The Ductile AI platform
An AI worker that builds your business automation jobs, and a platform that runs them
Describe a job in plain words. AnnE builds it as a business process from 1,100+ apps and Excel-style formulas, proves it on sample data, and once you approve it, it runs the same way every time, with every run on record.
How it works
From a sentence to a job that runs
- 1
Describe the job
Type or say it in plain words, in the console or in Slack, Microsoft Teams, Google Chat or WhatsApp. Paste a job post or an email as it is.
- 2
AnnE builds it and tests it
AnnE, Ductile AI's AI worker, plans the job, builds it as a business process from tested app actions and Excel-style formulas, and runs it on realistic sample data before anything real happens.
- 3
You approve it
You see the plan in plain words and the result on sample data. Nothing runs, sends, pays or deletes until you say yes.
- 4
It runs, and every run is on record
On a schedule, a webhook, a form or an app event, the same way every time. Each run records what it read, what it wrote and when.
What is in the platform
Everything a business automation job needs
The pieces AnnE builds with, and that run your jobs. More for developers on the developers page.
AnnE, the AI worker
Reads a request or a job post, plans it, builds the process, writes the formulas and proves it on sample data. Change a job by telling her what to change.
Business processes
A job is a process: steps that read, decide, transform, write and wait for a person. Versioned, explained in plain words, identical on every run.
Excel-style formulas
IF, SUM, LOOKUP, TEXT, FILTER and over 120 functions for every mapping, filter and route, across JSON, XML, CSV, Parquet and fixed-width data.
1,100+ apps, 5,700+ actions
Ready-made actions for the apps businesses use, each tested against the vendor's API: from Shopify, QuickBooks and Slack to Salesforce, SAP and Snowflake.
Ways to start a job
Schedules, a secure webhook for every process, hosted forms, app events such as a new order or a changed ticket, and a request in chat.
Data, built in
Tables with validation and a history of every change; files in CSV, Excel, JSON, XML, Parquet and PDF; warehouses and lakes; SOAP, EDI and mainframe records.
People in the loop
Approvals for anything that sends, pays or deletes, and worklists where exceptions go to a role, not one person's inbox.
Run history and audit
Every run kept with what it read and wrote. Search it, replay it, export it; every change to a job is kept with who made it.
Governance and security
Built for jobs that have to be right
The same result, every time
An approved job runs exactly as approved. The AI builds it but never runs it, so a model update changes nothing you rely on.
Sign-ins sealed per job
Each job keeps its own app sign-ins in the vault, handed to a step only while it runs. Neither AnnE nor any model ever sees them.
Access you control
Single sign-on, roles, access by space and job, SCIM and audit export, as your plan allows.
Development to Production
Build in Development and promote to Production, with Test on Enterprise, and source control in GitHub.
Run it your way
Our cloud or your network, our models or yours
See Bring your own model and pod and AnnE in chat.
Ductile AI cloud
Nothing to install. Jobs run on Ductile AI's runners, kept apart for each customer.
Your own runners
On-premise runners inside your network, next to databases, file shares and plant systems behind the firewall.
Your own AI model
AnnE and your AI steps on the model you already pay for. Its work uses none of your plan's AnnE builds.
AnnE in chat
Ask for a job, approve it and hear back in Slack, Microsoft Teams, Google Chat or WhatsApp.
Platform architecture
A hosted console, and runners where your jobs run
The console plans, keeps and governs; runners do the work. AnnE builds a job once; from then on the approved job runs on a runner without the AI.
The console
The control plane: jobs, processes, schedules, webhooks, forms, approvals, the vault, the AI gateway and run history. Hosted by Ductile AI, the same on every plan.
AnnE
Runs in the console and builds jobs through the AI gateway, on the model chosen for each part of her work. She reads names, settings and the shape of data, never records' values or sign-ins.
Runners
Where a job's steps run, on the JVM or in Python, from the app actions in the catalogue. Each app's code is checked before a runner loads it.
The vault
Sign-ins sealed per job and handed to a step only while it runs. A sign-in made for one job is never offered to another.
The AI gateway
Every model call, AnnE's and your jobs' AI steps', goes through it: the model per organisation and per job, budgets and daily caps, vendors allowed, identifiers taken out.
