# Ansvar AI > Source-grounded intelligence with exact citations. EU-hosted. An auditable knowledge layer for legal, compliance, and security work. Ansvar AI is an MCP gateway that connects AI clients (Claude, Cursor, VS Code, Open WebUI, GitHub Copilot) to audited legal, regulatory, and security-control sources. Teams use it for cross-jurisdiction legal research, DPIAs and DORA/NIS2 gap analysis, STRIDE/LINDDUN/TARA threat modeling, and ISO/NIST control reviews. Every answer is grounded in citable source text — no model training on customer data, no silent fallbacks, no uncited claims. Ansvar AI is built and operated by Ansvar Systems AB in Sweden. The service is EU-hosted, runs no server-side AI model — model traffic flows directly from your AI client to your model provider — and is open where it can be. ## Product - [How it works](https://ansvar.eu/how-it-works): MCP-based architecture, client integration, onboarding flow. - [Solutions](https://ansvar.eu/solutions): the workflow each field leads with — security (STRIDE), AI governance (EU AI Act high-risk readiness), privacy (DPIA), public sector (tender review), financial (DORA gap analysis), automotive (ISO 21434/UN R155), agriculture & machinery, robotics (robot TARA), rail & signalling (rail TARA) — grounded with every line cited. Corpus search is on every plan; workflows and the legal evidence layer are plan-gated — see /pricing. Hub for the thirteen sector deep-dive pages at /sectors/ (financial, privacy, ai, public, healthcare, industrial-ot, energy, automotive, drone, security, agriculture, robotics, rail). - [Pricing](https://ansvar.eu/pricing): Free €0 (100 search calls/day, B2B sign-up) · Solo €29/month or €290/year (full-fleet fan-out for one seat) · Premium €249/month or €2,490/year (legal evidence layer + 5 workflow runs/month) · Team €490/seat/month, self-serve (1-seat minimum; workflows on your own documents, 20 runs/seat/month) · Company custom, via talk-to-sales. A seat can belong to a person or an agent. - [Security](https://ansvar.eu/security): EU hosting posture, no-server-side-model architecture, compliance grid, data-flow architecture, security FAQ. - [Coverage](https://ansvar.eu/coverage): the live audited jurisdiction and corpus inventory — legal jurisdictions, security frameworks, and regulatory domains, with provision counts and per-country pages. - [Standards](https://ansvar.eu/standards): clause and control text of ISO/IEC 27001, 27002, 27005, 42001 and ISO/SAE 21434 served through the gateway under a SIS license — cited, not paraphrased. Optional per-user add-on; buyers pick their standards at checkout. Terms: /standards-addon-terms. - [Services](https://ansvar.eu/services): Fixed-scope consulting engagements on the Ansvar gateway — AI Act readiness, gap analysis, and custom workflow engagements. Citation-grounded, senior-reviewed, fixed-price. The four sample deliverables below are complete worked examples in the exact format of a paid engagement, every citation fetched. - [Sample threat model](https://ansvar.eu/services/sample-threat-model): a complete STRIDE threat model of a fictional LLM-backed returns API — threat register with fetched citations, control mapping, CRA duties, honest gaps. - [Sample DPIA](https://ansvar.eu/services/sample-dpia): a complete GDPR Art. 35 assessment of a fictional workplace-wellness app — screening, data flows, risk register, Art. 32 safeguards, an honest Art. 36 call. - [Sample gap analysis](https://ansvar.eu/services/sample-gap-analysis): a complete NIS2 gap analysis of a fictional logistics SaaS — Art. 21(2)(a)–(j) gap register, Art. 23 incident-reporting readiness, gaps marked rather than guessed. - [Sample AI Act readiness](https://ansvar.eu/services/sample-ai-act-readiness): a complete EU AI Act readiness assessment of a fictional HR-tech provider — role determination, Annex III classification, the Art. 6(3) derogation walk, and a cited obligation register. - [NIS2 gap analysis template](https://ansvar.eu/templates/nis2-gap-analysis): Free ungated XLSX/PDF — one row per Article 21(2) measure, Article 23 incident-reporting tracker, verdict dropdowns. - [DPIA template](https://ansvar.eu/templates/dpia): Free ungated XLSX/PDF — Article 35 screening with the WP248 criteria, risk register, Article 36 determination. - [EU AI Act high-risk checker](https://ansvar.eu/tools/ai-act-checker): Free interactive classification walk — Art. 5 screen, Art. 6 gates, Annex III, profiling override — cited to the Act. - Data access: programmatic access for developers is MCP-only via gateway.ansvar.eu over OAuth 2.1 — there is no REST data API. Client setup is under Docs: Quickstart below. - [Rate limits](https://ansvar.eu/limits): Per-tier limits reference for the gateway — rate limits, concurrency caps, the Free-tier daily search quota, and the error contract (caps surface as JSON-RPC -32000 `cap_exceeded` inside the MCP session, not HTTP 429; tier-hidden tools return -32601). Includes sample retry code. - [How Ansvar AI operates](https://ansvar.eu/ai-governance): Operating posture — your AI client talks to your model provider directly, deterministic citation validation, EU-hosted retrieval, and no training on customer data. ## Agent-readable surfaces - [Setup](https://ansvar.eu/setup): how to connect Claude, Claude Code, ChatGPT, Gemini, VS Code Copilot agents mode, Cursor, Copilot Studio, and custom MCP clients to the Ansvar Gateway. - [Docs: Quickstart](https://ansvar.eu/docs/quickstart): connect an AI client to the gateway and run a first cited query. - [Docs: MCP basics](https://ansvar.eu/docs/concepts/mcp-basics): what MCP is and how the gateway uses it — for readers new to the protocol. - [Docs: Connect