Latent

Developer tool

Hallucination detection for language-model apps: an SDK sends each finished answer from OpenAI, Anthropic, Gemini or Ollama calls to a reader that scores every token, flags unsupported sentences with the words and the source passage behind each flag, and can hold, replace or reroute a flagged answer; a review queue and Slack, PagerDuty, email or webhook alerts, recalibration on the customer's own traffic, and a deployment in the customer's own VPC that reads an open-weight model directly as it writes; free for 5,000 scored answers a month, then Pro at $299 and Team at $1,499 a month.

Find alternatives to Latent

TypeDeveloper tool

RoleStandalone tool

Bought forWatch an AI agent for suspicious or runaway behaviour

AvailabilitySold

DeploymentSaaS, Self-hosted

Intended forDeveloper, Enterprise

SecurityVendor-asserted

CertificationsSOC 2 Type I

Trains on customer dataUnknown

Zero data retentionAvailable

Data residencyNot recorded

Security featuresNot recorded

StatusActive

Sold withinSold on its own

PricingFreemium, Subscription, Usage-based, Enterprise quote — official pricing ↗

Pricing, Deployment, Security, Sold within — runlatent.ai · 18 Oct 2026

What it does

Guardrails and safety
Evaluation and observability

Features 6

Reads the model as it writesReads a model's internal state to flag likely made-up answers.
Hold, replace or rerouteHolds a flagged answer, swaps in fallback text or asks a stronger model.
Review queueFlagged answers wait with the words that gave them away.
AlertsSlack, PagerDuty, email or a webhook when flags spike or accuracy drops.
Calibrates on your trafficTunes its threshold to the customer's own answers.
Runs in your VPCNothing leaves the customer's environment when self-hosted.

runlatent.ai

Sources

Pricing

Pricing page · 18 Oct 2026

Pricing page: Free at $0 for 5,000 scored answers a month, Pro at $299 a month for 100,000 then $1.00 per 1,000, Team at $1,499 a month for 1,000,000, and an Enterprise plan priced in the contract.

Description

Official documentation · 18 Oct 2026

Home page: "Latent reads your model's internal state as it writes. Made-up answers are held before a customer sees them."

What it does

Official documentation · 18 Oct 2026

Home page: "Latent reads your model's internal state as it writes. Made-up answers are held before a customer sees them."

Constraints

Official documentation · 18 Oct 2026

Derived from the security facts on this page, from the same source.

Security

Official documentation · 18 Oct 2026

Security page: SOC 2 Type I complete, Type II in progress, ISO 27001 not held; the pricing page and home page describe a zero-retention term in the Enterprise contract and a VPC deployment where nothing leaves the customer's environment; vendor claims, no auditor listing.

Sold within