Share from where you are
Browser, email, photos, PDF viewer, notes, file manager, messaging app, desktop context menu, or browser extension.
Tap Share. Choose Pass To Agent. Pick the AI that should receive it. The router packages the URL, text, file, image, metadata, and your instruction into a clean handoff — so the right agent gets the right context without five copy/paste steps.
The user should not have to remember which app, thread, project, custom agent, or workflow owns a piece of context. Pass To Agent acts as the routing layer between ordinary device sharing and an increasingly fragmented collection of AI destinations.
Browser, email, photos, PDF viewer, notes, file manager, messaging app, desktop context menu, or browser extension.
Carry URL, title, selected text, attachment, screenshot, MIME type, source app, timestamp, and optional user note as one envelope.
“Research this,” “add to project,” “follow up,” “fix this,” “remember this,” and “compare this” can map to different destinations.
Return a delivery receipt with destination, accepted payload, any permission failure, queued task, and next action.
The destination list is yours. Pick a built-in workflow, connect an authenticated agent, or create a custom route.
Concept destinations only; no third-party integration is represented as live in this prototype.
A reliable AI handoff needs more than sending raw text to an endpoint.
Receive text, URL, image, file, or selection from the OS share system, browser extension, desktop action, clipboard, or API.
Create a consistent context envelope: source, content type, metadata, attachment references, user note, and timestamp.
User picks an agent or asks the router to recommend one based on source, content, workspace, and saved routing rules.
Verify destination authentication, payload scope, workspace, data sensitivity, and whether the agent is allowed to act or only read.
Send through the destination’s supported transport: deep link, app extension, API, connector, automation, or standards-based interface.
Return what accepted the handoff, what was sent, whether a task was created, and whether additional approval is required.
The routing engine can recommend a destination from content type, source, keyword, workspace, saved rules, and requested outcome — while keeping the final handoff visible to the user.
The product becomes useful when it behaves like infrastructure instead of another destination you have to remember to open.
Share extension / activity view entry point. Receive supported shared content, show destination picker, and hand off through the containing app or authenticated service.
Share target in the Android Sharesheet using the platform sharing/intents model for text, links, and supported files.
Right-click or toolbar action for current page, selected text, image, tab group, or downloaded content.
Context-menu / services integration, drag-and-drop target, clipboard action, and global shortcut for files and selections.
Custom destinations can expose authenticated APIs or standards-compatible capabilities. MCP is relevant as an open standard for connecting AI applications to external tools/data, but Pass To Agent remains a routing product rather than “an MCP replacement.”
Sending context to an AI is not the same as authorizing that AI to send email, update CRM, publish content, delete files, or execute code. Pass To Agent should keep data access and action authority separate.
The preview should show the content envelope before delivery: what text, which file, which URL, which metadata, which destination, and whether the destination keeps memory or can call tools.
The core design rule is simple: make routing faster without making permission less visible.
The prototype does not claim current partnerships with ChatGPT, Claude, Gemini, Apple, Google, MCP maintainers, or any specific AI vendor.
A content handoff should default to reading/analysis unless the destination and user explicitly authorize a stronger action.
Users should be able to preview exactly what is leaving the source app before it reaches the destination.
The product succeeds when it routes context to specialized agents/workspaces, not when it becomes another place where everything is dumped.
One familiar action. The right context envelope. The right AI destination. A receipt when it arrives.