Frequently Asked Questions

SotaDocs supports npm and PyPI packages, any public GitHub repository, websites/URLs, and local files. All content is converted to markdown for consistent processing.

No. We process content to create semantic indexes but never store raw source code. Private repo content is discarded after processing. Only the semantic embeddings and metadata are retained.

Sources are automatically synced on configurable schedules (hourly, daily, or webhook-triggered). You can also manually trigger a refresh at any time. The process detects changes and updates only what's new.

Summary: Prompt-optimized key points (low tokens, fast).

Full Pack: Complete documentation compiled (high tokens, comprehensive).

Vector Search: Ask questions, get cited answers (medium tokens, precise).

The blindspot report compares your model's likely knowledge cutoff (training data) with the current documentation state. It outputs a staleness score, list of changed/deprecated APIs, and suggested prompt patches.

Every retrieved snippet includes: source URL or file path, timestamp of last sync, version (for packages), confidence score, and section/heading context. This helps agents attribute information correctly.

Yes! Export options include markdown packs (.md), JSONL packs, and Skills.md files for agent startup injection. Downloaded packs work completely offline.

Skills.md files (like GEMINI.md, CLAUDE.md) are special instruction files that coding agents read at startup. SotaDocs generates these based on your indexed documentation.

SotaDocs works with any agent via MCP or API. We explicitly support Claude Code, Codex CLI, Gemini CLI, Cursor, and any MCP-compatible client.

Yes! Full REST API with source ingestion, query (all modes), export generation, sync control, and webhooks for events. SDKs available for Node.js and Python.

Free tier includes: 5 indexed sources, daily sync, 100 queries/day, summary and vector modes, public repos only, and community support.