# Gemina — Full Site Content > Gemina turns business documents into structured data you can search, total, and chat with. Document APIs for extraction, search & chat — through a REST API, five SDKs, MCP for agents (OAuth or API key), or a drop-in chat UI. Privacy-first, with a free FileTag tier (1,500 tags/month, no credit card). This file is the single-fetch full-content companion to https://www.gemina.co/llms.txt. It concatenates plaintext versions of the most important pages so a crawler can ingest Gemina's value proposition, product surface, and technical posture in one request. For the structured per-page index, see https://www.gemina.co/llms.txt. For MCP integration, see https://www.gemina.co/docs/mcp.md and https://www.gemina.co/.well-known/mcp.json. For machine-readable API docs (markdown), see https://www.gemina.co/docs.md, https://www.gemina.co/docs/mcp.md and https://www.gemina.co/docs/filetag.md. --- ## Homepage (https://www.gemina.co/) Eyebrow: Document APIs for extraction, search & chat Headline: OCR gives you text. Gemina gives you answers. Subhead: Upload any document. Get back clean, structured data. Then search it, total it, and chat with it. Through a REST API, five SDKs, MCP for agents, or a drop-in chat UI. Trust signals: - CCPA Compliant. - GDPR Compliant. - 99.9% uptime SLA. - 100M+ pages processed. - Trusted by product teams worldwide. Customer logos shown on the homepage include Ardani, Corz, EZMatch, Hash, Invoice4U, Jonnix, Morning, Rivhit, SmartBee, Smartbill, UCan2, WeBenefit, and iFreeLance. One document. Three ways to use it: 1. Extract. Get every field as typed JSON — headers, line items, and your own custom schemas. All major formats supported (PDF, PNG, JPEG, GIF, WebP, HEIC, AVIF — including iPhone camera photos, which are HEIC by default). Push results into your systems via webhooks or the REST API. 2. Search & Total. Find any document by meaning or by field. Get exact totals, computed in the database. 3. Ask. Your users ask questions in plain language. Answers come back grounded, with citations. More than an OCR — the difference. Three options measured against the same six parts (extraction, templates, search, exact totals, chat, tagging): - A plain OCR. You get text, that's it. None of the six. Still yours to build: everything after the text. - An LLM plus your own SDK. You get one answer per call — extraction, but only per call. Still yours to build: storage, search, chat and security. - Gemina. You get the whole pipeline: all six, production-ready. So you can build your product. Embedded chat for platforms and SaaS: a React component (@gemina/elements) plus scoped session tokens. Your users ask; your API key never reaches the browser. Agents & MCP: 13 MCP tools — extract, tag, search, aggregate, index — on one hosted server with discovery built in (FileTag 3, Extraction 7, Document Intelligence 3). Why the numbers are right: the database does the math, not the model. Totals run as database queries over extracted fields. Currency and units are never mixed. When data is missing, Gemina says so — it never guesses. Highlighted product surfaces: - Smart Templates. Upload a sample document, AI suggests an extraction schema, review and edit fields, then use the template ID in API calls. - Invoice Extraction. Vendor-agnostic. Pulls header fields (invoice number, dates, totals, tax, currency, vendor, payment terms) and line items (item ID, barcodes, description, quantity, unit price, tax, discounts). 90–100% accuracy depending on complexity and model selection. Header extraction from 4–6 seconds with Velox; other document types and models take longer. - FileTag — New. Free metadata tagging on every plan. Tag any PDF or image with structured metadata, smart filenames, and a downloadable enriched copy. Available via REST for backends, or via MCP for agents. 1,500 free tags per month, no credit card. - Document Intelligence — New. Opt-in per account. Every successful extraction is indexed into a searchable layer. Search across your whole document collection (structured filters, semantic, or hybrid), compute exact database-backed aggregations (never AI-estimated, never mixing currencies), and chat in natural language with grounded, cited answers. Past conversations are retained as chat sessions you can list, reread, and permanently purge, each with an auto-generated title and a retention-driven purge date. Available via the official SDKs (retrieval, chat, chat history, and browser-safe session-token minting), via REST, or as MCP tools (query_documents, aggregate_documents, index_document). Security and privacy posture (summary): - Your data is never used to train models. - Configurable retention