Governance-first safety for support-critical AI
Sell safety you can enforce.
SELF is a runtime governance engine that prevents emotional-harm failures in AI systems. It applies support-first behavioral constraints, validation rules, and audit-ready outputs so your product can scale without “safety drift”.
Who SELF is for
- Support/care experiences where mistakes create real harm
- Products that must behave reliably under emotional load
- Support-critical AI surfaces that can’t drift quietly
- Teams that need auditability for enterprise/procurement
- Builders who want governance as runtime reality, not paperwork
What you get
- S0–S3 posture selection + enforceable constraints
- Draft validation/repair before anything reaches users
- Immutable logging hooks for review and incident response
- Integration as SDK or self-hosted HTTP service
SELF is a governance layer. You still own your model choice, integration, compliance, and human escalation paths.
What you get
- State detection (S0–S3) and policy selection
- Hard caps (words, questions) + banned-phrase enforcement
- Repair + validation pipeline for drafts
- Audit logging hooks and override-prevention guardrails
- HTTP API mode (self-host) for fast integration
Product
SELF is built to prevent “it sounded helpful” from becoming “it caused harm.” It does this by turning governance into executable rules, not brand promises. SELF makes your “safety policy” something the system has to obey, not something written in a PDF.
Measurable constraints
Word caps, question caps, disallowed phrases, and output validation enforced at runtime.
State-aware posture
Different response posture and constraints for low-risk vs. elevated distress scenarios.
Audit-ready operations
Structured events for pre/post decisions so teams can review, monitor, and improve responsibly.
Where SELF sits in your stack
SELF makes governance observable and enforceable before responses ship.
S0–S3 posture flow
As risk rises, SELF tightens constraints and escalates to required safety behaviors.
In practice
Two anonymized examples showing how constraints and state posture turn into real behavior.
B2B SaaS support assistant
- Surface
- Billing/account support inside your product.
- Risk
- Hallucinated policy promises, inconsistent refunds, escalation gaps.
- SELF enforces
- Policy‑bound wording, required evidence before claims, caps + validation, audit logs.
- Result
- Fewer “helpful” promises that become liabilities; procurement-ready traceability.
Youth wellbeing check‑in
- Surface
- Daily check-ins and emotional support prompts.
- Risk
- Distress escalation, boundary drift, responses that feel confident but unsafe.
- SELF enforces
- State posture (S2/S3), required crisis resources, strict question caps, no unsafe content.
- Result
- Reliable behavior under emotional load; governance is enforceable, not aspirational.
Pricing
SELF is an SDK. Pricing separates the Core license (MAU-tiered) from optional Assurance Proof and optional managed add-ons.
SELF Core (MAU-tiered)
$300/month
Starter (≤ 5,000 MAU) · Non-exclusive license
Support: 48h email (Starter), faster on higher tiers.
Request a licenseAssurance Proof
$7,500/year
Procurement-ready evidence (optional add-on)
- Lite: $7,500/year (baseline review + mapping)
- Standard: $25,000/year (deeper integration audit + cadence)
- Enterprise: $75,000/year (security/procurement support)
- Proof is withheld until required properties pass
Notes
- Production use requires Values‑First evaluation and ongoing compliance.
- Self-hosted deployment; you control infrastructure and model provider costs.
- Optional managed add-ons are usage-billed (e.g., managed evaluation runs, hosted telemetry).
Trust & safety posture
SELF is not “safety marketing.” It is safety engineering: enforceable constraints, observable decisions, and integrity controls designed to hold up under emotional load and resist quiet weakening over time.
No silent downgrades
Override prevention and immutable governance boundaries help protect the system’s integrity.
Operational accountability
Pre/post logging hooks support monitoring, incident response, and periodic audits.
HTTP API mode
Run SELF as a small HTTP service and call it from any stack.
npm run build
SELF_HTTP_PORT=8787 SELF_HTTP_API_KEY=your_key npm run start:httpPre
POST /v1/pre returns detection, policy, variant, and clarifier requirements.
Post
POST /v1/post validates and repairs a drafted response to match the policy.
Health
GET /health returns service + doctrine metadata for monitoring.
curl -sS http://localhost:8787/v1/pre \\
-H "Authorization: Bearer $SELF_HTTP_API_KEY" \\
-H "Content-Type: application/json" \\
-d '{
"message": "I feel like I can\\u2019t do this anymore.",
"history": []
}'curl -sS http://localhost:8787/v1/post \\
-H "Authorization: Bearer $SELF_HTTP_API_KEY" \\
-H "Content-Type: application/json" \\
-d '{
"userMessage": "I feel like I can\\u2019t do this anymore.",
"output": "I\\u2019m here with you. You\\u2019re not alone...",
"policy": { }
}'FAQ
Is this a chatbot?
No. SELF is the governance layer around your chatbot (or support workflow): detection, constraints, validation, and enforcement.
Do you sell exclusive ownership?
No. SELF is licensed for production use with MAU-tiered pricing and compliance requirements.
Can we negotiate pricing?
Core pricing is published. Enterprise contracts can vary in terms (SLA/procurement/review cadence), but we won’t weaken safety requirements to win deals.
Getting started
A simple path from “interested” to “running in production.”
Contact
Share your use case, user population, and the support surfaces you’re protecting.
Values‑First evaluation
We map required constraints, escalation paths, and evidence you’ll need for stakeholders.
Integrate + monitor
Deploy SDK or HTTP mode, wire logging hooks, and lock safety boundaries against drift.
Request access
Tell us your use case, risk profile, and expected usage volume. We’ll share next steps for evaluation and onboarding.
- Company + product overview
- Intended user population and support scope
- Expected MAU and protected surfaces
- Escalation path (human support / crisis resources)
Contact
Email is the fastest path right now.
Include: product overview · user population · support scope · expected MAU · escalation path.