Openlayer
Openlayer is an AI governance and observability platform for regulated enterprises. It inventories every AI system a company runs, tests them before release, monitors live traffic, enforces guardrails in real time and produces audit-ready compliance evidence.
What is Openlayer?
Openlayer is an enterprise platform for governing, testing and observing artificial intelligence systems. It is published by Unbox Inc., a Delaware corporation headquartered in San Francisco that trades under the Openlayer name, and it targets regulated organisations that must prove what their AI actually does rather than simply describe it. The product is organised in seven layers, each useful on its own. Discovery builds an organisation-wide inventory of AI systems and automatically flags shadow AI. Registration turns that inventory into an accountable registry recording owner, purpose, data sensitivity and risk tier. Testing offers more than 175 ready-made evaluations plus custom metrics, runs regression suites against the customer's own data and blocks failures inside a CI/CD pipeline the way a failing unit test would. Observability traces prompts, responses, tool calls and spans, then watches quality, cost, latency and drift on live traffic with alerting. Security and Guardrails blocks prompt injection, jailbreaks, PII and PHI leakage and off-policy outputs in real time, enforcing access and routing rules at a gateway. Compliance maps controls to the EU AI Act, NIST AI RMF, ISO/IEC 42001, OSFI E-23 and internal policy, then generates audit evidence and attestations. Cost Controls reports token spend by project, team and user, applies hard budget caps per API key and can route heavy users to cheaper models. What separates Openlayer from a pure observability tool is how far it reaches. The site publishes a coverage matrix with four cases: code you own, instrumented through the SDK; traffic you can route, handled by the gateway; connected vendor AI attached by telemetry; and opaque vendor AI that can only be registered and tracked. Data arrives through SDKs, a CLI, a REST API, Git, OpenTelemetry or Snowflake, and roughly thirty-five named integrations cover the major model providers and agent frameworks. Openlayer runs as a multi-tenant cloud service on AWS, and also as a self-hosted or air-gapped deployment inside the customer's own VPC. It is a governance tool, not a generative one.
What it does
- Inventory every AI system running across an organisation, including undeclared shadow AI
- Register each system with its owner, purpose, data sensitivity and risk tier
- Test AI systems before release and block regressions inside a CI/CD pipeline
- Trace and monitor live AI traffic for quality, cost, latency and drift
- Block prompt injection, jailbreaks and PII or PHI leakage in real time
- Map controls to the EU AI Act, NIST AI RMF and ISO/IEC 42001 and generate audit evidence
- Cap and allocate token spend per project, team and API key
When to use Openlayer / When not to
A quick filter to help you decide if Openlayer is the right fit.
When to use Openlayer
- Regulated enterprises in finance, insurance, healthcare, telecom, cybersecurity or automotive that must show a regulator what their AI systems did and when
- MLOps and LLMOps engineers who want evaluation suites running in CI so regressions are blocked before a release ships
- Security and compliance teams mapping AI controls to the EU AI Act, NIST AI RMF or ISO/IEC 42001 and needing evidence rather than documentation
- AI governance leads who must inventory every model, agent and vendor tool in use, including the ones nobody declared
- Platform and data teams running a mix of in-house AI, connected vendor AI and opaque SaaS AI that all need governing the same way
When not to use Openlayer
- Individuals and hobbyists: the product is contractual and business-to-business, and the terms require you to be at least 18 and able to bind an entity
- Very small teams with no enterprise budget, since everything beyond the free Basic quotas is quoted through a sales form
- Anyone looking for a tool that generates AI output: Openlayer evaluates and governs models, it does not produce text, images or models of its own
- Non-technical buyers wanting something turnkey, as the richest modes require SDK, CLI or API instrumentation
- Users who need a mobile app or offline access: there is no iOS or Android application, only a web platform, an API and a self-hosted option
How to use Openlayer
A typical end-to-end flow, from setup to results.