periods. Built-in purge dates. - Choose the country where your data is stored. - AES-256 encryption at rest, TLS 1.2 or higher in transit. - Real-time admin dashboard with reports and full visibility. Capabilities: - Any language, with tuned support for Hebrew and English — print and handwriting. Automatic detection; right-to-left scripts handled correctly. - Handwriting recognition (ICR) for cursive, print, signatures, and form fills. - Auto-scaling — thousands of documents per minute at peak. - Per-field confidence on every extraction. When the system isn't sure, it says so. Calls to action: Start Free Trial (https://console.gemina.co/registration/create-account?planId=trial), View Features, View Documentation, Try Live Demo (https://www.gemina.co/demo/invoice). --- ## Product overview (https://www.gemina.co/features) Title: Extract it, search it, ask it, automate it. Tagline: Upload a document once. Everything the harness reads from it comes back four ways — typed JSON, search results, exact totals, and answers you can trace. The four products — one upload, four things to do with it. Each reads from the same extraction, so nothing is processed twice and nothing disagrees with anything else. 1. Extract (https://www.gemina.co/product/extract). Headers, line items, and your own schemas, returned as typed fields with a confidence score on every value. 2. Search & Analytics (https://www.gemina.co/product/search). Three ways to find a document, then sums, averages and counts computed in the database rather than guessed by a model. 3. Chat (https://www.gemina.co/product/chat). Questions in plain language, answered from the documents themselves, with the document IDs the answer came from. 4. Agents & MCP (https://www.gemina.co/product/agents). 13 tools on one hosted MCP server, so an agent can do everything above without you writing the glue. What's in the harness — why the four agree with each other Most AI extraction is a single LLM call wrapped in a prompt. Gemina is everything around it — specialized agents, validation layers, and compliance controls. One reading of the document, which is why search, totals and chat never contradict the JSON. 1. Agents that reason. Specialized agents that understand context, cross-check values, and refuse to guess — not a single model hoping for the best. 2. Works on any layout. No per-layout training and no templates to maintain. Upload a document and the harness handles it. 3. Compliance built in. Data residency by region and configurable retention. Your data never trains a model. How the harness is built: https://www.gemina.co/technology True of everything you send — these are not per-product features; they hold for every document that enters the harness. - Any language, any script. Latin, Cyrillic, Arabic, Hebrew and CJK, with tuned support for Hebrew and English. Mixed-language documents handled in one pass; right-to-left scripts read correctly. - Handwriting, not just print. Cursive and print handwriting, signatures, annotations and form fills, read alongside printed text in the same document. - Confidence on every value. High, medium or low per field, not per document. A missing field comes back empty and flagged rather than filled in with something plausible. - Speed you choose per call. Three model tiers on one parameter. Header extraction starts from 4–6 seconds on the fastest; the slowest trades time for accuracy on hard documents. - Your data stays yours. Choose the country your data is stored in, set automatic purge dates or delete via API, and your documents never train a model. - REST, five SDKs, or MCP. The same capabilities over a REST API, official clients for Python, Node.js, Java, C# and PHP, or a hosted MCP server for agents. Security and compliance detail: https://www.gemina.co/trust-center By document family — the API is the same for all of them. - Invoice & spend (https://www.gemina.co/solutions/invoice-spend). Key every invoice, then total the spend across all of them by vendor, category or period. - Inventory & procurement (https://www.gemina.co/solutions/inventory-procurement). Purchase orders, delivery notes and packing lists — what arrived, in what unit, at what price. - Contracts & legal (https://www.gemina.co/solutions/contracts-legal). Parties, dates, renewal terms and amounts, then ask the whole repository where a clause lives. - Identity & forms (https://www.gemina.co/solutions/identity-forms). Applications, intake forms and ID documents, typed — handwritten or printed, in any script. Building this into a product your own customers use: https://www.gemina.co/solutions/platforms Evaluation checklist: 12 questions to ask any extraction vendor, with Gemina's