Your agents follow your rules within your environment.

Form.io's Universal Agent Gateway (UAG) sits between AI agents and your Form.io project. Agents read your forms to learn what to collect and how to behave. Every value they send back is checked against the form's own validation and saved according to the user's permissions, all within your own environment.

Or install the Form.io Agentic Coding Plugin and have your coding agent make your forms agent-ready.

npx skills add formio/ai
How it works
Self-hosted MCP native Any model Open source, MIT
YOUR SECURITY BOUNDARY CClaude OOpenAI GGemini AIYour agent Any agent or model UAG Universal Agent Gateway ① Authenticate ② Read the form ③ Validate ④ Write once ⑤ Audit MCP · Agent API Forms & rules Submissionsvalidated data Workflows Your systems ERP CRM Legacy

One form for people and agents.

Form.io's renderer turns form JSON into an interface for people. The UAG turns the same JSON into context for agents: the fields, the rules, the allowed values and the instructions.

  • Change a field or an instruction in the form builder, and agents pick it up on the UAG's next refresh.
  • People and agents are checked by the same validation rules.
  • Agents automatically find the forms you tag, so you write no per-form agent code.
{ "key": "urgency", "type": "radio", "validate": { "required": true }, "values": [ "low", "high" ] } Form JSON one source of truth Renderer UAG Urgency * Low High Submit For people ## urgency - type: choice - required: yes - allowed: "low" | "high" - ask: "How urgent is the issue?" For agents

Agents inherit your users' permissions.

A prompt can't enforce permissions, so the UAG does. Every agent request goes through the same authentication, permissions and validation as a person using your application.

Agent request tool call or API AuthenticateOAuth PKCE, your SSOor a scoped token Authorizethe caller's own rolesand permissions Validatetypes, options, rules,conditionals Write once & recordthe UAG saves, never the model,with engine, model and confidence

No model holds a write tool

Engines return answers, and the UAG validates each value and writes it once, itself. Text in a submission can't get a model to change your records.

Your RBAC, unchanged

Agents authenticate through your existing Form.io login, SAML or OIDC, and act with exactly the permissions of the person or service they represent.

A record of every decision

Each decision records its engine, model, confidence and source. With Form.io Enterprise, submission revisions show who changed what, human or agent.

Confidence thresholds

Decision engines answer closed questions with a confidence score. Below the threshold you set, the field is left for a person to answer.

Re-decided on the server

Values decided in a browser can be tampered with. An "Agent: Run task" action before the save decides again on the server, and fields marked authoritative take only its answer.

Change it without a deploy

Rewrite the instructions in the form builder and agents follow them from the UAG's next project refresh, with no restart. To undo a change, restore an earlier form revision.

Your data stays with you.

Run the UAG next to Form.io in your own cloud, data center or air-gapped environment. It needs no public endpoint, and with a local model no data leaves your network.

  • One container on Docker or Kubernetes, beside your Form.io deployment.
  • Private by default: your applications connect out to the UAG; no public endpoint, no tunnel.
  • Local models through Ollama or vLLM, or engines isolated on their own server so model keys stay out of the gateway.
  • Enterprise ready: RBAC, FIPS-compliant deployments, form and submission revisions, stage deployments.
  • Open source under the MIT license, so you can read the code that handles your data.

Deploy it in your environment

YOUR NETWORK Your applicationsportals, chat, services Form.ioforms, data, actions Universal Agent GatewayMCP · Agent API · validation · audit Engine servermodel keys stay here Local modelOllama, vLLM: no egress Model APIs optional, outbound no inbound access

The right engine for every field.

Closed choices go to a decision engine that reports its confidence. Open values go to an agent engine such as Claude. Changing models is a configuration change.

A task's fields Claim type · radio Injury? · yes / no Date of loss · date Summary · free text tagged in the form builder Decision engines closed sets · calibrated confidence · fast JEVOpenAI Decisions Agent engines open values · read-only research tools Claude OpenAI Gemini Ollama UAG validates & saves confident · uncertain · suggested people confirm the rest

With and without the UAG.

Agents wired straight to your data

  • Business rules live in prompts and drift between agents
  • Models hold write access to your systems
  • Every new form means new agent code
  • No record of which model decided what
  • Data leaves your security boundary

Agents through the UAG

  • One JSON schema governs people and agents alike
  • The UAG validates and writes; models only answer
  • New forms are discovered, not coded
  • Provenance on every decision and value
  • Self-hosted, with local models when you need them

Let your coding agent build it.

The Form.io Agentic Coding Plugin gives Claude Code, Cursor, Codex and other coding agents that read skills the Form.io skill library, including the formio-uag skill. Describe what the agent should do, and your coding agent tags and writes your forms for agents, designs their tasks, deploys the UAG and its engines, and wires your application to it.

  • Its example prompts build a chat that fills in a form, an agentic workflow and a self-filling wizard in your own project.
  • The plugin writes forms through the Form.io MCP server and shows you each change before it saves.
  • Settings that open a form to anonymous users or spend model credits are listed separately for you to confirm.
bash
# In your application's folder
npx skills add formio/ai

# Or, in Claude Code, the plugin with the Form.io MCP server
/plugin marketplace add https://github.com/formio/ai.git
/plugin install formio-ai@formio
Then ask your coding agent
Put a chat on our public support page where customers describe their problem to an AI agent. The agent then interviews the user for the necessary information and submits our "serviceRequest" form for them.

Take it to production.

Enterprises have built their data collection on Form.io for more than a decade. Ready to go live? Run the UAG in your own environment, on your infrastructure and under your control. Talk to Sales about a deployment license and support options.