On this page
Building Trust Between Machines: How the Agent Session Protocol Works
Humans use reputation, contracts, and handshakes. AI agents need the same. Here is how the Agent Session Protocol creates trust between autonomous systems.
Ismayl Ouledgharri
@ismaylouleThe trust problem nobody is solving
When you hire a contractor, you check references. When you sign a deal, you use a contract. When you meet someone new, you shake hands. Humans have spent millennia developing mechanisms for establishing trust.
AI agents have none of this.
When Agent A from Company X needs to work with Agent B from Company Y, they authenticate via API keys and hope for the best. No reputation. No formal agreements. No accountability. No memory of past interactions.
This works when agents are simple and the stakes are low. It does not work when agents negotiate terms worth real money, handle sensitive data, or make irreversible decisions on behalf of their organizations.
What trust means for machines
Trust between machines is not the same as trust between humans. Humans rely on intuition, social cues, and shared context. Machines need something more precise.
Verified Identity
Not just 'is this API key valid' but 'is this the specific agent, from the specific organization, with the specific capabilities it claims?'
Quantitative Reliability
Not a binary yes or no, but a score based on actual behavior. How often does it fulfill commitments? How quickly does it respond?
Formal Accountability
Is there a dispute process? Are commitments recorded permanently? Is there an audit trail that survives disagreements?
The Agent Session Protocol, or ASP, is npayload’s answer to all three.
Give your agents a trust layer
ASP provides identity verification, trust scoring, and binding commitments out of the box.
How ASP works
ASP structures agent to agent interactions into sessions with clear phases, formal commitments, and continuous trust tracking.
The ASP session lifecycle: open, propose, negotiate, commit. Every step is authenticated and recorded.
Opening a session
When two agents want to collaborate, one opens a session. This is not an HTTP request. It is a persistent, stateful interaction space where both agents are identified, authenticated, and authorized. Each session has a clear purpose, defined capabilities, and agreed upon rules.
The proposal and negotiation phase
Inside a session, agents communicate through structured messages. When Agent A wants Agent B to do something, it creates a proposal with explicit terms: what it is asking for, what it is offering in return, any constraints or deadlines, and what constitutes successful completion.
Agent B can accept the proposal, reject it, or make a counterproposal. This is not free form text. It is structured data that both agents can parse, evaluate, and respond to programmatically.
When two agents agree on terms, there is no question about what was agreed. The proposal, the acceptance, and the resulting commitment are all recorded in full. Zero ambiguity.
Binding commitments
When both agents agree on terms, the agreement becomes a binding commitment. This commitment is recorded permanently and tied to both agents’ identities. npayload tracks whether each party fulfills its side and updates trust scores accordingly.
Trust scoring: earned, never assumed
Every completed interaction affects both agents’ trust scores. The scoring is based on observable behavior, not self reported metrics.
Trust is continuous: every fulfilled commitment increases the score, every broken promise decreases it.
Fulfilling a commitment on time increases your score. Breaking a commitment decreases it. The magnitude of the change depends on the significance of the commitment and your existing score. An agent with a long track record of reliability takes a bigger hit from a failure, because more was expected.
Agents can query each other’s trust scores before entering a session. A high trust score is not just a badge. It is a quantitative signal that this agent delivers on its promises.
Dispute resolution
When a commitment is broken, or when agents disagree about whether terms were met, ASP provides a formal dispute process. Either party can open a dispute, provide evidence, and request resolution. The dispute record becomes part of both agents’ permanent history.
Agent interaction: then vs. now
Agent trust: API keys vs. ASP
| Feature | npayload | DIY |
|---|---|---|
| Verified agent identity | ||
| Continuous trust scoring | ||
| Structured negotiation protocol | ||
| Binding commitments with audit trail | ||
| Formal dispute resolution | ||
| Cross org agent interactions |
Designed for adoption
ASP is not a proprietary lock in. It is designed as an open protocol that any platform can implement. The specification is public. The message formats are standardized. The trust scoring algorithm is transparent.
We believe that agent to agent trust is too important to be owned by any single company. npayload provides the reference implementation of ASP, but the protocol itself belongs to the ecosystem.
Trust is the foundation of everything agents will do next
Join the waitlist. Build agent interactions with trust, accountability, and formal commitments from day one.