# Gemina

> Gemina gives you document APIs for extraction, search and chat. Upload any document once; get back typed structured data, then search it, compute exact totals over it, and chat with it — grounded, with citations. Reachable over REST, five SDKs, 13 MCP tools (OAuth or API key), or a drop-in React chat UI. Privacy-first, with a free FileTag tier (1,500 tags/month, no credit card).

An OCR returns text. Gemina returns answers. One upload gives you four things through one API: typed extraction (headers, line items, and your own custom schemas, each field carrying its own confidence), hybrid search across everything you have extracted, exact aggregations computed in the database rather than estimated by a model, and grounded chat that cites the documents it used.

Why the numbers can be trusted: totals are SQL, not inference. Money metrics are never summed across currencies — ask without a currency filter and the server adds the currency grouping and reports `currencyGroupingAdded: true`. Re-uploaded documents are collapsed and counted in `duplicatesIgnored`. Line items with no unit data are reported in `linesWithoutUnitData` instead of being silently dropped. A response tells you when a number would mislead you.

**FileTag** is the free tier of the Gemina MCP server and REST API: it turns any PDF or image into structured metadata, smart filenames, and an enriched copy. Callable from Claude Desktop, Cursor, OpenClaw, Hermes-Agent, and any MCP-compatible host.

For a single-fetch dump of the full site content, see https://www.gemina.co/llms-full.txt. For a plaintext index of every page (with links to the markdown copies), see https://www.gemina.co/sitemap.md. For the MCP server reference, see https://www.gemina.co/docs/mcp (markdown: https://www.gemina.co/docs/mcp.md). For agent-ready install instructions, see https://raw.githubusercontent.com/tommyil/gemina-mcp/main/llms-install.md (canonical) or browse the repo at https://github.com/tommyil/gemina-mcp. For MCP-host discovery, see https://www.gemina.co/.well-known/mcp.json.

## Product

- [Homepage](https://www.gemina.co/): What you get back from one document — extraction, search and totals, and chat — plus how Gemina differs from a plain OCR and from an LLM you call yourself.
- [Extract](https://www.gemina.co/product/extract): Typed extraction depth — every header field and line item as JSON with per-field confidence, templates built from one sample, and three model tiers (Invictus highest accuracy, Praetorian balanced, Velox fastest) selected with one parameter.
- [Search & Analytics](https://www.gemina.co/product/search): Structured, semantic and hybrid search over everything you extract; exact sums, averages and counts computed in the database, with the meta flags that declare when a total would mislead. Includes the honest scope note: money and measurable quantity are summable today, custom-template numbers are searchable and filterable but not yet aggregate metrics.
- [Chat](https://www.gemina.co/product/chat): Plain-language questions answered from the tenant's own documents with citations. Covers the drop-in React chat component, browser-safe scoped session tokens, and chat history. Indexing is opt-in and not retroactive; chat is REST/SDK, not an MCP tool.
- [Agents & MCP](https://www.gemina.co/product/agents): The Gemina MCP server — connect with OAuth (Claude Code, claude.ai) or an API key, install snippets for every host, all 13 tools by group, the anonymous discovery mount, and a worked ingest-find-total-cite agent flow.
- [How we compare](https://www.gemina.co/compare): Category-level comparison plus the developer-tool set (Reducto, Extend, Chunkr, Nanonets, Unstructured, Sensible) described by their own live positioning as captured 2026-08-12, what each is good at, and three cases where something else is the better buy.
- [Invoice & Spend](https://www.gemina.co/solutions/invoice-spend): Invoices, credit notes, receipts and POs — vendor, dates, taxes and line items, then exact spend totals per vendor, currency or month. Gemina is upstream of an AP workflow tool, not one itself.
- [Inventory & Procurement](https://www.gemina.co/solutions/inventory-procurement): Packing lists, bills of lading and customs forms — line items with units captured alongside quantities, so quantities are summed per unit and never across units.
- [Contracts & Legal](https://www.gemina.co/solutions/contracts-legal): Parties, dates, renewal terms and commercials from agreements, then semantic search across the whole shelf. Extraction and retrieval, not legal advice.
- [Identity & Forms](https://www.gemina.co/solutions/identity-forms): Intake forms, applications and identity documents, handwritten or typed, any language. Documents are extracted, not verified against an issuing authority.
- [Platforms & SaaS](https://www.gemina.co/solutions/platforms): Embedding document chat in your own product — server mints a scoped session token, the browser renders the component, the API key never leaves your server.
- [Product overview](https://www.gemina.co/features): Map of the four products (Extract, Search & Analytics, Chat, Agents & MCP) and what holds for every document whichever one you use — language, handwriting, confidence, model tiers, residency.
- [Technology](https://www.gemina.co/technology): How the multi-agent architecture works — specialized agents, validation layers, and the orchestration pattern.
- [FileTag](https://www.gemina.co/filetag): Free tier of the Gemina MCP server and REST API — document tagging, renaming, and metadata enrichment. 1,500 calls/month free.
- [Invoice Demo](https://www.gemina.co/demo/invoice): Interactive in-browser demo — upload an invoice and see the extraction with line items, validation, and confidence indicators.