- Open a free Basic account through self-service signup, without going through a sales team
- Create a project in the workspace, then generate an API key from the workspace settings
- Decide how each AI system will be attached: SDK instrumentation, gateway routing, telemetry export or plain registration
- For code you own, install the Python or TypeScript SDK and instrument your application
- For traffic you cannot instrument, point it at the Openlayer gateway, which routes requests, manages API keys and applies guardrails
- For third-party AI that emits logs, export traces through OpenTelemetry, OpenLLMetry or OpenLIT
- Register opaque vendor tools declaratively so they still appear in the inventory and the risk registry
- Build test suites from the library of more than 175 evaluations, add custom metrics and set thresholds
- Add Openlayer as a step in your GitHub CI/CD pipeline so a failing test blocks the deployment
- Configure alerting by email, Slack or webhook, then set roles, SAML SSO and access groups for the team
Pros & Cons
Pros
- Unusually broad scope: testing, observability, guardrails, compliance and cost control sit in one platform instead of four
- Four documented ways to attach a system, including opaque vendor AI that cannot be instrumented at all
- Serious and verifiable security posture: SOC 2 Type II, HIPAA, AES-256 at rest, TLS 1.2+, third-party penetration tests and continuous monitoring with Vanta
- A categorical commitment, repeated in three separate legal documents, never to train models on customer data
- Complete and consistent legal corpus dated 28 July 2026, including a published DPA with Standard Contractual Clauses and named Article 27 representatives
- Self-hosted and air-gapped deployment inside the customer's own VPC, with product telemetry disabled by default
- Public, well-structured documentation with an API reference, roughly thirty-five named integrations and a changelog of twenty-seven dated entries
Cons
- No public paid price at all: the Enterprise plan is simply listed as Custom, so any budgeting starts with a sales form
- The free Basic plan is tightly capped at one member, five projects, twenty tests per project, 20,000 inference logs a month and three months of retention
- Everything a regulated enterprise actually needs — role-based access control, SAML SSO, on-premise deployment, compliance reports, SLA, data export — sits on the Enterprise side
- The entry plan is labelled a Trial on the pricing page while being presented as free and open to everyone, and nothing on the site settles whether it is permanent
- Default cloud hosting is in the United States; EU storage is offered but the site states neither the conditions nor the plan required
- The authoritative subprocessor list lives on an external trust centre rather than in the DPA, which gives only categories and examples
- No mobile app, no browser extension, and no interface language other than English is announced
Pricing & Plans
Openlayer publishes a free entry plan and no paid price whatsoever. The Basic plan costs nothing and is opened by self-service signup; it allows one member, five projects, one inference pipeline per project, 20,000 inference logs per month, twenty tests per project, a single workspace, three months of data retention and community support through Discord. The Enterprise plan is listed only as Custom — tailored for larger businesses and is obtained through a sales conversation, so no amount, currency or billing period is published anywhere on the site. Quote-only enterprise pricing is normal practice for this class of B2B software, but it does mean the cost of a real deployment cannot be estimated from the website alone.
- Basic — free
- self-service
- labelled 'Trial — ready to start for everyone'
- Enterprise — 'Custom — tailored for larger businesses'
- quoted on request
Data, GDPR & hosting
A consolidated view of how Openlayer handles your data.
GDPR overview
Openlayer states GDPR compliance explicitly rather than leaving it to inference: the security page carries a GDPR badge alongside SOC 2 Type II and HIPAA, and the about page repeats the claim. The privacy policy and the DPA are both dated 28 July 2026. Section 11 of the privacy policy sets out EEA rights — access, rectification, erasure, restriction, objection, portability, withdrawal of consent and the right to complain to a supervisory authority. Two Article 27 representatives are named, one for the EU in Dublin and one for the UK in Brighton, and a Data Protection Lead is identified by name and email. The published DPA incorporates the 2021/914 Standard Contractual Clauses under Irish law, adds the UK and Swiss addenda, and gives customers fifteen days' notice and a right to object before any subprocessor change.
Who owns the data?
Under section 4.2 of the terms, the customer keeps all right, title and interest in Customer Data, a category defined broadly enough to cover datasets, model inputs and outputs, prompts and completions, traces, spans, inference logs, evaluation results and their metadata. Openlayer receives only a non-exclusive worldwide licence to host, copy, process, transmit and display that data, and solely to provide, maintain, secure and support the service or to act on the customer's documented instructions. Openlayer is the processor for Customer Data and the controller only for account, billing and telemetry data. It keeps ownership of its own software and of any feedback submitted to it.
Reuse rights
The customer stays free to use and reuse its own data without asking Openlayer for permission: the licence granted runs the other way, and it is narrow. Openlayer commits, in three separate documents — the privacy policy, section 4.4 of the terms and section 2 of the DPA — never to use Customer Data to train, fine-tune or otherwise develop its own or any third party's models, never to sell it and never to share it between clients. Service improvement relies instead on configuration, performance and usage data, and on aggregated or de-identified data. Managed AI features do send portions of Customer Data to OpenAI, engaged as a subprocessor under terms the site describes as intended to prohibit training and limit retention. Model providers a customer connects with its own credentials remain that customer's own vendors, not Openlayer's subprocessors.
