Skip to main content
This is the framework that powers every Mindset AI agent on SDK 3. Your agent runs its entire response loop directly in the user’s browser: calling the LLM, executing tools, and rendering interactive UI. A secure proxy handles credentials and platform services on our infrastructure. The result is a conversation that feels immediate and behaves with awareness of the page it lives on.

​
Key capabilities

Low-latency responses

Each session loads the agent’s configuration once: prompt, tool schemas, and conversation context. Every message a user sends after that goes straight to the LLM, with no server-side setup between the user and the model.

Live interactive elements

Agents can pause mid-conversation to present choices, request a confirmation, run a quiz, or guide a sign-up flow. Because the orchestration runs locally, the UI for these interactions appears the moment the agent decides to show it. There’s no network hop between the agent’s question and the user’s click.

Continuous capability delivery

Agent behaviour is shaped by small, composable rules delivered to the SDK at session start. When we ship new capabilities (multi-step plans, user memory, policy updates, new tool types), your customers pick them up automatically on their next session. No API changes or engineering coordination required on your side.

​
How it works

Mindset AI splits the agent stack into three tiers.
1

Display tier

The chat UI in your user’s browser. Renders streamed responses, handles input, and shows interactive elements. The UI lives inside a Shadow DOM so it doesn’t conflict with your site’s styles.
2

Orchestration tier

The agent graph itself, running in the browser alongside the display tier. Decides which tools to call, assembles the system prompt, and runs the reasoning loop.
3

Platform tier

Our infrastructure. Holds credentials, handles LLM and tool calls on the agent’s behalf, persists state, and runs analytics. Everything that needs to be secure, multi-tenant, or proprietary stays here.
The browser handles the user-facing experience and the agent’s decision-making. Our infrastructure handles credentials and platform services.

​
Current limitations

The SDK requires a modern browser: Chrome 140+, Firefox 140+, Safari 17+, or Edge 140+. Chrome, Firefox, and Edge auto-update, so most users are on a supported version by default.