ChatGPT](https://ansvar.eu/docs/setup/chatgpt): a custom connector in ChatGPT Developer mode, Business/Enterprise rollout notes, Codex CLI, and the API path. - [Docs: Connect Gemini](https://ansvar.eu/docs/setup/gemini): Gemini CLI and Code Assist via OAuth, Antigravity, Gemini Enterprise, and the headless API path. - [Use cases](https://ansvar.eu/use-cases): captured production runs, each published verbatim on its own page — a NIS2 gap analysis with the policy attached (licensed ISO 27001 clauses from SIS fetched in the loop), a security-policy gap analysis, DPIA for an HR vendor, Swedish LOU tender review, STRIDE threat model, EU AI Act classification, and a paragraph-cited document review — with the cited answers exactly as the gateway returned them. - [Coverage JSON](https://ansvar.eu/coverage.json): machine-readable coverage endpoint — the audited jurisdiction, framework, and corpus inventory behind the Coverage page, as JSON. - [MCP discovery](https://ansvar.eu/.well-known/mcp.json): structured site-level pointer to the gateway endpoint, browser-WebMCP tools, and documentation links. - [Plugin manifest](https://ansvar.eu/.well-known/ai-plugin.json): legacy ChatGPT-plugin-format descriptive pointer at the MCP gateway and docs — informational only, no API block. - Browser WebMCP: when a browser supports `navigator.modelContext`, ansvar.eu registers read-only informational tools for agents. Current browser WebMCP tools: - `describe_ansvar`: explains what Ansvar does, the trust model, and important boundaries. - `list_ansvar_use_cases`: returns the main workflow examples and links. - `get_ansvar_gateway_setup`: returns the MCP endpoint, OAuth model, client setup links, and the no-free-usage boundary. - `try_use_case(id)`: replays a deterministic gateway response for one of the documented use cases (`gap-analysis`, `national-gdpr`, `dpia-hr-vendor`, `tender-review-lou`, `ai-act-cross-jurisdiction`, `refusal`). The JSON is identical for every caller — static demo, not a live gateway call. Returns `use_case_not_captured_yet` for use cases whose authentic captures have not landed. These tools are informational and demo-only. They do not run live gateway queries, execute workflows, upload files, or bypass subscription gates. Live cited answers require an authenticated Ansvar subscription through `https://gateway.ansvar.eu/mcp`. ## Company - Contact: team@ansvar.eu - Registered: Ansvar Systems AB, Sweden. ## Legal - [Privacy](https://ansvar.eu/privacy) - [Terms](https://ansvar.eu/terms) - [DPA](https://ansvar.eu/dpa) - [Subprocessors](https://ansvar.eu/subprocessors) - [ISO Standards Add-on terms](https://ansvar.eu/standards-addon-terms) ## Blog - [Blog index](https://ansvar.eu/blog): field notes on auditable AI for legal, compliance, and security work. RSS at https://ansvar.eu/blog/feed.xml. - [Stop letting threat models die in a wiki: make STRIDE output double as compliance evidence](https://ansvar.eu/blog/threat-model-to-compliance-evidence): A threat model and a NIS2/DORA/ISO gap analysis describe the same system. Map each threat to the measure it satisfies, with article-level citations. - [EU AI Act: your obligations depend on your role, not your tech stack](https://ansvar.eu/blog/eu-ai-act-obligations-by-role): Provider, deployer, importer, distributor, GPAI provider — each EU AI Act role carries its own obligations. The role test, article references, and live dates. - [Why RAG over a document dump fails regulated work](https://ansvar.eu/blog/cited-answers-vs-rag-for-regulated-work): RAG citations are decorative — no provenance contract links chunk to answer. Regulated work needs typed corpus tools, deterministic validation, and refusal. - [Swedish law as an MCP server: how SFS statutes become a queryable corpus](https://ansvar.eu/blog/swedish-law-as-an-mcp-server): 6,041 Swedish statutes from Riksdagen, segmented to section level and served as an MCP — query by SFS number, chapter, and paragraf, every result cited. - [What you can and can't automate in a DPIA](https://ansvar.eu/blog/automating-the-dpia-evidence-trail): A GDPR Article 35 DPIA has an automatable core and a judgment core. AI assembles the cited evidence trail; the DPO signs necessity and proportionality. - [Working through DORA Article 28: third-party obligations, the contract checklist, and what the auditor asks for](https://ansvar.eu/blog/dora-article-28-third-party-mapping): DORA Article 28 sets the third-party obligations; the contract clauses live in Article 30. Each subsection mapped to ISO 27001 and SCF controls. - [NIS2 vs ISO 27001: a clause-by-clause working mapping (and where ISO stops short)](https://ansvar.eu/blog/nis2-vs-iso-27001-clause-mapping): Every NIS2 Article 21(2) measure mapped to ISO 27001:2022 Annex A — and the three real gaps: reporting clock, management liability, supply chain depth. - [What an MCP gateway is, and why compliance work needs one](https://ansvar.eu/blog/what-is-an-mcp-gateway-for-compliance): Chat and RAG hallucinate regulatory citations. An MCP gateway adds routing, fan-out, tier auth, and deterministic citation validation — and when you need one. - [Effective risk: turning the LLM-era CVE firehose into a triage queue you can actually work](https://ansvar.eu/blog/effective-risk-cve-noise-llm-era): AI tooling files more CVEs than any analyst can read. Effective-risk rescoring deterministically scores CVE × asset × controls, citing every score change. - [How we threat-model AI systems: STRIDE meets MCP](https://ansvar.eu/blog/how-we-threat-model-ai-with-mcp): STRIDE was built for 1999 monoliths. How we adapted it for agentic systems and shipped it as a workflow your own AI client runs through the Ansvar gateway.