One provider connection
OAuth is centralized instead of reimplemented per feature, so revocation and audit stay coherent.
Multi-agent graphs across chat, voice, phone, email, calendar, and enterprise workflows. Self-hostable end-to-end. Built in Vienna.
Widget, voice, phone, webhook — same graph, same agents, same memory. Switch channels without losing context.
Or no channel at all — drive ChatFlow as a pure REST + SSE backend, with full KB CRUD and push APIs. Build mobile, embed, headless.
USER: Can I book an appointment for tomorrow?
Connections gives ChatFlow one OAuth, refresh, revocation, and audit substrate for provider accounts. Email, calendar, company cloud files, and attached MCP servers are no longer side features: they can trigger runs, carry attachments, send replies, book meetings, sync drives into knowledge bases, and feed governed workflows.
Microsoft 365, Google Workspace, SMTP/IMAP, transactional mail.
Connect, refresh, revoke, audit — one place for provider accounts.
Inbound events become trigger payloads, run state, tools, and graph steps.
Reply, book, attach, promote documents, or call enterprise APIs.
OAuth is centralized instead of reimplemented per feature, so revocation and audit stay coherent.
Inbound mail, outbound replies, threading, and attachment handling share the trigger pipeline.
Calendar events and booking actions run through provider scopes, guardrails, idempotency, and scheduling checks.
Files from chat, email, or company cloud drives (OneDrive, SharePoint, Google Drive) enter governed document intake and KB promotion paths instead of ad-hoc storage.
Theme every color, font, and corner. Then let the agent deliver more than text — images, video, forms, detail cards, carousels, action pills, follow-ups, and inline citations all flow through the same widget, agent-driven.
Inline media in any reply — uploaded, pasted, dragged, or rendered from KBs and URLs. Lightbox preview on tap.
The agent renders the right form for the question — booking, intake, lead-qual, tickets — and reads the answer back.
Single-row, multi-row, or media-led detail views materialised from structured KBs with one retrieval.
Hero image, title, copy, action — fully themable cards rendered from structured KB rows or live tool results.
Quick replies and next-step suggestions, agent-chosen and persona-aware. One tap continues the conversation.
Suggested follow-ups under the answer, ranked by what the agent thinks the user is likely to ask next.
Both options are pet-friendly[1] and within 200 m of a tram stop[2]. Mariahilf has a south-facing balcony, Wieden adds basement storage[3].
Inline citations link back to the exact KB chunk, with a numbered reference list at the end of the answer.
10,000+ MCP servers pre-catalogued, plus a curated shortlist of Enterprise Picks — Slack, Atlassian, Salesforce, ServiceNow, Stripe and more — you connect in one click. OAuth where the vendor supports it, and every call governed by your access policies. No custom code.
Harness selection, instruction source, deterministic graph routing, run-state writes — every step inspectable. Drag the timeline to peel back the curtain.
The Node Workbench is now the canonical editor for multi-agent data flow: deterministic nodes, trigger contracts, run state, fanout, waits, approvals, and step-by-step inspection of real harness runs.
Branches, waits, approvals, transformations, and outputs sit beside agent nodes — guaranteed steps, not model guesses.
Sample payloads, strict schemas, normalized fields, and auto-injection make event-driven runs predictable.
Schema-governed state survives handoffs, waits, approvals, resumptions, and cross-run inheritance.
Operators can inspect why a harness run branched, paused, wrote state, or failed.
A tenant can run several harnesses side by side. Each one selects its own graph, instructions, triggers, memory and analytics — without forking your KBs, tools, or model registry.
Open-weight models for embeddings, reranking, and speech — plus your local LLM. Zero data leaves your network.
MCP, OpenAI-compatible endpoints, open-weight models. Export your full configuration as JSON. Leave whenever.
Vienna-based. GDPR role matrix. EU AI Act per-Article mapped. DACH VAT and reverse-charge built in.
Concrete features your DPO and procurement team can audit — not a marketing claim.
Subject-rights endpoints (Art. 15 / 16 / 17 / 20) implemented in product, with audited purge and export.
Article-by-article feature mapping documented inside the platform.
Per-tenant action permissions over every tool, connection, and MCP server — allow, require approval, or deny — with a filterable, exportable audit of what agents actually did.
Automated retention windows — 30 / 90 / 180 / 365 days, per-tenant configurable.
EU-only deployment, hosted in Vienna. GoodGuys GmbH (Austria). EU invoice with valid VAT ID.
Production can stay Docker Compose, but the runtime is now organized through deployment profiles, encrypted platform secrets, deploy preflights, Admin Deployment visibility, and a Kubernetes/OpenShift handoff path for enterprise environments.
Production deploys run through one guarded operator path — preflight checks and health verification included.
Platform secrets are encrypted end-to-end and never sit in plain config files on a host.
Docker Compose today; Kubernetes and OpenShift adapters follow the same deployment profile model.
Databases, caches, search indexes, and object storage can run as managed enterprise services instead of bundled containers.
Operators can see profile, secret freshness, health, and readiness from the Admin Deployment page.
True multi-tenancy — full white-label rebrand (name, logo, colors, email identity), per-tenant branding, API keys, model registry, billing with EU VAT. Cross-tenant Platform mode for you (with audited impersonation), isolated worlds for them.
ChatFlow's predecessor, AI-Concierge, has served tourists, citizens, and patients across Basel, Zurich, Salzburg, Vienna, Villach and the Dolomites — every day, in six languages.
Talk to ChatFlow about ChatFlow. The voice stack, live.
Live demo. Audio is processed in real time. 3-minute cap. privacy.
Contextual retrieval, multi-agent orchestration and self-hostable voice — the real mechanics behind ChatFlow, written up in plain language.
Walk through your exact use case with us, live in the product. No slides, no sales deck — just your stack against ours, and honest answers.