NudgeIn development
Our first application: personal agents making plans between friends, within each person’s rules.
Meet NudgeYour personal permission wallet
bleib lets AI work for you without handing over the keys to your life.
In development. Starting with Nudge.
The world we’re building for
Agents will coordinate
with other agents.
To plan, book and buy on your behalf.
Within the permissions you set.
The agents do the work. You keep the authority.
What you get
Do more.
Give up less.
Delegate
Give your agent a goal. It handles the back-and-forth within your permissions. You get the finished plan.
Meet NudgeOwn
The people, preferences and rules you approve stay in your bleib. Each authorised agent can use the context you allow. No starting from zero.
Inside your walletControl
For a clinic visit, allow your current medication list. Your broader health data stays out of scope.
How permissions workIn your own words
Your rules.
Built as you go.
When a new decision comes up, your agent asks.
Speak or type your answer. Your bleib keeps what you approve.
Your agent
Dinner with Marek is €25.
Shall I confirm it?
You
Yes. And keep future dinner plans under €30.
Your approved rule
- Dinner plans
- Under €30
- Over the limit
- Don’t book
Saved to your bleib.
Your bleib saves it with your approval.
How your wallet works
Private context.
Verifiable control.
Your wallet issues permission.
The source verifies it.
Agent requests access
Off-chainPrivate
Your bleib
Context stays encrypted.
On-chainPublic
Public registry
Your context stays off-chain.
The source verifies both.
Only approved access proceeds.
Notes & reading
Why the boundary matters.
What we’re building, and the research behind the need for it.
OpenAIIncident report
OpenAI’s report of unauthorised access during internal cybersecurity evaluations. August 2026.
Read the report (opens in a new tab)
AnthropicControlled study
Researchers tested models that leaked sensitive information in hypothetical corporate scenarios. June 2025.
Read the research (opens in a new tab)
NIST / MicrosoftDisclosed vulnerability
CVE-2025-32711: an AI command-injection vulnerability in Microsoft 365 Copilot involving information disclosure.
Read the disclosure (opens in a new tab)
European CommissionFurther reading
Europe’s approach to carrying and selectively sharing digital credentials.
Explore the wallet (opens in a new tab)
Agent2AgentOpen standard
Inside the open protocol for agent-to-agent communication.
Read about A2A (opens in a new tab)A few questions
Good to know.
A personal wallet for the context and permissions your AI agents use. Your agents do the thinking and the work. Your bleib keeps the approved context and rules they work within.
That’s what we’re building to remove. Your approved context stays in your bleib. A compatible agent can use the parts you authorise, without you explaining everything again.
No. Your agent can ask when a new decision comes up. Answer naturally, add your conditions, then approve the specific rule before it is saved to your bleib.
No. An agent can request more; it cannot approve its own request. Connected sources and action gateways must verify the permission before releasing information or carrying out an action.
Your approved personal context is designed to stay encrypted off-chain. Records such as your calendar and health data can stay with their existing sources. The public registry verifies control and revocation; it does not hold your personal context.
bleib is in development, starting with Nudge. The website shows the intended experience. A live wallet and public integrations are not available yet.
Let AI do more.
Keep control.
Your agents handle the task. You set the limits.
bleib is in development.
Starting with Nudge.