answer to each — https://www.gemina.co/features#checklist Category-level comparison against the alternatives: https://www.gemina.co/compare FileTag — the free tier below the four, shown under the product map rather than in it. A document's identity rather than its contents: vendor, date, type, total, a suggested filename, and an enriched copy, over REST or MCP, with 1,500 tags a month on every plan. Search, totals and chat are built on the full extraction, so FileTag is the step before that. See https://www.gemina.co/filetag. --- ## FileTag (https://www.gemina.co/filetag) Tagline: Tag, rename, and enrich documents from your agent. FileTag turns any PDF or image into structured metadata, smart filenames, and a downloadable enriched copy. Call it via MCP or REST. Free forever tier: 1,500 tags every month, no credit card required. One call, three outputs: 1. Tag. Structured fields — vendor, date, total, document type, document number — not free text. Example: vendor "Acme Corp", date "2026-02-15", total "7,200 ILS". 2. Rename. Six filename patterns returned. Your agent picks the one that fits your filing convention. Example: "2026-02-15_Acme-Corp_Invoice.pdf" plus five more patterns. 3. Enrich. Metadata embedded into PDF properties or image EXIF. Download via a short-lived signed URL (15 minutes). HEIC and AVIF are the exception: they come back renamed with embedded: false, because embedding would require re-encoding the photo. Read the embedded flag rather than assuming. Use cases for FileTag - Organize files. Your agent files user uploads to consistent folders — no custom parsers per document type. - Enrich RAG. Tag documents on ingestion so you can filter vector search by vendor, date, or type. - Route documents. Send receipts to accounting, contracts to legal, invoices to AP — by structured tag, not regex. - Tag uploads. Enrich user uploads in your product backend — cheaper and more consistent than per-document LLM calls. How it works — three lines, one tool call, structured JSON ``` # 1. Reserve an upload slot upload = mcp.call("files_create_upload", filename="invoice.pdf") # 2. PUT the file to the signed URL upload_to(upload.url, file_bytes) # 3. Tag it result = mcp.call("tag_file", file_id=upload.file_id) # → metadata, six filename patterns, enriched-file URL # Then: pull back the enriched copy whenever you need it enriched = download(result.enriched_file_url) ``` Alternative: use `tag_url` if your file already lives at a public HTTPS URL. Response shape: ``` { "document_id": "abc-123", "suggested_filename": "2026-02-15_Acme-Corp_Invoice_12345.pdf", "metadata": { "document_type": "invoice", "vendor": "Acme Corp", "date": "2026-02-15", "amount": 7200, "currency": "ILS", "document_number": "12345", "title": "Invoice", "tags": ["vendor", "invoice"] }, "filename_patterns": { "date_first": "2026-02-15_Invoice_12345.pdf", "type_first": "Invoice_12345_2026-02-15.pdf", "vendor_first": "Acme-Corp_Invoice_2026-02-15.pdf", "date_vendor": "2026-02-15_Acme-Corp.pdf", "vendor_date": "Acme-Corp_2026-02-15.pdf", "compact": "Acme-Corp_Invoice.pdf" }, "enriched_file_url": "https://api.gemina.co/files/tmp_abc123.pdf", "enriched_file_expires_at": "2026-02-15T12:15:00Z" } ``` Install — compatible MCP hosts: Claude Desktop, Cursor, Claude Code, VS Code, Cline, OpenClaw, Hermes-Agent, Codex, Windsurf, MCP Inspector, plus any client that supports Streamable HTTP MCP. Sign in with OAuth from Claude Code or claude.ai, or drop the snippet into your client config, paste your API key, restart. Setup snippets and platform-specific install instructions live on /product/agents (Connect section) and /docs/mcp. Privacy & retention - 7-day deletion. Files deleted within 7 days of upload (configurable per plan). - No model training. Your documents are never used to train AI models — Gemina's or anyone else's. - GDPR & CCPA compliant. AES-256 at rest, TLS 1.2 or higher in transit. Configurable data residency on paid plans. FAQ (selected) - What is MCP? Model Context Protocol — an open standard for AI agents to call external tools. Claude Desktop, Cursor, OpenClaw, Hermes-Agent and many others support it natively. - Do I need MCP, or can I just call REST? Both work; same endpoints under the hood. Over MCP you can sign in with OAuth instead of pasting a key. - I'm building a RAG pipeline / LangChain app — does this fit my stack? Yes. Call POST /v1/filetag per document during ingestion, attach metadata to your vector store entries, filter retrieval at query time. Cheaper and more consistent than per-document LLM calls. - What document types are supported? PDF, PNG, JPEG, GIF, WebP, HEIC, AVIF. Up to 50 MB. HEIC and AVIF files come back renamed with a suggested