorize with one click (Microsoft 365, HubSpot, Holded; Gmail and Google Drive coming soon) and agents can work on them. For your own software, we build a custom integration.
Base: integration layer · OAuth + credential vault
Token-efficient connections
No runaway token spend
MCP-style connections load context with no control and drive token spend up. Our tooling layer brings only what each step needs: in our latest architecture review we cut the cost per decision from ~2,500 to ~500–800 tokens. Predictable cost as the process grows.
Base: cost optimization iPricings/CSOP · CAiSE 2025
Document generator (PDF/Word)
Deliverable reports, no token waste
without going back through the AI
Generates ready-to-deliver Word and PDF reports. The key: the model only provides content; a dedicated engine lays out the document without going back through the AI, so you don’t waste tokens on formatting. Long documents are built in fragments.
Base: cost efficiency · proven microservice (115 tests)
Cognitive onboarding
Understands your company and proposes the agents
Before configuring anything, it understands your operation: in two phases it finds where you lose the most time, proposes which agents would give you maximum automation, and leaves their work plans ready. Onboarding stops depending on filling forms.
Base: process discovery / process mining
Real-time agent view
Watch the flock work
See the flock at work. Each agent appears with its name and the task it is doing right now, live. Automation stops being a black box: you know who did what and in what order.
Base: operational traceability
And also, built in
For developers and consultancies
Want to build on the AI Flock Engine?
API access, white-label and a dedicated instance to create your own AI solutions on the AI Flock Engine: European and compliant.
Where AI Flock fits
We mapped the AI-platform market on two axes: from chat to business processes, and from coding to business language.
Chat / assistant ——— Business process automation →
▲ No-code · business language · ▼ Needs development · technical detail
Each category solves one part. Only AI Flock sits in the top-right corner: it automates business processes in business language, no coding.
Every execution, audited. Designed for the EU AI Act obligations.
When a plan runs, every step is logged: which agent did each task, with what result and when. And each execution is classified into an N0–N4 risk level —assessed when the plan launches— based on whether it touches personal data, sends it to third parties or leaves the EEA. This is the audit trail the EU AI Act will ask for; a chatbot does not give you that.
N0–N4 scale: from "no personal data" (N0) to "external send of sensitive data or outside the EEA" (N4).
Detail by agent
Documents (2)
Analysis (1)
Communication (1)
You are not captive to the AI model
You never get locked into an AI vendor. Switch models whenever you need to and move to a European or local model —without rebuilding anything—. The value is in the platform, not the model: if regulation, the market or the vendor shifts, your operation is unaffected.
Model-agnostic by design · European infrastructure (Hetzner) · models on European servers.
Security & Compliance
Built for European regulation from day one.
Infrastructure on Hetzner (Germany), ISO 27001:2022 certified.
Models and your data run on European servers: Claude on AWS Bedrock (EU), Azure OpenAI (EU) and Hetzner (Germany). They don't train on your data.
Data processed and stored in Europe. No transfers to the US.
Human oversight, traceability and FRIA by design.
You confirm the plan before it runs and emails stay in your drafts.
Proprietary RAG infrastructure with Hybrid Search and cutting-edge architecture.
Each agent handles a bounded problem with specific context. Architecture designed to maximize precision.
Why act now
The competitive advantage of AI automation has an expiration date.
Your competitors use AI too
- The difference is no longer using AI, but having it work on your processes and your documents.
- Every month without agents is repeated work your team keeps doing by hand.
You correct it once
- Every correction stays in the agent: you don't have to explain it again.
- Repeated work becomes a process that runs with your approval.
The AI Act enters into force
- From August 2025, non-compliant AI systems face fines of up to 7% of global turnover.
- Building compliant from day 1 is cheaper than retrofitting.
- AI Flock is designed for the AI Act and GDPR-native from day one.
Source: EU AI Act, Article 99
Technical FAQ
Want to see how it works with your processes?
30 min · With our tech team · Written summary the next day
See pricing