A canvas for agents that get work done.
State an intent. Approve the scope. Get structured results. Orchestration is built in.
State the intent.
The address bar routes to the right agent. Not a search engine. A delegation surface.
Approve the scope.
Before anything runs, the exact data the agent will access is shown. Reject it, modify it, or approve it. Nothing executes without sign-off.
Get structured results.
Results are schema.org JSON-LD. Composable — pipe one block's output into the next agent.
Works without a model.
Search, knowledge, and data agents resolve on-device without any LLM. Adding a model unlocks natural language routing and agent generation. Either way, nothing is sent anywhere without a mandate.
Every run leaves a receipt.
Each execution produces a co-signed receipt recording what ran, what scope was granted, and what the agent actually did. Signed by both parties. Verifiable independently.