AI and API access
Base2026 API & AI Access
Public read-only files and agent-readable entry points for using Base2026 as an attributed source intelligence library.
Base2026 is built to be useful to humans and AI agents without forcing either one to scrape the visual interface. The public site exposes a small read-only data layer with the same public boundary as the website: reviewed source records, public passages, creator/topic metadata, and source-backed intelligence only.
Private research vaults, raw captions, raw ASR, media files, credentials, logs, and unreviewed pipeline artifacts are not part of the public API surface.
Public entry points
- Human search workspace:
/ - Agent-readable context file:
/llms.txt - Public data dictionary:
/data-dictionary.json - Public API index:
/api-index.json - Public sitemap:
/sitemap.xml - Research-to-audit bridge:
/apply-research.html
Static public data
These files are safe read-only exports for external tools, notebooks, audits, and AI workflows.
/static/manifest.json- release counts and public export metadata./static/documents.jsonl- source-level search documents and public source records./static/passages.jsonl- searchable public evidence passages linked to source records./static/insight_cards.jsonl- reviewed public source-backed insight cards./static/topic_signal_briefs.jsonl- deterministic summaries for strong topics./data-dictionary.json- field descriptions and public/private boundary notes.
Search endpoint
The public UI and compatible integrations use a read-only Meilisearch-compatible Cloudflare Worker API backed by D1 FTS5:
POST /api/search/multi-search
No browser key is required. The endpoint does not expose Meilisearch credentials, private data, raw captions, or write access. External integrations should prefer the static JSONL files unless they need live search ranking.
AI usage
AI agents can use Base2026 as a source-backed research layer by reading /llms.txt, then using the data dictionary and public JSONL files. When citing Base2026, prefer canonical source, topic, creator, or comparison pages rather than copying raw rows.
For business-specific implementation, use /apply-research.html as the public bridge from Base2026 source intelligence to Alex Yarosh's AI Visibility Snapshot, Diagnostic Audit, and service workflow.
Good use cases:
- find creators who discuss a topic;
- compare repeated tactics across creators;
- inspect source-backed SEO/GEO/AEO claims;
- build internal research notebooks from public records;
- cite a canonical Base2026 source page with original creator attribution.
Not supported:
- raw transcript harvesting;
- creator impersonation;
- private lead or admin data access;
- writes, corrections, or moderation through public endpoints;
- replacing the original creator channel.
Future MCP contract
The planned MCP layer should remain read-only at first:
- search sources by query and filters;
- get one source by item ID;
- get one topic by topic ID;
- list creators with filters;
- compare creator viewpoints for one topic.
Every tool response must preserve attribution, canonical URLs, original source links, and public/private policy flags.