Agents can use it.
People can use it.
Your organization can govern it.
TongFlow is the Agent-Human-Friendly, Modality-First, Enterprise-Level GenAI Workflow Platform. Text, images, audio, video, 3D, documents and links connect on one canvas into workflows you can run again and again.
AI is already at work. Is it really being used?
- AI agents
Nobody can say how it got there.
→ Agent-Friendly - Frontline staff
I ask one question in a chat box and get one answer back.
→ Human-Friendly - Business material
Images, text, audio and video each live in their own tool.
→ Modality-First - Managers
Who is using it, what it costs, where the data goes: nobody can say.
→ Enterprise-Level
People use AI on their own. The organization doesn't yet.
AI has taken hold with individuals, but it hasn't become something the organization can do. What works stays inside one person's chats, and a new person or a new batch of material means starting over.
Where TongFlow sitsBetween models and the business, the missing layer is a generative AI platform.
- Applications Business systems and services
- Generative AI platform
- AI models Text · image · audio · video · 3D
- Infrastructure Cloud and data centers
- Chips GPUs and other compute
- Energy Power
Four strengths that only work together.
Friendly to people alone, and agents can't get in. Friendly to agents alone, and staff can't use it. Text only, and the business doesn't connect. Without enterprise controls, the more people use it, the messier it gets.
- Agent-Friendly
Agent-Friendly
Agents work inside a workflow, and every step is recorded.
Agent output you can check and trace - Human-Friendly
Human-Friendly
People who don't write code can turn what they know into a workflow.
Your staff put AI to real use - Modality-First
Modality-First
Start from the material and the result. Choose the model last.
Change the model, keep the workflow - Enterprise-Level
Enterprise-Level
See it, manage it, look it up later.
AI your organization can govern
Agent-Friendly
Agents work inside a workflow, and every step is recorded.
“Nobody can say how it got there.”
What agents do stays on record.
- 01
Agents and people build with the same tools
A local Model Context Protocol (MCP) server and a Claude Code plugin give agents the same tools as the canvas.
- 02
The output is a workflow file
Not a one-off answer: people can open it on the canvas, edit it and run it again.
- 03
Each step leaves a record
Each run is kept in the run history. Before a paid step runs, the agent checks with a person first.
Also as custom delivery: human sign-off on key steps, and a remote MCP server.
Result Agent output you can check and trace

Human-Friendly
People who don't write code can turn what they know into a workflow.
“I ask one question in a chat box and get one answer back.”
Using AI well shouldn't be for a few people.
- 01
Three actions: add, transform, combine
Add your material, transform it into the form you need, and combine several pieces.
- 02
Each step's output stays on the canvas
Pick up from an intermediate result and keep going.
- 03
Workflows are reusable
Run them again on new material, or hand them to a colleague.
Result Your staff put AI to real use

Modality-First
Start from the material and the result. Choose the model last.
“Images, text, audio and video each live in their own tool.”
The business is multimodal. One workflow should carry it.
- 01
Seven modalities on one canvas
Text, image, audio, video, 3D, document and link.
- 02
Capabilities are separate from models
Swap the implementation of a capability, and the steps around it stay as they are.
- 03
60+ capability slots, 50+ official plugins Own compute: custom delivery
Connect through commercial APIs, model gateways, open models or your own compute.
Result Change the model, keep the workflow

Enterprise-Level
See it, manage it, look it up later.
“Who is using it, what it costs, where the data goes: nobody can say.”
The more people use it, the more it needs governing.
- 01
Runs in your own environment
The core is open source, so the code can be reviewed. For a private deployment, we set it up in your environment.
- 02
Accounts, permissions, spend approval, audit logs Custom delivery
Connected and configured to fit your policies.
- 03
Compute on your own contracts
Models and compute run on your own accounts, with no markup. We don't sell generation credits.
Custom delivery
Developed and delivered to your requirements as part of a private deployment.
- Account and identity integration (single sign-on, SSO)
- Role-based permissions (RBAC)
- Approval for paid runs: request, approve, run
- Run logs and audit: search and export by person, workflow and time
- MCP server (a remote server at a URL)
- Human approval of key steps
- Local models, your own compute, data kept inside your environment
Result AI your organization can govern Delivered through a private deployment.

Get one process running first.
Output: a clip from the key visual Marketing content
A campaign brief → Copy options for social, a product page and email, two key visuals and a short clip
- Product page headline
- Dimmable LED Desk Light for Remote Work
Content localization
An English product page → Japanese, Chinese and Spanish pages, and a Japanese narration script
- English (source)
- A daypack that folds into its own pocket.
- Japanese
- 本体のポケットにコンパクトに収納できるデイパック。
- Chinese
- 一款能直接收纳进自带口袋的便携双肩包。
- Spanish
- Una mochila que se pliega en su propio bolsillo.
Meetings and knowledge
A meeting recording → A transcript, then a summary, decisions and action items
- Decisions (excerpt)
- Website relaunch will go live on March 4th.
- Action item
- Daniel: complete load testing and send results by Friday.
![Slide illustration: a window with the placeholders [CUSTOMER] and [ORDER] next to a person at a laptop](/scenes/en-training-960.webp)
Output: slide illustration Training and internal communication
A policy memo → A training script, its narration and two slide illustrations
- Training script (excerpt)
- Second, never put customer personal data into AI prompts.
The example inputs are fictional. The organizations, products and people in them are not real.
Turn a finished workflow into a form.
App Mode shows only the fields to fill in. One click runs it, and the result appears below. The people who build workflows maintain them; more colleagues use them by filling in the form, even from a phone browser.

Self-hosting
Start from source or Docker and run it on your own machines. TongFlow is open source under the AGPL-3.0, an OSI-approved license. A commercial license is available.
Private deployment
We set up a dedicated workflow platform in your own environment, and deliver the custom-delivery capabilities you need.
For pricing, contact us.
- Open it in a browser or the desktop app. There is nothing to deploy.
- You use your own model API keys and compute accounts. Usage is billed to your accounts, and your data stays with you. We don't sell generation credits.
- It is for evaluation only, with no guarantees.
- It is free for 7 days. To keep using it after that, you subscribe inside the product.
From design to operation, we work alongside you.
Private deployment
A dedicated platform, set up in your own environment.
Custom nodes and plugins
Your business systems and in-house models, connected as nodes on the canvas.
AI workflow consulting
We start from where AI should be used, and work it through with you.
How we work together
- 01
Discovery
Map the work, the current setup and the security requirements.
- 02
Proof of concept (PoC)
Pick one process and test it with real material.
- 03
Deployment
Set up the environment and build the nodes you need.
- 04
Rollout
Extend it across teams and keep improving the workflows.