filename but without embedded metadata (embedded: false), since embedding would require re-encoding the photo. - What happens when I run out of free credits? Clear error. Upgrade to a paid plan for more, or wait until your monthly reset. - Can I use the same API key for both MCP and REST? Yes. One key, both surfaces. The same key also works for Data Extraction and other Gemina REST endpoints. Or sign in with OAuth from Claude Code or claude.ai — that mints a key for you. - Is there a rate limit? ~10 tags/second per API key, on top of the monthly credit cap. - What's coming next? Indexing across documents and natural-language queries. Sign up now — your existing API key will work for those when they ship. --- ## Technology (https://www.gemina.co/technology) Title: Inside the harness. Tagline: Specialized agents, validation, scaffolding — the layer above raw AI. Why a harness. Most AI extraction is one model call. When the model gets it wrong, you get a mistake. No safety net, no second opinion, no validation. The harness is everything Gemina puts around that call so production doesn't break. Specialized agents — not a single model — each contribute: extracting, validating, cross-checking, correcting. The harness is callable from both REST and MCP. One hosted MCP server exposes 13 tools: FileTag (3), Extraction (8), and Document Intelligence (3 — search, aggregate, index). An anonymous mount answers tools/list without credentials; running a tool needs an OAuth sign-in or an API key. Graph-based workflow orchestration - Composable workflows. Build complex extraction pipelines from simple, reusable components. - Checkpointing. Pause, resume, or debug extractions at any point. - State persistence. Full auditability of the decision-making process. - Extensible. Add new document types without rewriting core logic. Multi-agent reliability - Multi-agent validation. Results cross-verified before delivery. - Confidence scoring. Know how certain each extraction is. - Intelligent tiebreaking. Disagreements resolved systematically, not randomly. - Auditable decisions. Every extraction includes metadata explaining how results were determined. Defense-in-depth quality assurance - Automated validation via deterministic heuristics — no extra AI calls needed for validation. - Per-field confidence. Every field carries its own confidence score. Uncertain values are flagged for review, never silently guessed. - Deterministic corrections. Pure computational logic — the correction process itself cannot introduce new errors. - Every decision explained. Each extraction is logged with full metadata. Fault tolerance - Fault isolation. Component failures contained, not propagated. - Automatic retries with exponential backoff for transient failures. - Circuit breaking. Failing services isolated before they impact the system. - Dynamic routing. Documents routed based on characteristics and capacity. Performance and scale - Documents processed in seconds, not minutes. High-performance JSON serialization and response compression. - 99.9% uptime SLA. Globally distributed infrastructure, automatic failover, multi-region redundancy. - 100M+ documents processed across diverse industries. - Async-first architecture. Non-blocking I/O, intelligent connection pooling. Stack at a glance - API Layer: FastAPI with automatic OpenAPI docs, strict runtime type validation, high concurrency. - Task Processing: distributed task queue, horizontal scaling, automatic retries with exponential backoff. - Storage: multi-region cloud storage, automatic failover, configurable data residency, cryptographic integrity verification. - Caching: multi-tier with configurable TTLs, circuit breaking, graceful degradation. - Observability: structured logging with request correlation, real-time metrics, distributed tracing, health checks. - Security: defense-in-depth — JWT auth, API key management, IP allowlisting, encryption at rest and in transit. Models available across the harness: - Velox. Fastest, lowest cost per page. - Praetorian. Balanced accuracy and speed. - Invictus. Highest accuracy. --- ## Pricing (https://www.gemina.co/pricing) Title: Simple, Credit-Based Pricing. Monthly credit packages with no hidden fees. Use your credits anytime during the billing period. Free Trial. Full API access, all AI models included, no credit card required. https://console.gemina.co/registration/create-account?planId=trial Starter (free forever) - 1,500 FileTag tags / month. - Free forever. No credit card required. - Best for: trying FileTag, building agents on the free MCP tier, individual developers. - Sign up: https://console.gemina.co/registration/create-account Paid plans (live values fetched