Trust begins with independent measurement.
AI agents increasingly perform actions on behalf of organizations. Trust requires those actions to be observable, attributable and measurable.
KYDE computes an independent Trust Score from the observable behavior of AI agents at the execution boundary.
Deploy through a single gateway. Existing agents continue to operate unchanged.
KYDE Trust Score™
CRM Sales Agent · Acme Corp
84
/100
Within the expected range for comparable enterprise agents.
Coverage
92%
92% of observable agent activity contributed to the score.
Independent measurement
Organizations increasingly rely on AI agents across models, providers and enterprise systems.
Existing governance tools describe individual environments. They cannot provide a consistent measure of behavior across organizational boundaries.
The Trust Score addresses this by measuring observable actions where they occur: at the execution boundary.
Every score requires a denominator.
A Trust Score without Coverage describes only part of the observed reality.
Coverage expresses how much of an agent's observable activity passed through governed paths and therefore contributed to the score.
Coverage of 92% and Coverage of 41% may produce the same score, but they describe fundamentally different levels of confidence. For this reason, every Trust Score is accompanied by a Coverage Report.
Coverage Report · Acme Corp
Observed, measured and cryptographically signed.
Known paths that have not yet been routed through the gateway.
Provider-internal execution outside enterprise visibility, such as internal reasoning, delegated sub-agents or provider-native integrations.
The execution boundary
Unobserved
Provider-internal
Internal reasoning, sub-agent delegation, provider-native connectors. Quantified in the Coverage Report.
KYDE Gateway
Execution boundary
Verifies agent identity, evaluates actions against mandate, computes the Trust Score, signs every event, and enforces organizational policy where required.
Governed
Enterprise systems
APIs, databases, MCP servers, SaaS endpoints. Every action measured, signed and recorded.
The resulting measurement is independent of model provider or framework.
The score is derived from four observable signals.
No access to model internals is required. No provider-supplied assessment is used.
Identity
Who performed the action?
Each agent receives a cryptographically verifiable identity. Self-hosted agents are identified per instance. Cloud-executed agents are bound to enterprise-issued credentials and connector authorizations.
Mandate
Was the action consistent with the agent's declared role?
Every agent operates within a defined mandate. Observed actions are continuously compared against that mandate.
Example
Baseline
Is the behavior consistent with this agent's established pattern?
Normal operating characteristics are learned over time, including call sequences, timing, endpoint usage and execution profiles. Persistent deviation may indicate compromise, misuse or unintended behavior.
Benchmark
How does this agent compare with similar agents?
Anonymized comparison across comparable role classes provides additional context. A CRM agent is evaluated against CRM agents. A coding agent is evaluated against coding agents. The comparison reflects behavior rather than implementation.
role class: crm-agents · anonymized · n=214
Why independence matters
Trust measurements are currently produced within individual platforms.
Model providers observe their own environments. Enterprises observe only their own deployments. Neither provides an independent measure across providers, infrastructures and organizational boundaries.
The execution boundary is the only location where every externally observable action can be measured using the same methodology. KYDE provides that common measurement layer.
Observe activity within their own platforms.
Observe their own deployments.
Computes one independent Trust Score across providers, models and enterprise environments.
Deployment
Deployment requires a single endpoint change.
No SDKs. No application changes. No framework dependency. Existing agents continue operating while every observable action becomes attributable, measurable and cryptographically recorded.
Supported environments include OpenAI, Anthropic, Gemini, self-hosted models and MCP-based architectures. Target latency: below 100 ms.
$ export OPENAI_BASE_URL=https://kyde.company/v1
$ kyde fleet init
✓ Gateway active
✓ Trust Score initialized
✓ Cryptographic ledger started
$ kyde fleet status
14 agents
Coverage: 94%
Trust Score: 87
Context
Large-scale systems of trust have consistently relied on observable behavior rather than inferred intention. Credit assessment is based on repayment behavior. Telematics evaluates driving behavior. Financial audit evaluates recorded transactions. AI governance requires an equivalent behavioral measurement. The Trust Score applies the same principle to autonomous AI systems.
For enterprises
Deploy the gateway and observe how Trust Score and Coverage evolve from the first recorded action.
Kyde Gateway Starter is free. No SLA, no commitment. Self-hosted, deployed from GitHub in minutes.
For insurers, auditors and certification bodies
Building the next generation of AI assurance frameworks?
The Trust Score provides an independent behavioral signal that can complement underwriting, assurance and certification processes. We welcome collaboration with organizations developing AI assurance infrastructure.
hello@kyde.com →