Data retention & training
Hosting summary
The cloud platform runs on Amazon Web Services in the United States, principally the us-west-2 region, with the marketing website served by Vercel. Account, workspace, billing and configuration data sit in managed PostgreSQL; inference logs, traces, spans and evaluation results in ClickHouse and MongoDB; files and datasets in Amazon S3; and Redis handles queuing and caching. The security page advertises multi-region hosting and states that data can be stored in the EU or the US, though it does not say what activating the EU option requires. Data is encrypted with AES-256 at rest and TLS 1.2 or above in transit, secrets are encrypted before storage, and storage is redundant with automated backups. Platform availability is quoted at 99.9%, rising to a 99.99% SLA on the Enterprise plan. Self-hosted and on-premise deployments are available, including air-gapped installation inside the customer's own VPC with no external network connection. Transfers out of the EEA, the UK and Switzerland rely on the Standard Contractual Clauses. Named subprocessors include AWS, Vercel, OpenAI, Sentry, PostHog, HubSpot, WorkOS, Resend and Stripe, with the authoritative list kept on an external trust centre.
Things to keep in mind
Risks and trade-offs to weigh before adopting Openlayer.
- No paid price is public, so a team can invest weeks of integration work before discovering what the platform will actually cost
- The entry plan is called a Trial on the pricing page yet presented as free and open to everyone with monthly quotas; nothing on the site settles whether it expires, and a team could build on it and be caught out
- Managed AI features send portions of customer data to OpenAI as a subprocessor under terms the site describes as intended to prohibit training — a statement of intent rather than a guarantee, and worth reading closely before sending sensitive material
- Because the vendor says it never trains on customer data, no opt-out mechanism exists: there is nothing to switch off, but equally nothing for a customer to verify on their own side
- Default cloud hosting is in the United States, so an organisation assuming EU residency by default would be wrong; the EU option exists but its conditions are not published
- A governance platform can manufacture false confidence: a green dashboard proves the configured tests passed, not that the AI system is safe, and mistaking coverage for assurance is the classic failure mode of this kind of tooling
- The contract imposes binding arbitration in San Francisco under Californian law, with waivers of class actions and jury trial, which is a real constraint for a non-US customer in dispute
Setup & Integrations
Technical difficulty
Difficulty varies sharply with the path chosen. Opening a free account and creating a project and an API key is straightforward and needs no sales contact. Attaching your own code requires SDK work in Python or TypeScript plus a CI/CD step in GitHub — routine for an engineering team, out of reach without one. Routing traffic through the gateway or exporting OpenTelemetry traces is configuration rather than code. Registering an opaque vendor tool needs no technical skill at all. On-premise or air-gapped deployment is a genuine infrastructure project, reserved for Enterprise, with guided onboarding included.
Deployment
Integrations
Behind Openlayer
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What does Openlayer actually do?
How much does Openlayer cost?
Is there a free plan, and what are its limits?
Will my data be used to train AI models?
Where is my data hosted?
Is Openlayer GDPR compliant?
Does Openlayer offer an API?
Which tools does it integrate with?
Is there a mobile app?
Who is behind Openlayer?
Should you pick Openlayer?
Openlayer occupies a narrow but genuinely difficult position: not another observability dashboard, not another GRC checklist, but an attempt to run compliance controls against live AI systems and let audit evidence fall out as a by-product. The collected pages support that ambition rather than merely asserting it. The documentation is public and well structured, roughly thirty-five integrations are named and individually documented, the changelog carries twenty-seven dated entries, and the legal corpus is complete, internally consistent and dated 28 July 2026. The security posture is the strongest part of the file. SOC 2 Type II, HIPAA, AES-256 at rest, TLS 1.2+, third-party penetration testing, continuous monitoring with Vanta, air-gapped deployment and a categorical refusal to train on customer data are all published, and the DPA backs them with Standard Contractual Clauses and named Article 27 representatives. For a buyer in a regulated sector, that is a materially better starting point than most tools in this category offer. The friction is commercial. Beyond the free Basic tier, capped tightly enough to be an evaluation rather than a deployment, nothing carries a public price, and every capability a regulated enterprise would actually require sits behind an Enterprise quote. That is standard for this class of software, but it means the website cannot tell you what the product costs. Two questions deserve an answer before signing: whether the Basic plan is genuinely permanent, since the pricing page calls it a Trial, and what activating EU rather than US hosting requires. The free tier at least lets a technical team judge the product before any of that is negotiated.
- Choosing a selection results in a full page refresh.
- Opens in a new window.