The app catalogue
1,100+ apps and 5,700+ actions, each tested against the vendor's API and released to runners as a checked bundle.
Run history
Every run's record - what it read, what it wrote, what it declined and why - kept in the console, searchable and exportable.
Doors in and out
The console, Slack, Microsoft Teams, Google Chat and WhatsApp for people; a REST API, an MCP server and GitHub for code.
Your own runners
Run jobs inside your own network
For systems behind the firewall and data that must stay where it is.
Inside your network
Runners next to your databases, file shares, ERP and plant systems, for the data that must not leave the building.
They call out, nothing calls in
An on-premise runner connects out to the console over an encrypted connection; nothing has to reach into your network.
The same jobs, the same results
A job runs the same steps on your runner as in our cloud. Move it between them without rebuilding it.
By plan
An add-on on Team, one on-premise cluster included on Business, included on Enterprise. Steps of Premium apps use credits as anywhere; standard apps are free.
Bring your own model
AnnE on the AI model you already pay for
Set it up in Settings > AI gateway. Step by step on Bring your own model and pod.
Any vendor
Anthropic, Google Vertex AI, Amazon Bedrock, Microsoft Azure AI Foundry or any OpenAI-compatible service, on every plan, Free included.
A model per part of the work
Your strongest model for planning, a faster one for filling in steps, the fastest for talking - to bring the cost down.
Your key, sealed
The key you paste is sealed in the console and never shown again; a Check button proves the model answers.
No limit from us
AnnE's work on your model uses none of your plan's AnnE builds or AI allowance; your vendor bills you directly.
At a glance
The Ductile AI platform, line by line
| What it is | An AI worker, AnnE, that builds business automation jobs from plain words, and a platform that runs them the same way every time. |
|---|---|
| Apps | 1,100+ apps and 5,700+ ready-made actions, each tested against the vendor's API. |
| How a job starts | Schedules (down to every minute, by plan), webhooks, hosted forms, app events, chat. |
| Formulas | Excel-style, over 120 functions. |
| Architecture | A hosted console (the control plane: jobs, vault, AI gateway, run history) and runners where steps run, on the JVM or in Python. |
| Where jobs run | Ductile AI cloud - pooled with each run isolated, or not shared with other customers, by plan - or on-premise runners that connect out to the console. |
| AI models | Ours, included in every plan, or your own: Anthropic, Google Vertex AI, Amazon Bedrock, Microsoft Azure AI Foundry or any OpenAI-compatible service. |
| Where you work | The console, Slack, Microsoft Teams, Google Chat, WhatsApp, a REST API and an MCP server. |
| Environments | Development and Production; Test on Enterprise; GitHub source control. |
| Security | Sign-ins sealed per job; single sign-on, roles, SCIM and audit export; your records never used to train models. |
| Plans | Free to start; paid plans by seats and credits; Enterprise by contract. See pricing. |
Questions about the platform
What is the Ductile AI platform?
A platform for business automation jobs. You describe a job in plain words; AnnE, its AI worker, builds it as a business process from tested app actions and formulas, proves it on sample data, and once you approve it the platform runs it the same way every time, with every run on record.
Who is AnnE?
AnnE is Ductile AI's AI worker. She reads what you ask for, plans it, builds it, writes the formulas and tests the result before anything real runs. You change a job by telling her what to change.
Is it an AI agent that acts on my systems?
No. AnnE builds the job; the job runs as an approved business process without the AI. That is why it gives the same result every time and a model update changes nothing you rely on.
Do I need to write code?
No. Describe the job and AnnE builds it. Formulas read like Excel, and Python is there for the rare step that needs real code.
Which apps does it work with?
1,100+ apps, from Shopify, QuickBooks, Slack and HubSpot to Salesforce, SAP, Snowflake and Databricks. Browse them all on the apps page.
Where do jobs run?
In Ductile AI's cloud, or on on-premise runners inside your own network for systems behind the firewall.
Can I use my own AI model?
Yes, on every plan. Connect Anthropic, Google Vertex AI, Amazon Bedrock, Microsoft Azure AI Foundry or any OpenAI-compatible service; AnnE's work on it uses none of your plan's AnnE builds. See Bring your own model and pod.
Is my data used to train AI models?
No. Your records are never used to train anything, and your records' values and sign-ins never reach any model.