Pretxt
by Pretxt AI
Runtime AI agent supervision that enforces plain-language business policies in milliseconds, without an LLM judge
Pretxt is a runtime supervision layer for enterprise AI agents, aimed at CIOs, security, compliance and platform teams in regulated industries. It checks each agent action against company policies written in plain language and redirects or halts the agent before an unauthorised refund, skipped verification or other violation reaches a customer or system of record.
Pretxt came out of stealth on November 17, 2026 with $5 million in seed funding led by Deciens Capital, with Howdy Partners, Service Provider Capital and NSV Wolf Capital participating. It was founded by CEO Brian Soby, previously co-founder and CTO of SaaS security company AppOmni and a former director of product security at Salesforce, and CTO Aaron Costello, AppOmni's former chief of security research, whose work uncovered serious flaws in Salesforce and ServiceNow AI agents. The company has leadership in Aspen, Colorado and Dublin, Ireland. The product, which Pretxt calls AI Supervision, sits in the path of an agent's model calls and evaluates each proposed action against the task and against policies, limits and allowed promises that the customer writes in plain language. Pretxt compiles those into rules and a readable plan, and it uses a near-deterministic engine instead of a second LLM as judge, which the company says takes about 50 ms per decision against more than three seconds for an LLM judge and gives identical outcomes for identical requests. Teams can test a plan against their own examples, and Pretxt suggests rule changes from real agent traffic that a human must approve, so agents cannot rewrite their own rules. Supervision spans security, accuracy, privacy, brand, escalation and compliance, and the company maps it to SOC 2, ISO 42001 and the EU AI Act. On Zapier's AutomationBench, Pretxt says policy-violating actions fell from about 18% to zero, a company-run result. The product is in early access, with published per-transaction pricing.
A head of AI platform, CISO or compliance lead at a bank, insurer, healthcare or customer-support operation putting autonomous agents into production.
Agent actions that break written business policy are caught and corrected while the task is still running, with a signed audit trail of what was attempted and decided.
At a Glance
- Category
- Governance & Security
- Pricing
- Freemium, Subscription, Usage-based
- Target Market
- CIOs, CISOs, Compliance Officers, Platform Engineers
- Deployment
- Cloud-first, API-based
- Headquarters
- Aspen, United States
Key Features
- ✓Plain-language policy plans
Teams write policies, limits and allowed promises in ordinary language and Pretxt turns them into enforceable rules and a readable plan.
- ✓Runtime intervention
Checks each proposed agent action during execution and redirects or halts the agent before a violation reaches a customer or system.
- ✓Deterministic engine, no LLM judge
Roughly 50 ms per decision with identical outcomes for identical requests, avoiding the cost and inconsistency of a second model.
- ✓Plan testing on your own examples
Run a supervision plan against edge cases and limit-pushing scenarios before it goes live, so rule gaps surface early.
- ✓Human-approved rule suggestions
Pretxt learns from real agent traffic and proposes plan changes, but a person must approve them, so agents cannot rewrite their own rules.
- ✓Evidence and rationale records
Keeps an audit record of what was asked, attempted and decided, with human-checkable reasons, and a digitally signed trail on Enterprise.
Capabilities
Use Cases
- •Refund and discount control
A support agent attempting an unauthorised refund is intercepted mid-task and steered back to the approved policy before the customer sees it.
- •Loan and claims verification
A financial-services agent that skips a borrower verification step is flagged, and complex insurance claims are routed to a human reviewer.
- •Escalating the unexpected
When an agent meets a situation no rule covers, Pretxt halts it and asks a human to define the rule instead of letting it guess.
- •Compliance evidence for audits
Compliance teams use the decision record and rationale to show auditors how agent behaviour maps to SOC 2, ISO 42001 or EU AI Act controls.
Ideal For
Best For
- ✓Customer-support agents that must not promise unapproved refunds, discounts or commitments
- ✓Financial services and insurance agents that must follow verification steps and route complex claims to humans
- ✓Healthcare and life sciences teams that need agents held inside strict, auditable boundaries
- ✓Platform teams that want deterministic, low-latency policy enforcement rather than an LLM-as-judge layer
Not Ideal For
- ✗Buyers who need a proven product with reference customers today: Pretxt is in early access and its benchmark figures are company-reported
- ✗Teams that want to self-serve immediately, since even the free tier is currently gated behind an early-access request
- ✗Organisations whose main concern is prompt injection or model-level attacks, which a dedicated AI security firewall may address more directly than business-process supervision
Deployment
Market Analysis
Pros
- ✓Founders have a track record finding real vulnerabilities in Salesforce and ServiceNow AI agents
- ✓Low-latency, deterministic checks suit high-volume production agents
- ✓Published, transaction-based pricing with a free tier and SSO on every plan
- ✓Human approval is required before supervision rules change
Cons
- ✗Early access only: no named production customers, and the free tier is not open for self-serve signup
- ✗The 60x speed claim and the AutomationBench result are company-run, and coverage did not describe the methodology
- ✗Supported agent frameworks and integrations are not publicly documented beyond MCP and the console
- ✗No independent user reviews or Hacker News discussion exist yet
Pricing
Free
$0
- ✓5,000 transactions/month
- ✓1 active agent
- ✓Console and MCP access
- ✓SSO
Essentials
From $175/mo
- ✓75,000 transactions/month
- ✓Up to 10 active agents
- ✓Custom checks
- ✓Guided setup with a forward-deployed engineer
- ✓$1.50 per 1,000 overage
Professional
From $2,200/mo
- ✓500,000 transactions/month
- ✓Up to 25 active agents
- ✓Use-case-tailored supervision
- ✓Shared support channel
- ✓$3 per 1,000 overage
Enterprise
Contact for pricing
- ✓Digitally signed audit trail
- ✓Compliance reporting
- ✓Region choice
- ✓HIPAA BAA
- ✓SCIM and JIT provisioning
- ✓Optional single-tenant deployment
Metered by transactions, where one transaction is one model call routed through Pretxt counting request and response, billed by count rather than tokens; plans run from a free 5,000-transaction tier to $175 and $2,200 per month, with signed audit trails, HIPAA BAA, region choice and single-tenant deployment reserved for custom-priced Enterprise, and every tier currently requires an early-access request.
Security & Compliance
Sources
This page was written from 5 sources, 3 on domains other than pretxt.ai.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Portal26
AI adoption management platform for shadow AI discovery, AI governance and Claude usage control
Akeyless Agentic Runtime Authority
Intent-based runtime access control for AI agents: credentialless access, per-action policy checks and a session kill switch
Darktrace / SECURE AI
Behavioral monitoring of enterprise AI use: shadow AI blocking, prompt risk analysis and AI agent identity mapping
Abnormal AI Governance
Shadow AI discovery, AI agent inventory and AI spend tracking built from email, identity and OAuth signals