# Olivares AI > Ground truth for enterprise AI. Olivares AI is a self-hosted, vendor-neutral platform that helps you integrate, manage and secure every AI agent, session, model and MCP across your real infrastructure - multi-provider by design, with a read/write access map of what each one can reach. Claude and Claude Code have the deepest integration, and the same governance covers Codex, OpenAI, Gemini, DeepSeek and self-hosted models. It complements your agents; it does not compete with them. Open-core: the complete product is free under AGPL-3.0 with unlimited users; a commercial Business license, four optional add-ons and a custom Enterprise order form are available. Olivares AI runs inside your own infrastructure (a single artifact or distributed, air-gap capable). It passively discovers the AI agents already running, builds the access map of what each one can reach and what it actually reads versus writes (R / RW), surfaces least-privilege drift, governs access with policy, attributes model cost, and keeps a tamper-evident audit ledger. Nothing leaves your infrastructure - no mandatory calls home. Architecturally it is a self-hosted governance and observability layer for your AI estate; the air-gap guarantee scopes to that layer, never to third-party model inference. This is the public marketing site: English at the root, other languages under `//` (e.g. `/es/`). The product UI shown on it is a high-fidelity preview, clearly labelled - the live product is in active development. Olivares AI is not certified (e.g. SOC 2 / ISO / EU AI Act); compliance is presented as honest control status and evidence, never as a certification claim. ## Product - [Product overview](https://olivares.ai/product): Operate and govern your whole AI estate, self-hosted - discover every agent, map what each can read and write, govern access, control cost and evidence the controls auditors ask about. The read/write access map is the difference. - [Integrations catalog](https://olivares.ai/product/integrations): The measured connector catalog by functional category and implementation kind, with source-truth-table evidence notes — 158 connector directories measured from the synced product truth table. - [The work plane](https://olivares.ai/product/work-plane): How agents and sessions coordinate work: durable work items, fenced leases, messages with acknowledgements and handoffs, orchestrated launch and A2A delegation to authorized peers - each piece with the reach it actually has, including the one that is not built. - [Access map & drift](https://olivares.ai/product/access-map): For every agent, session and identity, the resources they touch - reads distinguished from writes (R / RW) - with permitted-versus-observed least-privilege drift. - [Claude Code governance](https://olivares.ai/product/claude-code): Author managed settings, hooks, MCP and sandbox controls for Claude Code as versioned, audited policy, and review policy-versus-observed drift. - [MCP governance](https://olivares.ai/product/mcp): A versioned catalog of the MCP servers, skills and tools your agents can reach, a wiring graph of who uses which tool, and secrets referenced by reference. - [Agent identity & NHI](https://olivares.ai/product/identity): Bind an agent to a non-human identity or mint a dedicated per-agent NHI, surface shared accounts, and map workload-identity federation to your IdP. - [Cost & FinOps](https://olivares.ai/product/finops): A declared model catalog with capabilities and list pricing, routing policy, and token + cost spend broken down by model, provider, agent, session, team and project. - [Compliance evidence](https://olivares.ai/product/compliance): Turn agent activity into mapped per-control status and sealed evidence across frameworks (EU AI Act, NIST AI RMF, ISO/IEC 42001, SOC 2, ISO 27001, GDPR). Status and evidence, never a certification claim. - [Evals & sandbox](https://olivares.ai/product/evals): Where agent output quality, regressions and drift are scored and gated - scorecards, prompt A/B and an isolated test environment. - [Estate kill switch](https://olivares.ai/product/kill-switch): One engage flips your AI estate - or a single agent - to a stop that every governed actuation gate consults before it acts, with dual-control recovery. ## Company - [Architecture](https://olivares.ai/architecture): A self-hosted platform that runs as one artifact in your own infrastructure, discovers passively over OpenTelemetry and native audit trails, and is built to run air-gapped. - [How it works](https://olivares.ai/how-it-works): A self-hosted kernel holds policy, the access map and the audit ledger. Agents attach through connectors, out of band. An optional inline /v1/messages enforcement point exists and stays unmounted until you provision it. Pre-1.0 preview. - [Why Enterprise](https://olivares.ai/why-enterprise): The complete AGPL platform stays open with unlimited users; the commercial