AI features
Where the forage actually sits
Truffen is in the business of building a core AI/ML product for agents and GTM teams. Named jobs: search sized for a context window, scrape to schema, enrich with citations, monitors, and waterfall routing. This is model infra — not a hiring interviewer and not an agent framework.
AI feature · Waterfall routing
Source, then fallback, on one credit
For agents that cannot stall on a thin first hit
Waterfall routing picks a source, falls back, and still meters one credit. Each field carries a source, a confidence, and a URL the agent can show.
Objective → source 1 → fallback → typed row
AI feature · Search for agents
Ranked excerpts sized for a context window
For builders who POST a natural-language objective
Declarative search returns ranked excerpts, not a dump of HTML. Same credit meter as scrape, enrich, and monitor.
POST /v1/search → ranked excerpts
AI feature · Scrape to schema
JS pages and PDFs become markdown or JSON
For GTM and research desks that need structure
Messy DOM, JavaScript pages, and PDFs land as markdown or a JSON schema your tools already expect.
POST /v1/scrape → markdown / JSON schema
AI feature · Enrich and monitors
People, companies, and a watch on the ground
For desks that need citations, then a webhook when the page changes
People and companies from a name or a domain, with source citations. Monitors watch a query and POST a change object. Model families in plan: Claude 3.7 Sonnet, GPT-4o, Gemini 2.5, Llama 3 70B. Access: Bedrock, Anthropic, OpenAI Direct, Groq. Frameworks: LangChain and custom/homebrew. Truffen is not fine-tuning those models.
POST /v1/enrich · POST /v1/monitors
Marketing / analytics. Core AI/ML product. MVP stage — no invented customers, funding, or uptime claims. Feeds agents; does not ship an agent runtime.