Knowledge ingestion

Keeping Ada in sync with Mayday

BlaBlaCar's support content already lives in Mayday — both the public Help Center and the private, agent-facing knowledge base. Rather than duplicate any of it, Ada connects to Mayday through the Ada Knowledge API and keeps a live, always-current copy, so the AI agent answers from the exact same source of truth your team maintains today.

Public Help Center

Private knowledge base

Ada Knowledge API

How the sync works

A lightweight connector — built once, running on its own schedule — is the only new piece. It does three things:

Read

Pull the latest content from Mayday's Content API — both the public Help Center collections and the private knowledge base.

Map

Transform each Mayday article into Ada's format — title, body, URL, language, and which knowledge source it belongs to.

Upsert

POST new articles and PATCH changed ones to the Ada Knowledge API. Matching on Mayday's ID makes it idempotent — no duplicates.

Two sources, one agent

Mayday holds two very different bodies of content. Ada ingests them into separate knowledge sources so each keeps the right visibility and can be governed independently.

Public source

Help Center

Customer-facing FAQs, policies and travel guides. Powers instant generative answers to the everyday questions — cancellations, refunds, ID verification, luggage, pets.

e.g. "How do I cancel and get a refund?"

Private source

Internal knowledge base

Agent-only procedures, edge cases and escalation rules. Used by Ada to resolve correctly and consistently — never exposed verbatim to members, and governed by guardrails.

e.g. driver-dispute handling · goodwill exception policy

The Knowledge API call

Each article is mapped from Mayday and upserted into Ada. Keying on Mayday's own content ID means the same article updates in place — publish in Mayday, and Ada reflects it on the next sync.

SE Mayday → Ada Knowledge API (illustrative)

// 1 · Read the latest content from Mayday
GET https://public-api.mayday.fr/knowledge/contents?collection=help-center

// 2 · Map each Mayday article to Ada's article shape
{
  "id": mayday.contentId,              // idempotency key
  "name": mayday.title,
  "content": mayday.bodyHtml,
  "url": mayday.publicUrl,
  "language": "fr",
  "knowledge_source_id": "mayday-help-center"  // or "mayday-private-kb"
}

// 3 · Upsert into Ada — bulk, idempotent by id
POST https://blablacar-ai-agent-demo.ada.support/api/v2/knowledge/articles/bulk
Authorization: Bearer <ADA_API_KEY>
Content-Type: application/json

{ "articles": [ ... mapped Mayday articles ... ] }

🔒 Private content stays private

The private knowledge base is ingested into its own source with restricted visibility. Ada reasons over it to answer correctly, but it isn't surfaced word-for-word to members — the public Help Center remains the customer-facing voice.

What BlaBlaCar controls

Sync cadence — hourly, nightly, or triggered the moment content is published in Mayday.
Which collections feed Ada — start with the Help Center collections that drive the most contacts, then expand.
Public vs private scoping — each Mayday collection maps to a public or private Ada source, with its own visibility rules.
One source of truth — your team keeps editing in Mayday; no second content system to maintain, no copy-paste drift.

This shows one viable pattern. Ada can also ingest knowledge via a live website crawl or direct API upload — the Mayday connector is the recommended fit here because it keeps Mayday as the single system your team already works in.