1
Define
Set the acceptable mistake rate for automatic release. For example: no more than 5% of automatically released AI outputs or actions should be mistakes.
Coming soon · Runtime protection for AI workflows
The runtime Risk Gate for AI workflows.
EyeTrustAI helps protect AI workflows from costly mistakes before they affect users, tools, or business processes. You define the mistake budget. Once the Risk Gate is calibrated on real workflow mistakes, EyeTrustAI guarantees that the mistake rate among automatically released outputs and actions stays within the budget you defined. It plugs into the AI systems, workflows, and orchestrators you already use.
Keep your workflow. Add a calibrated Risk Gate before AI releases outputs or actions automatically.
How it works
From first install to calibrated enforcement.
EyeTrustAI is designed to work with the AI systems and workflow tools teams already use. It starts in warm-up mode, learns from labeled mistakes, and turns calibration into automatic release control.
2
Plug in
Add EyeTrustAI before a consequential AI answer, action, or tool call.
3
Warm up
Run in advisory mode while users, reviewers, tests, or downstream outcomes label mistakes.
4
Calibrate
Use conformal risk control to learn which AI outputs and actions can be certified for automatic release in this deployment.
5
Release or route
Once calibrated, allow automatic release only when the mistake rate is controlled. Otherwise, defer for evidence, escalate to a human, or block.
Example workflow
Example: a Risk Gate inside a support workflow.
This is one example of how a user would see EyeTrustAI: as a decision step inside an existing workflow. The same pattern can apply to agents, RAG systems, coding workflows, enterprise approvals, and orchestration platforms.
Customer Support Assistant
Handles incoming requests and routes safe, supported answers.
New support request
Trigger→
Retrieve knowledge
Retrieval→
Draft answer
LLM→
EyeTrustAI Risk Gate
Contract + calibration + evidence + auditReusable module
PROCEED → Send answer
Continue only when the output is inside the calibrated release region.
DEFER → Retrieve missing evidence
Ask the workflow to collect the evidence needed to retry safely.
ESCALATE → Human review
Route uncertain or high-impact cases to a person with context.
BLOCK → Stop + audit
Prevent hard policy violations and record the incident.
risk_contract.yaml
mistake: unsupported_or_wrong_answer loss: bounded_0_1 risk_target: "≤ 3%" evidence: - retrieved_docs - source_policy
SDK / API call
decision = eyetrust.guard({
module: "support-answer-risk",
input, output, evidence
})
The Differentiator
From guardrails to guaranteed automatic release control.
Most AI guardrails enforce static rules: blocklists, policy checks, schemas, or confidence thresholds. EyeTrustAI goes further. It learns from real workflow mistakes and calibrates what your AI workflow is allowed to release automatically.
The guarantee is concrete: once calibrated, EyeTrustAI controls the mistake rate among the AI outputs and actions released automatically, under the mistake budget you define. If the budget is 5%, the Risk Gate learns which outputs and actions can be released automatically while keeping the mistake rate of those releases within that budget.
Define the mistake budget
A Risk Contract makes the target explicit: what counts as a mistake, what evidence is required, and what mistake rate is acceptable for automatic release.
Calibrate from real mistakes
During warm-up, labeled workflow events become the calibration set. EyeTrustAI uses conformal risk control to learn the release boundary for that deployment.
Release only what is certified
When an output or action is inside the calibrated release boundary, it can proceed automatically. When it is not, EyeTrustAI returns DEFER, ESCALATE, or BLOCK with a reason and an audit trail.
What EyeTrustAI adds
A runtime control layer, not another workflow builder.
EyeTrustAI is the runtime control layer behind the Risk Gate: it observes guarded events, applies Risk Contracts, uses calibration when available, and returns decisions your workflow can route.
Risk Contracts
Define what counts as a mistake and what mistake budget applies to automatic release.
Guarded Events
Capture inputs, outputs, actions, context, evidence, and later labels.
Conformal Risk Control
Turn real workflow mistakes into calibrated rules for what can be released automatically.
Evidence-seeking DEFER
Ask the workflow for missing evidence before retrying automatic release.
Audit Trail
Record decision, evidence, contract version, calibration version, and label status.