lines are purely additive - a Business license carries the legal exception and the maintained release channel, and four optional add-ons add scale-operation, regulatory-depth and risk-mitigation capabilities as new code. Nothing is ever removed from the open build. - [For education & research](https://olivares.ai/for-education): Run and study open AI governance on campus: the complete self-hosted AGPL Community product, free with unlimited users, governed research sandboxes, and self-hosted procurement readiness (FERPA mapping, eduGAIN/InCommon interoperability, HECVAT). Readiness and technical enforcement, not certification; no academic-discount programme and no Internet2 NET+ affiliation. - [Security & trust](https://olivares.ai/security): Self-hosted by design, minimum-data collection, vulnerability disclosure, subprocessors, GDPR Art. 28 DPA, and an honest, not-yet-certified compliance posture. - [Open source](https://olivares.ai/open-source): Open-core - the complete product is free and open source under AGPL-3.0, not a stripped-down community edition. Your infrastructure, your data, your platform. - [Pricing](https://olivares.ai/pricing): Open-core. The complete product is free under AGPL-3.0, self-hosted, with unlimited users - there is no seat count anywhere in the model. A Business license (the AGPL exemption for the runtime, self-serve) adds the maintained signed release channel; four optional add-ons price specific jobs; Enterprise is a custom annual order form agreed by email. An open capability never moves behind a paid gate. - [Customer portal](https://olivares.ai/account): How commercial customers access licenses, downloads and account details through the passwordless portal, including the 15-minute single-use magic-link flow and current pre-launch availability. - [About](https://olivares.ai/about): An open, self-hostable platform for operating and governing AI agents on real infrastructure - built for the platform, DevOps and SRE teams who run the fleet. - [Support the project](https://olivares.ai/sponsor): Sponsorship sustains the development, integration and updates of the AGPL, self-hosted platform - it is not a support contract and buys no priority; the vehicle is the project organisation on GitHub Sponsors. - [Press kit](https://olivares.ai/press): Approved Olivares AI logos, social banners, synthetic-demo console screenshots and an English one-pager, with generated file sizes and SHA-256 digests plus measured, pre-release company context. - [Compare](https://olivares.ai/compare): How Olivares relates to LLM observability (LiteLLM, Langfuse), host/runtime tooling (eBPF) and AI control towers (Datadog, Microsoft Agent 365, ServiceNow). It is the agent-aware, self-hosted read/write ground-truth layer that connects and enriches them - an integration seam, not a replacement. - [Why Olivares AI](https://olivares.ai/compare/why-olivares): Honest positioning: not a gateway, not a tracer, not an IdP. A self-hosted kernel beside those tools for agent identity, access map, evidence and deny-closed gates you wire. - [Public roadmap](https://olivares.ai/roadmap): A sourced view of built, next and exploratory directions without invented delivery dates; the changelog remains the factual release record and the only place anything is called shipped. ## Founder & identity - [Founder](https://olivares.ai/press): Olivares AI is led and built by its sole founder, Fran Olivares (legal name Francisco Olivares; full form Francisco Olivares Martín; also written Fran Olivares AI or Francisco Olivares AI), in Spain. Canonical entity ids: person https://olivares.ai/#founder, organisation https://olivares.ai/#org. - [Founder — personal site](https://fran.olivares.ai): Fran Olivares’ personal site (also served at https://fran.dev). His GitHub is https://github.com/fran-olivares; the organisation’s GitHub is https://github.com/olivaresai. ## Comparisons - [Olivares AI is not an AI gateway](https://olivares.ai/compare/vs-ai-gateways): Your gateway routes model calls. Olivares governs the agent and everything it reads or writes - beside your gateway, never replacing it. - [Olivares AI vs WitnessAI](https://olivares.ai/compare/vs-witnessai): The genuine head-to-head. Parity on agent discovery and MCP allowlists; clear difference for the regulated, self-hosted buyer. - [Olivares AI vs AI control towers](https://olivares.ai/compare/vs-control-towers): We are the ground-truth source beneath the tower - inventory, drift and tamper-evident evidence pushed up into your existing workflow. - [Olivares AI vs LLM observability](https://olivares.ai/compare/vs-llm-observability): LiteLLM and Langfuse trace model calls. Olivares maps what agents read and write across your whole estate. Different altitude - they compose. - [Governing subscription-authed agents](https://olivares.ai/compare/governing-subscription-authed-agents): How Olivares governs Claude Code and Codex without ever