from the backend at request time — see /pricing for current numbers) - Bronze. Entry-level credit pack for the production harness. - Silver. Mid-tier; balanced accuracy and volume. - Gold. Higher volume, full feature set, bonus credits. Per-credit costs scale down with plan size. Credit costs per page vary by operation type (extraction, OCR, validation, correction, template build) and model selection (Velox = cheapest, Praetorian = balanced, Invictus = highest accuracy = highest cost). Templates also incur a per-field cost above 10 fields. Each plan also lists Document Intelligence terms when enabled: included chats per month (with per-chat overage pricing beyond that), per-document indexing cost, per-GB-month document storage cost plus an included storage allowance, and search included for free. Search is never metered (rate-limited only); indexing and chat overage are billed in credits. Enterprise. Pay-as-you-go with no commitment. Volume discounts available. Dedicated account manager. SLA guarantees. Data residency options (choose region/country for regulatory compliance). Contact sales: https://www.gemina.co/contact. --- ## Product pages ### Extract (https://www.gemina.co/product/extract) Title: Every field, typed. Every field, scored. - Typed JSON rather than a text blob: header fields and line items, each value carrying its own confidence (high / medium / low). - Refuses to guess — a field that isn't there comes back empty and flagged, not filled in with something plausible. - submit_extraction_feedback closes the loop: send verified values and get a per-field comparison of what was right, wrong or missed. - Templates are built from one sample, with no labelled training set; template fields are searchable and filterable like everything else. - Three model tiers behind one parameter: Invictus (highest accuracy, most credits), Praetorian (balanced default), Velox (fastest; header extraction from 4-6 seconds). Optional thinking, evaluation, correction and coordinate passes trade time for accuracy. - Size and page limits: extraction takes documents of 1 KB to 10 MB — note this is not the 50 MB FileTag ceiling. The page ceiling is per extraction type (custom_template 30, ocr 15, invoice_headers / invoice_line_items 10, FileTag 10); ask for several types in one request and the most permissive ceiling applies. Full table at https://www.gemina.co/docs.md ### Search & Analytics (https://www.gemina.co/product/search) Title: Find the right document. Calculate the right answer. - Three find modes: structured (exact filters over extracted fields), semantic (natural-language similarity over document summaries), hybrid (keyword + semantic fused with Reciprocal Rank Fusion — the best default for free text). - Exact aggregation: sums, averages, minimums, maximums and counts computed in the database, grouped by vendor, currency, document type, expense type, payment method, end user, month or year. - Response meta declares when a number would mislead: currencyGroupingAdded (money metrics were requested without a currency filter, so currency was added to the grouping), duplicatesIgnored (re-indexed documents collapsed), uomGroupingAdded (quantities in different units kept apart), linesWithoutUnitData (line items excluded from measurable sums). - Field catalog: ask the API which fields your documents actually carry and build filters from the answer; new document types become filterable without a deploy. - Scope, stated honestly: anything you extract is searchable and filterable, including custom templates and FileTag metadata. Money and measurable quantity are summable today. Numbers inside custom templates are searchable and filterable but are not yet available as aggregate metrics. - Search is never metered — rate-limited only. ### Chat (https://www.gemina.co/product/chat) Title: Ask your documents. Get answers you can trace. - Plain-language questions routed to the right engine (filter, search, or exact aggregation) and answered from what was found. Amounts come from the database, not the model. - Every answer returns the document IDs it used, so a UI can link to the source. - Ships as a drop-in React chat component (@gemina/elements, ) with citations, confidence handling and RTL support, or call the API and build your own. - Browser-safe scoped session tokens: your server exchanges its API key for a short-lived token pinned to one end user. Tokens can query, aggregate, chat and read history, but can never purge. - Chat history: sessions you can list, reread and permanently purge, each with an auto-generated title and a retention-driven purge date. - Worth knowing: indexing is opt-in and not retroactive (documents processed before