## Documentation

- [API Documentation (HTML)](https://www.gemina.co/docs): Official SDKs for TypeScript, Python, C#, Java, and PHP — process a document in one call, then search, chat, and manage chat history over your results. Raw REST API too, including the error envelope, status codes and `error_code` values, and the per-extraction-type page limits.
- [API Documentation (markdown)](https://www.gemina.co/docs.md): Same content, plaintext markdown — easier for agents to ingest.
- [MCP Server Docs (HTML)](https://www.gemina.co/docs/mcp): Endpoint, OAuth 2.1 or API-key authentication, all 13 tools by group, error codes, and a headless curl session.
- [MCP Server Docs (markdown)](https://www.gemina.co/docs/mcp.md): Same content, plaintext markdown.
- [FileTag Docs (HTML)](https://www.gemina.co/docs/filetag): FileTag REST API reference, the three FileTag MCP tools, and the LangChain recipe.
- [FileTag Docs (markdown)](https://www.gemina.co/docs/filetag.md): Same content, plaintext markdown.
- [Install guide for agents (canonical)](https://raw.githubusercontent.com/tommyil/gemina-mcp/main/llms-install.md): Step-by-step FileTag MCP install with copy-pasteable config snippets for Claude Desktop, Cursor, Claude Code, VS Code, Cline, Windsurf, Codex CLI, OpenClaw, Hermes-Agent, and curl smoke-tests. Hosted in the [gemina-mcp repo](https://github.com/tommyil/gemina-mcp) — single source of truth.
- [MCP manifest](https://www.gemina.co/.well-known/mcp.json): Machine-readable discovery file for Gemina's MCP server — all 13 tools, grouped by FileTag, Extraction, and Document Intelligence.

## Pricing

- [Pricing](https://www.gemina.co/pricing): Credit-based pricing for the production harness. Free FileTag tier with 1,500 tags/month, no credit card. Transparent per-page rates.

## Company

- [About](https://www.gemina.co/about): Mission, team, and the reasoning behind the harness approach to production AI.
- [Trust Center](https://www.gemina.co/trust-center): Security posture, data handling, compliance commitments, and privacy-first guarantees. Data residency is chosen per account from four named areas — North America (United States), Europe (Frankfurt and Berlin, dual-region), Asia, and the Middle East (Israel) — with storage and AI processing both staying in the area you pick.
- [Contact](https://www.gemina.co/contact): Sales, support, and partnership inquiries. Email: info@gemina.co. LinkedIn: linkedin.com/company/gemina-ai.
- [FAQ](https://www.gemina.co/faq): Common questions across general, FileTag, getting-started, technical, billing, security, and enterprise categories.

## Integration

- **Positioning**: Gemina is the extraction and intelligence layer. It sits upstream of an AP workflow tool or inside your own product; it is not a review cockpit with approvals and ERP connectors.
- **MCP server**: One hosted server at `https://api.gemina.co/api/v1/mcp/`, exposing 13 tools — FileTag (3), Extraction (7), and Document Intelligence (3: search, aggregate, index). An anonymous read-only mount at `https://api.gemina.co/api/v1/mcp/public/` answers `tools/list` without credentials, so registries and clients can introspect the catalog before signing up. Authenticate with OAuth 2.1 (sign in from Claude Code or claude.ai) or an API key. Config snippets for every host at /product/agents#install; reference at /docs/mcp.
- **REST API**: Authenticated with an `X-API-Key` header (FileTag also accepts `Authorization: Bearer`). Base URL and endpoint reference at /docs.
- **Sign up**: https://console.gemina.co/registration/create-account?planId=trial (free trial, no credit card for FileTag tier).
- **Console**: https://console.gemina.co (account, billing, API keys).

## Policies

- [Privacy Policy](https://www.gemina.co/privacy-policy)
- [Terms of Service](https://www.gemina.co/terms-of-service)
- [Cookie Policy](https://www.gemina.co/cookie-policy)
- [Refund Policy](https://www.gemina.co/refund-policy)

## Notes for AI agents

- All marketing pages are server-rendered HTML — content is available without executing JavaScript.
- Structured data is provided as JSON-LD (Organization, WebPage, SoftwareApplication, FAQPage, BreadcrumbList).
- The contact form at /contact is reCAPTCHA-protected; for programmatic outreach, prefer info@gemina.co.
- The console (console.gemina.co) is authenticated. Public marketing content lives at www.gemina.co.