THE GUIDE LAYER FOR YOUR APPLICATION
One guide for people and AI agents. Turn your application knowledge into helpful conversations and approved workflows, with clear permissions and a record of the result.
Reasoning becomes an authorized action.
Explore the layers. Select one to inspect · architecture illustration.
ONE PLATFORM. TWO WAYS IN.
Your users get a conversation. Compatible agents get structured context and approved actions. Your team stays in charge of what each can do.
A familiar chat inside your app. Ask a question, find a page, follow a workflow or review a page action in one conversation. Type or speak, then restart whenever you need a fresh start.
Help me create a purchase order.
I’ll guide you through it. First, choose your supplier. Review each step before continuing.
Give compatible agents relevant app context, exact inputs and published actions. The gateway checks access, follows your workflow and reads back the saved result.
WHY USE THE SMART365 SDK?
Connect your product knowledge and approved actions to an agent. Give customers a clearer path through your software and give your team a record of the work completed through Smart365.
Give authorized, compatible agents a structured guide to your application: relevant fields, required inputs and approved actions through an MCP or custom adapter. Reduce repeated discovery and confirm saved results with verified receipts.
Keep questions, page explanations, navigation and reviewed actions in one chat. Voice input and workflow tools stay in the same place. Our roadmap connects these steps into guided journeys across screens and connected services.
Review the delegated agent-key label, actions, provider calls, approvals, errors, verification and timing in the console. Revoke access when needed.
A SHARED GUIDE FOR COMPATIBLE AGENTS
Codex · Claude · ChatGPT · local models — compatible adapter required.
The percentages below were measured with Codex in a controlled web pilot.
For N tasks, give an authorized agent relevant application knowledge and an approved path instead of repeated discovery. In our controlled Codex pilot (9 runs per arm, 3 synthetic tasks), the normalized token index fell from 100 to 51.6 with no loss in median completion time, including SDK startup. These are measured pilot ratios, not a forecast for every workload. Model cost depends on pricing and caching.
Explore compatible agent integrationsVisibility covers actions routed through Smart365. Agent-key labels are not independently verified bot-provider identities.
See the SDK in use
Give compatible tool-calling agents structured app knowledge and approved workflows through an MCP or custom adapter. This 1 minute 59 second narrated animated video shows how your app, Smart365Guide, and an agent connect, with smooth AI-generated narration and measured Codex pilot results.
Animated architecture illustration, with AI-generated voice narration and English captions. It does not connect a live agent or perform a customer action. Codex, Claude, ChatGPT and local model clients require a compatible adapter and validation. The efficiency figures come from the recorded controlled Codex pilot, including SDK startup; other agents need their own measurements. Results vary, and token savings do not guarantee equivalent billed savings.
THE ARCHITECTURE, EXPLAINED
Tools tell an agent what it can call. Smart365 connects those tools to a user, a business workflow, and a result you can verify.
LAYER 03 / AUTHORIZE
Check identity, account consent, permissions and business policy. Actions that need approval wait for a different authorized reviewer before execution.
consent ✓ scopes ✓ policy ✓HOW THE APP AND AGENT CONNECT
Share relevant application knowledge through the SDK. Let a compatible adapter connect the agent to published actions, with access checks and verification along the way.
Animated architecture illustration. No agent is connected or action performed here.Web SDK · selected app context
Fields · published workflows
Structured guide · scoped access
Execution · verification
Codex · Claude · ChatGPT
Local tool-calling models
The licensed SDK exposes selected routes, field information and app-owned action references. Your team supplies and maintains the relevant knowledge.
Compatible tooling and an approved integration are required for each client. The controlled efficiency measurements use Codex; other agents need their own validation.
BUILT AROUND YOUR AUTHORITY
Give agents a path to action while keeping account permissions and business policy explicit.
Explore security controlsUsers connect their accounts and grant scopes. Agent keys inherit the user’s current permissions.
High-risk actions pause for independent review. Reasoning alone never authorizes execution.
Check provider state, record the evidence, and keep unverified outcomes visible.
START WITH ONE WORKFLOW
Bring a web app, a developer and one repetitive task. We can demonstrate the guide, help map a bounded workflow, and compare results against the existing path before you expand.
Register your application and allowed domain. Add the licensed browser SDK to your site.
Connect your REST or MCP backend. Publish exact inputs, required permissions and checks for the saved result. Supply scoped app context through the SDK guide.
Use an agent adapter to consume the guide. Grant scoped access, test in staging, and review correctness, tokens and completion time with your team.
Free demo and assisted pilot planning. Any ongoing service or model charges are agreed separately. Current pilots focus on web apps; native adapters need separate validation.
A FEW PRACTICAL QUESTIONS
See the full feature catalogue for delivered capabilities, setup requirements and planned work.
Register your web app and allowed domain, then embed the licensed SDK. Add your product knowledge and reviewed workflows. For external agents, connect a REST or MCP provider, publish action contracts and configure a compatible adapter.
Compatible tool-calling clients can connect through MCP or a custom adapter. Codex is the tested reference. ChatGPT, Claude and local model clients require their own adapter setup and validation.
Your team publishes capabilities and their policies. The gateway checks delegated scopes, consent and any required approval. Your backend still enforces its own permissions; an embedded chat does not bypass them.
The current web prototype includes unified chat, optional browser voice, workflow tools, structured agent guides and the capability gateway. Automatic workflow discovery, cross-service planning and native adapters are roadmap work. Start with a staging pilot and measure your own results.
START WITH ONE TRUSTED WORKFLOW
Make your product easier for people to use.
Give agents a trusted way to act.