Makra delivers web content as structured data using an embeddings model, allowing AI agents to work directly with organized information. By managing context extraction and representation, it lets developers focus on building agent logic rather than data preprocessing, and the service can be accessed via a simple API for integration into AI workflows.
Funding
Funding not disclosed
Founders
Product
Problem
AI agents that need up-to-date web information typically fetch raw HTML and rely on large language models to parse it, which consumes high token counts and repeats costly inference for the same pages across requests.
Solution
Makra provides a memoization layer that converts webpages into structured JSON schemas and extracts requested data without repeated LLM inference. On the first visit, Makra analyzes a page’s DOM, generates a JSON Schema describing its full structure, and stores this representation. Subsequent extraction calls use the stored schema to perform direct DOM reads and semantic lookups, reducing per‑request cost to that of a simple embeddings lookup. The service exposes two APIs—`schema` for structure discovery and `extract` for targeted data retrieval—allowing AI engineers to integrate live web data into agents and pipelines with minimal token usage and consistent performance.
Target Audience
Makra is aimed at AI engineers and developers building agentic systems, batch extraction pipelines, or intelligent scrapers that require frequent, low‑cost access to structured web data.
Features
- `schema` API returns a complete JSON Schema of a page’s structure, including semantic field descriptions
- `extract` API delivers precise JSON data matching a provided schema, eliminating the need for post‑processing
- Memoization of page structures after the first analysis, enabling cost‑effective reuse across multiple requests
- SDK and quickstart guides for rapid integration with popular agent frameworks such as LangChain
- Supports autonomous navigation and pagination, allowing agents to scrape multi‑page sites intelligently