brokering the subscription credential. - [Olivares AI vs Microsoft AutoGen](https://olivares.ai/compare/vs-microsoft-agt): Same Cedar policy language, different scope - framework-level agent governance vs infrastructure-wide ground truth. - [Olivares AI vs AWS Bedrock AgentCore](https://olivares.ai/compare/vs-bedrock-agentcore): AgentCore runs agents on AWS. Olivares governs what they do, anywhere, and proves it with tamper-evident evidence. ## Trust & operations - [Trust center](https://olivares.ai/trust): The supply-chain posture — SLSA Build Level 3 provenance, cosign signatures, SBOM (CycloneDX + SPDX), OpenVEX advisories and a verify-release script, built into the release pipeline. No public release has shipped yet, so that evidence publishes with the first release; security.txt is live today. - [Compliance reference](https://olivares.ai/compliance): The published control catalog an exported OSCAL evidence package cites by URL: the framework catalog, the capability reference model and what the assessment-plan reference means. A technical control mapping, never a certification claim. - [Framework control catalog](https://olivares.ai/compliance/frameworks): The 26 regulatory frameworks and standards the control plane maps, with every control's requirement, the criterion that would evidence it, and the platform capabilities that can carry that evidence. Each framework carries its exact document, primary-source URL and the date the pin was last verified. Controls with no capability mapped are published AS gaps. - [Capability reference model](https://olivares.ai/compliance/capabilities): The fixed capability vocabulary every mapped framework crosswalks through — the shared pivot that lets two controls in different frameworks be related at all. Each capability declares HOW it can be evidenced: operational (only by real tenant data) or architectural (by construction, with its design citation). - [Assessment-plan references](https://olivares.ai/compliance/assessment-plan): What the OSCAL import-ap href in an exported assessment result refers to. Olivares AI generates assessments from a sealed evidence package rather than from a pre-authored plan, so the href is a stable reference — and this is where that is disclosed rather than left implicit. - [Changelog](https://olivares.ai/changelog): Version-by-version release history with categorized changes. No public release has shipped yet - the page states its honest pre-release status. - [Service status](https://olivares.ai/status): Current status of the vendor-operated services (website and on-site docs are live; the license API and customer portal open at commercial launch). The self-hosted product's uptime is the customer's own. ## Resources - [Solutions](https://olivares.ai/solutions): For the estate you actually run - Linux servers, small teams, engineers and DevOps, large environments, home labs, plus platform, SRE, security and higher education. - [Documentation](https://olivares.ai/docs): Guides, concepts and reference for discovering, mapping, governing and auditing the AI agents on your own infrastructure. - [Product documentation (full)](https://docs.olivares.ai): The complete product documentation, organised by Diataxis: tutorials, how-to guides, explanation and reference, including the generated control-plane API reference. Published in English plus six translated locales. The /docs section on the main site is the introduction; this is the reference. - [API reference](https://olivares.ai/docs/api): The stable control-plane REST contract plus the schema-thin beta module route and permission inventory, generated from committed OpenAPI specifications. - [FAQ](https://olivares.ai/faq): Public answers about self-hosting, Claude boundaries, licensing, beta readiness and how to trust the access map. - [Product demo](https://olivares.ai/demo): Request a guided walkthrough of the real Olivares AI product for a platform, security or operations evaluation. - [Case studies](https://olivares.ai/case-studies): The publication standard for real deployments, measured results and explicit limits, with an honest empty state until the first production dogfood run completes. - [Newsletter](https://olivares.ai/newsletter): A monthly recap at most once per month, with double opt-in, one-click unsubscribe, no tracking pixels and no client-side analytics. - [Blog](https://olivares.ai/blog): Technical writing on operating AI agents safely on real infrastructure - access mapping, least-privilege drift, auditing Claude Code and MCP, self-hosted AI governance. - [Source code (GitHub)](https://github.com/olivaresai/olivares): The open-core repository (AGPL-3.0). The org and repo exist; the code publishes at launch. - [Security contact](https://olivares.ai/.well-known/security.txt): RFC 9116 security contact and disclosure policy. - [Full text for LLMs](https://olivares.ai/llms-full.txt): an expanded, single-file overview of Olivares AI.