Four things bundled behind one endpoint
Every token issued through this gateway carries all four — there's nothing extra to wire up per project.
Stable model aliases
Call a name, not a vendor. The underlying model can change without touching your code.
MCP tools, built in
UNESDOC, the Data Hub, cross-platform Search, and the Administrative Manual — reachable as MCP tools from any model call, no extra integration.
mcp.unesco.orgEthics guardrails
Every request is screened by GLiGuard, grounded in the UNESCO Thesaurus and the 2021 UNESCO Recommendation on the Ethics of AI.
ai-guardrails.unesco.orgThree inference sources, one fallback chain
On-premise GPU (Ollama) first, HuggingFace cloud burst second, Azure-hosted APIs as the managed fallback — routed automatically per alias.
Monthly credit packages, reinvoiced per project
No individual vendor invoices to reconcile — usage is tracked per token, per model, per source, and settled monthly.
How it works
Your project is allocated a monthly credit package sized to expected usage. Spend against that package is tracked per API token, broken down by model alias and inference source (on-prem / HuggingFace / Azure). At month end, actual spend is reinvoiced to your project/sector code — the same cost-recovery model used elsewhere across UNESCO's shared services.
Getting a token
- Raise a Smart Ticket for the Data & AI team with your project name and sector/reinvoicing code.
- Tell us which aliases you need (e.g. Daily + Vision) and your expected monthly volume.
- You'll receive a scoped API token with its monthly credit package and budget already configured.
Starting an AI project?
Request an API token sized to your project's monthly credit package.
Raise a Smart Ticket
for the Data & AI team