Module Registry
Install reusable risk-control modules from EyeTrustAI, your team, or partners.
Risk Modules
Reusable modules. Calibrated guarantees.
Risk Modules package the contract, event schema, evidence requirements, labeling protocol, scoring logic, and decision routes for a class of AI workflow. Install a module, calibrate it on your deployment, and reuse it across orchestrators.
Modules are reusable. Guarantees are deployment-scoped.
A module can be shared, sold, or kept private. The guarantee becomes active only after calibration on the workflow, data, model, evidence, and mistake definition where it runs.
Open modules
Reusable templates for common AI workflow risk patterns.
Private team modules
Internal contracts and evidence logic for company-specific workflows.
Gated partner modules
Controlled access for sensitive, specialized, or partner-maintained modules.
Commercial modules
A future path for experts and consultants to distribute risk-control logic.
Module Registry coming as early pilots develop.
Official modules
The first reusable Risk Modules we are building.
These modules are reference implementations of one platform logic. They are not separate architectures; each one uses the same Risk Contract, Guarded Event, calibration, decision, and audit model.
Generic LLM Answer Risk
For support, document QA, and generated responses.
RAG Evidence Risk
For answers that must be grounded in retrieved evidence.
Coding Agent Action Risk
For file edits, commands, tests, and repository actions.
Clinical / Enterprise Evidence Risk
For simulated evidence workflows where missing context should trigger defer or escalation.
Workflow Routing Templates
Example routes for PROCEED, DEFER, ESCALATE, and BLOCK inside common orchestrator patterns.
Module Starter Kit
A package format for contracts, event schemas, labeling protocols, examples, and tests.
Built for workflow builders
Use your stack. Add a calibrated Risk Gate.
You should not need to rewrite your workflow to get calibrated risk control. EyeTrustAI is designed to plug into the stack you already use.
AI automation builders
Add risk routing to AI-powered business processes.
Agent developers
Control consequential tool calls before they happen.
Platform teams
Give internal AI workflows a shared governance layer.
Module authors
Package reusable risk-control logic for others to install.
n8n
Zapier
Make
Workato
LangGraph
Temporal
Airflow
MCP-enabled agents
FAQ
Automatic release guarantees, operationally scoped.
What exactly does EyeTrustAI guarantee?
EyeTrustAI does not guarantee that an AI model is always correct. It guarantees automatic release control: for a defined workflow, mistake definition, calibration set, and mistake budget, EyeTrustAI controls the mistake rate among the AI outputs and actions it allows to be released automatically. Outputs and actions that cannot be certified are deferred, escalated, or blocked instead.
Is EyeTrustAI an orchestrator?
No. You keep your orchestrator. EyeTrustAI is the Risk Gate inside the workflow.
Is EyeTrustAI a Risk Gate or a control layer?
Users plug EyeTrustAI into workflows as a Risk Gate. Under the hood, EyeTrustAI provides the runtime control layer that applies Risk Contracts, calibrates risk, evaluates evidence, records audit trails, and returns decisions your workflow can route.
Is this another guardrail?
Not exactly. Guardrails usually enforce predefined rules. EyeTrustAI learns a calibrated release boundary from real workflow mistakes, then uses that boundary to decide when AI can proceed, defer, escalate, or block.
Does it work before calibration?
Yes, in advisory/warm-up mode. Automatic release guarantees activate only after enough deployment-specific labels exist.
Can modules be reused across companies?
Yes. Module logic can be reused. Guarantees are calibrated for each deployment unless a shared population is explicitly certified.
Can this run privately?
The first version is API/SaaS-first. Local, private, and enterprise deployment paths are planned pilot options for teams with sensitive workflows.
Founding Team
High-stakes workflow expertise meets machine learning reliability.
Built for consequential AI workflows.
EyeTrustAI combines frontline experience in high-stakes decision workflows with deep expertise in uncertainty, reliability, and safety in machine learning.
That combination shapes the product: mathematically grounded, deployment-oriented, and built for teams who need AI systems to know when to proceed, defer, escalate, or block.
Become one of the first EyeTrustAI design partners.
We are looking for AI teams, workflow builders, agent developers, and high-stakes automation teams interested in piloting EyeTrustAI as we build the first Risk Modules and orchestration adapters.