Roadmap
Where Rover is headed — what is live today and what ships next on the way to every website becoming a callable, composable agent surface for humans and machines.
| Initiative | Status |
|---|---|
| Agent-to-Web Protocol (A2W) | Live |
| Agent Identity & Analytics | Shipped |
| Per-Agent, Per-User, Per-Geo Customization | Parts live |
| WebMCP and Browser-Native Tools | Early |
| Open Compatibility Suite | Planned |
| Voice Accessibility | Early access |
Agent-to-Web Protocol (A2W)
Live. The open protocol for AI agents, CLIs, and autonomous systems to interact with any Rover-enabled website.
POST /v2/a2w/runswith a canonical prompt-native envelopeGET /v2/a2w/runsmarkdown execution for URL-fetch chatbots- Structured results via JSON polling, SSE, or NDJSON streaming
- Continuation input for multi-turn agent workflows
- Workflows and delegated handoffs — every run belongs to a workflow, and
POST /v2/a2w/runs/{id}/handoffscarries one lineage across Rover-enabled sites - Discovery marker plus
rover-site.jsonandagent-card.jsonfor endpoint detection - No Playwright, no MCP servers, no middleware required for the base protocol
Agent Identity & Analytics
Shipped. First-party visibility into AI agent traffic on your site — direct signal from real interactions, not estimates from third-party GEO or AI SEO tools.
- Which AI agents visit your site and what they attempt
- Success and failure rates per agent, per workflow
- Trust tiers in production:
self_reported,heuristic,anonymous, andsigned_directory_only;verified_signedis reserved for verified HTTP Message Signatures - Rover accepts the calling agent's identity and records verification method, discovery surface, capability ID, page ID, execution path, and workflow lineage
- Intent analytics: what are AI agents trying to do on your site?
- Ground truth in place of guesswork from third-party AI traffic estimators
Per-Agent, Per-User, Per-Geo Customization
Parts live. User and account identity, bounded visitor context, verified account lookup, conversation memory, and agent attribution are in production. Per-geo policy and per-calling-agent capability control remain future work.
- Live: identify users and accounts, pass bounded plan and goal context, and keep optional recent-request memory
- Live: attribute visiting agents with explicit trust tiers and connect account outcomes in analytics
- Future: vary owner policy by geography
- Future: grant different capability sets to different verified calling agents
- Future: combine user, region, and calling-agent policy in one owner control surface
WebMCP and Browser-Native Tools
Early. Rover already registers discoverable tools with the page's WebMCP registry for WebMCP-aware agents. Browser-native tools compile from the same capability graph, without making WebMCP the only path.
- WebMCP stays an optional accelerator for live-tab execution, not the base standard
- The same capabilities remain reachable through
rover-site.json,agent-card.json, and A2W - Turn any web app into a composable building block for agent workflows
- Auto-index site capabilities for browser-native tool discovery
Open Compatibility Suite
Planned. Ship the Rover discovery ladder and compatibility suite publicly before pushing the model into formal standards bodies.
- Beacon-first UX spec with silent, beacon, integrated, and debug modes
- Compatibility tests for raw HTTP, agent-card consumers, DOM/a11y agents, screenshot/VLM agents, and WebMCP-capable browsers
- Clear distinction between
rover-site.jsonas the rich profile andagent-card.jsonas the interop card - Standards work follows deployment reality rather than blocking product rollout
Voice Accessibility
Early access. Browser-native voice-driven workflows for hands-free interaction with any Rover-enabled site.
- Speech-to-action: speak a task, Rover executes it
- Browser-native dictation — no external services required
- Configurable silence detection and auto-stop
- Full voice-first experiences for accessibility and mobile