Klira AI

Klira AI describes itself as a safety and risk lab for AI agents. Its stated approach combines three instrument lines: PROBE, secret rotating adversarial attack suites used to measure an agent's failure boundary; ATTEST, tamper-evident monitoring that produces signed snapshots of an agent's configuration at every change; and PUBLISH, an open registry of real incidents, measured deployments and analysis. The company frames agent risk as structurally hard to measure for seven reasons it enumerates — non-determinism, non-stationarity, absent loss history, self-reported vendor evidence, correlated failure across a small number of shared foundation models, unmapped severity distributions, and shifting regulation — and argues that because most agents run on the same few foundation models, a single bad model update becomes a common-cause event across many deployments.

The technical product is the Klira AI SDK, positioned as a universal LLM observability and governance layer. It provides tracing, guardrails, evaluations and policy enforcement through decorators (@workflow, @task, @agent, @tool, @guardrails, @mcp_guardrails) and is built on OpenTelemetry. The SDK includes automatic framework detection with adapters for OpenAI Agents, LangChain, LangGraph, CrewAI, LlamaIndex, Vercel AI and custom frameworks, and LLM adapters for OpenAI, Anthropic, Gemini and local Ollama. Governance is evaluated in three layers: a regex-based fast rules engine, context-aware policy augmentation of prompts, and an LLM fallback for complex policy decisions. A hierarchical context model (organization, project, agent, conversation, user) supports cost attribution, performance analysis, audit-trail generation and role-based access control.

Alongside the SDK, Klira sells access to a cloud AI Control Center covering observability dashboards, policy management and compliance reporting, with pre-built policy templates for regulatory frameworks including HIPAA and GDPR. Healthcare-specific features include clinical event logging and PHI redaction, masking, replacement or hashing. The company also markets an AI compliance consulting and ethical risk assessment service that evaluates chatbots, virtual assistants, recommendation systems, automated decision-making tools and content generators for issues such as prompt injection risk and data leakage.

Business model

Klira operates a two-tier commercial model around its AI governance platform. An open-source SDK, distributed under the Apache License 2.0 and installable via pip (klira) and npm (@klira-ai/sdk), runs locally and works without a connection to Klira's cloud. Paid access to the Klira Platform — a cloud-based AI Control Center with observability dashboards, policy management and compliance reporting — is sold in a Pro tier (standard API limits, email support, policy templates, compliance reporting) and an Enterprise tier (custom API limits, dedicated support, custom SLAs, self-hosted deployment, BAA availability, premium compliance features, and selectable US or EU data residency). The company also offers a free 30-minute AI compliance consulting and ethical risk assessment call, delivering a risk score, vulnerability findings and a written compliance report within 2-3 business days as an entry point to paid engagements.

Subscription access to the Klira cloud platform across Pro and Enterprise tiers, with the SDK itself free and open source. Enterprise arrangements include custom API limits, dedicated support, custom SLAs, self-hosted deployment, Business Associate Agreements for HIPAA customers and Data Processing Agreements for GDPR/CCPA customers. Paid compliance solutions may follow the free assessment call, which carries no purchase obligation.

Traction

Public directory listings place headcount at 1-10 employees. Klira maintains an incident registry it describes as containing 800+ incidents on record, drawing on public sources such as the AI Incident Database.

Latest developments

Klira's terms of service for its AI compliance platform, covering the open-source SDK, cloud services and API, took effect January 26, 2026. The company lists adjacent offerings on its site: Tessero for agent commerce (tessero.dev), Attacker for agent red-teaming, and Coboling, an open-source equivalence gate for AI-driven changes to COBOL systems.