Linkworld Platform
Agentic features in your product.
EU-sovereign. No rewrite.
Add agentic features to your product without operating your own AI infrastructure: point your existing OpenAI client at Linkworld — no rewrite, no new SDK — and it runs on EU-sovereign, multi-LLM infrastructure with tenant isolation, per-tenant usage metering, an approval layer in front of every high-stakes action, and a complete audit trail.
Early access · key issued in your workspace · starter quota included · no sales call required
Best model per task — Claude, GPT, Mistral, open source. Provider outages fail over automatically.
Every tenant runs isolated; all usage is metered per tenant — billing-ready for your business model.
A security gate in front of high-stakes actions: payments, outgoing mail, data access — only after approval.
EU data centers, GDPR-native, EU-AI-Act-ready. Every action logged in a structured, exportable way.
Platform API · Sovereign models
EU-sovereign LLMs. Per token, on demand.
Open models, operated EU-sovereign under our control — including the best reasoning models, always in their latest version. Create an API key in your workspace, start building, billed per tenant. Frontier models run through the same router.
| Model | Class | Input · € / 1M tokens | Output · € / 1M tokens | |
|---|---|---|---|---|
| gpt-oss-120B | Reasoning — GPT open-weight: math, code, planning | EU-sovereign | €0.10 | €0.48 |
| Qwen 3.5 397B | Flagship — reasoning & vision, long context | EU-sovereign | €0.72 | €4.32 |
| Qwen 3.6 27B | Latest generation — analysis & vision, efficient | EU-sovereign | €0.48 | €3.24 |
| GLM 5.2 | Agentic reasoning — tool use, code, long chains | EU-sovereign | €0.95 | €3.10 |
| Llama 3.3 70B | All-rounder — text, summarization, extraction | EU-sovereign | €0.80 | €0.80 |
| Mistral Small 3.2 | Fast & cheap — triage, routine, agents | EU-sovereign | €0.11 | €0.34 |
| gpt-oss-20B | Compact reasoning — the cost sweet spot | EU-sovereign | €0.05 | €0.18 |
| Claude · GPT · Gemini | Frontier reasoning — through the same router | Routing | Provider price + fixed platform fee | |
Always the latest model version — upgrades automatic · metered per tenant to the second · volume tiers from 100M tokens/month
Batch: non-urgent bulk jobs via /v1/batches at ~half price — we pass the provider batch discount straight through.
Managed RAG & embeddings
Text and PDF documents in, answers out: the platform handles extraction, chunking, embeddings and the index — isolated per tenant, via /v1/rag.
| Service | Unit | Price | |
|---|---|---|---|
| Embeddings · Qwen3-Embedding 8B | per 1M tokens | EU-sovereign | €0.12 |
| Embeddings · bge-m3 (multilingual) | per 1M tokens | EU-sovereign | €0.02 |
| Vector storage | per GB / month | EU-sovereign | €0.25 |
| RAG queries (search + context) | per 1,000 queries | EU-sovereign | €0.40 |
Collections, documents, files and query through one API · ingestion is billed as embedding tokens, nothing else
Early access: API keys right in your workspace · starter quota included
Your OpenAI client already works.
The API is OpenAI-compatible: swap the base URL, drop in a key — and your existing code talks to sovereign EU models. No new SDK, no migration.
from openai import OpenAI"https://api.linkworld.ai/v1","rk_live_…","gpt-oss-120b", …OpenAI-compatible · /v1/chat/completions
Personal data never touches a non-EU model.
Every call carries its intent: tier picks the model class, contains_pii arms the PII gate — enforced centrally, every model choice logged as evidence for EU-AI-Act audits. The default path stays EU end-to-end, with no US processor; frontier models are a deliberate, logged opt-in for tasks without personal data.
"tier": "eu-sovereign""contains_pii": true"tier": "frontier" + "contains_pii": trueModel policy · enforced centrally
Headless · Platform API
Your product. Our runtime.
You're building a SaaS product and want to ship agentic features — without running agent infrastructure yourself. Your software consumes Linkworld headless: your customers only ever see your product. Metered and billed per end customer.
Isolated data, credentials, audit trail and usage numbers — per customer of your product.
Agents work with your domain APIs via MCP — the business logic stays in your product.
The approval layer and audit log apply headless too — configurable per end customer.
POST /v1/agents/{agent}/invocationsPUT /v1/tenants/{customer}/mcp-bindings/crmGET /v1/tenants/{customer}/usagePlatform API · tenant-scoped
MCP-native
Your systems speak MCP.
The Model Context Protocol isn’t an add-on here — it’s how agents reach tools, yours and ours alike.
Attach your CRM or ERP per tenant as an MCP tool — agents use it like a platform tool, scope-gated.
Apps reach platform tools via MCP with skill gating: the agent only sees what the tenant is entitled to.
Claude Code, Cursor & co. connect directly to the platform — lint, deploy and test console as MCP tools.
The architecture
One runtime. Every layer.
From UI to isolated execution: your apps and products sit on top — everything underneath is Linkworld.
Agentic apps · your SaaS via API · AI chat
Orchestration, states, approvals
APIs, databases, SaaS, MCP
Short- and long-term context, RAG
Best model per task, failover, EU-sovereign by default
Security gate, policies, audits
Isolated runtime, scaling, usage per tenant
Agent levels
From answers to autopilot — one scale.
The same runtime carries every autonomy level. You decide per app and per tenant how far it may go.
Knowledge agents and tool access — controlled, auditable.
Multi-step workflows with rules, intermediate steps and approvals.
Operator agents and whole agentic apps — up to the business process.
Compliance & trust
European, transparent, verifiable.
Compute and storage exclusively in EU data centers. Sub-processors documented.
DPA available, data-subject rights as product features — no compliance theater.
Risk classes, transparency logs and human oversight built in.
Every action logged in a structured way, retention per tenant, export for audits.
Frequently asked
Before you integrate.
Do I need to rewrite my code to use the Platform API?
No. The API is OpenAI-compatible — you just swap the base URL and key, and your existing code keeps running unchanged. No new SDK, no migration.
Where are the models that run behind the Platform API operated?
EU-sovereign, under our control — including the best reasoning models, always in their latest version. Frontier models like Claude, GPT or Gemini run through the same router.
What happens if a call contains personal data?
The PII gate enforces an EU-sovereign model centrally. Combining the "frontier" tier with contains_pii is blocked with a 403 policy_violation and logged — evidence for EU-AI-Act audits.
How is usage billed?
Per token and per tenant, metered to the second. Volume tiers apply from 100M tokens/month; non-urgent bulk jobs run through /v1/batches at roughly half price.
Can I use retrieval (RAG) through the API too?
Yes — managed RAG via /v1/rag handles extraction, chunking, embeddings and the index, isolated per tenant. Only embedding tokens, vector storage and queries are billed.
Can my own domain APIs talk to the agents?
Yes, via MCP. You attach your domain API per tenant as an MCP tool, and the agent uses it scope-gated like a platform tool — the business logic stays in your product.
Is the Platform API generally available yet?
It's currently in early access: the API key is issued right in your workspace, with a starter quota included and no sales call required.
Start right now.
No framework agreement, no sales cycle: request an API key and get your starter quota through early access. As your volume grows, the tiers grow with you.
Start free Build apps →