LeadPaca: data sourcing and privacy
Last updated 20 November 2026. This page describes the listed API (leads-api, v0.2). It applies on top of the general privacy policy.
1. What a scan reads
| Source | Kind | Access | What is read |
|---|---|---|---|
| Xiaohongshu | public social | third-party data API (paid, keyed) | public posts matching the pack's keywords, their public comment threads and replies, and the public profile page (bio, follower count) of accounts that commented |
| Douyin | public social | third-party data API (paid, keyed) | public videos matching keywords and their public comment threads |
| Google News, Bing News | news | public RSS, no key | headlines and snippets, then the linked publisher article |
| World Bank procurement | procurement | open JSON API, no key | published procurement notices, including the contact the notice itself names |
The service reads only content its authors have made public on these platforms. It does not log in to any platform, does not read private messages, group chats, or login-walled networks (LinkedIn and WhatsApp are explicitly not reachable), and does not buy, import or resell contact lists. The live source list is published by the API at GET /v1/leads/sources.
2. What a lead contains
- the platform and public account handle (and the public display name) of a party that revealed a commercial role in a post or comment;
- a verbatim quote of that public post or comment and its public URL, so the claim can be checked at the source;
- the role, region, capacity, timeline or budget only as stated by the author;
- contact paths that the author published (for example a contact method written in a bio or post), ranked by reachability;
- two scores with written rationale, and a status:
verified(the extraction was checked against the source text) orneeds_human(rule-based only).
The service does not infer or output special-category information (health, religion, political opinion, ethnicity, sexual orientation and so on), does not profile private individuals outside the stated commercial context, and does not combine sources to de-anonymise anyone. Hub accounts and pasted advertisements are flagged rather than treated as leads.
3. How the model is used
Extraction and translation run on Google Vertex AI (Gemini) in the Tokyo region. The model receives the text of individual public posts and comments, returns structured fields, and every returned lead is verified against the source text before it is marked verified. Prompts and outputs are not used for model training. The project-only preview operation uses no model at all.
4. Retention
- Scan results are kept for 24 hours so the buyer can poll and download them, then deleted automatically (
TASK_TTL_HOURS=24). - No raw platform data is kept beyond the running scan. No database of people or accounts is built across scans.
- Logs hold metadata only (task id, timings, source counts, spend), never post content or handles.
- Buyers are identified only by the paying wallet address in the x402 payment payload; there are no accounts.
5. Legal basis and the buyer's obligations
LeadPaca processes publicly available business-context information for the legitimate interest of B2B lead discovery, with data minimisation (only stated facts, only what the author published, short retention) and transparency (every lead cites its public source). The buyer, as the party that contacts a lead, is responsible for complying with the laws that apply to their outreach, including the GDPR, China's PIPL, Japan's APPI and local anti-spam rules, and with the terms of the platform on which the contact was published. The API guidance states this to every calling agent.
6. Removal requests
Anyone who appears in a result, or a platform acting for them, can ask for removal. Write to the contact address on the About page with the platform and handle. The account is added to a suppression list within 7 days and excluded from all future results; because results are deleted after 24 hours, no stored copy remains to erase.
7. Changes
New sources (for example TikTok or WeChat public articles, which the invite-only web beta is trialling) are added to the listed API only under the same rules: public content, no login, no private channels, short retention, and an update to this page.