you enabled it need a backfill); chat is REST/SDK, not an MCP tool; chats are included per plan with per-chat pricing beyond the allowance. ### Agents & MCP (https://www.gemina.co/product/agents) Title: Let agents work with business documents. - Connect section: sign in with OAuth from Claude Code (claude mcp add --transport http gemina https://api.gemina.co/api/v1/mcp/, then /mcp) or claude.ai / Claude Desktop (Settings → Connectors → Add custom connector), or paste an API key using the config snippet for your host (Claude Desktop via mcp-remote, Cursor, VS Code, Cline, Codex, Windsurf, OpenClaw, Hermes-Agent, MCP Inspector, curl). - 13 tools on one hosted MCP server, in three groups: FileTag (3), Extraction (7), Document Intelligence (3). - Anonymous discovery mount answers tools/list without credentials, so registries and clients can introspect before signing up. - Worked flow: ingest (files_create_upload, extract_document), find (query_documents), total (aggregate_documents), cite (get_document, get_extraction). - Chat is deliberately not an MCP tool — it carries conversation state a stateless tool call doesn't. Agents compose conversational answers from query_documents and aggregate_documents and keep control of the reasoning. ### Platforms & SaaS (https://www.gemina.co/solutions/platforms) Title: Your customers' documents, inside your product. - The B2B2B pattern: your server holds the API key and mints a session token scoped to the signed-in end user; the browser renders the chat component with that token; the component calls Gemina directly. The API key never reaches the browser. - A leaked token expires in minutes and can only see one end user's documents. - Whole integration is one server route plus one component — no vector store to run, no retrieval to tune, no chat state to keep, no citation UI to build. - Brandable: your own header text, header color and intro copy. ### How we compare (https://www.gemina.co/compare) Category-level comparison (plain OCR / raw LLM / cloud extraction API / document-processing platform), plus the developer-tool set described by their own live positioning captured 2026-08-12: Reducto, Extend, Chunkr, Nanonets, Unstructured, Sensible. Each is credited with what it is genuinely good at; none is given a feature-by-feature scorecard. Also states three cases where something else is the better buy: you need an AP workflow, you want the lowest cost per page at huge scale, or you already own the retrieval layer. ## Solutions by document family ### Invoice & Spend (https://www.gemina.co/solutions/invoice-spend) Invoices, credit notes, receipts, purchase orders, statements. Returns vendor and identifiers, dates, money (gross subtotal, discounts, rounding, each tax with its rate, total) and full line items, with per-field confidence. Then: exact spend totals per vendor, currency or month. Honest line: Gemina is the extraction layer, not an AP workflow tool — no approval routing, no three-way match, no ERP connectors, no review cockpit. ### Inventory & Procurement (https://www.gemina.co/solutions/inventory-procurement) Packing lists, delivery notes, bills of lading, customs declarations, goods received notes, supplier price lists. Line items carry units alongside quantities. Honest line: quantity totals group by unit of measure and are never summed across units; line items indexed before unit extraction shipped are reported in the response rather than silently excluded. ### Contracts & Legal (https://www.gemina.co/solutions/contracts-legal) Service agreements, SOWs, amendments, leases, NDAs, insurance policies. Returns parties, dates (effective, term, renewal, notice), commercials, and your own clause fields via a template. Honest line: this is extraction and retrieval, not legal advice — citations exist so a lawyer can check the source. ### Identity & Forms (https://www.gemina.co/solutions/identity-forms) Intake and registration forms, applications, claim forms, identity and license documents, surveys, consent forms. Handles handwriting and print, checkbox selections, and any language with tuned support for Hebrew and English. Honest line: identity documents are extracted, not verified — no issuing-authority check, no forgery detection, no identity verification. --- ## Trust Center (https://www.gemina.co/trust-center) Title: Trust Center — Enterprise-Grade Security From Day One. Security & compliance - GDPR Compliant. Full compliance with EU data protection. We support data subject access requests, right to erasure, data portability, and maintain transparent data practices. - CCPA Compliant. California Consumer Privacy Act. Rights to know, delete, opt-out. We never sell personal information. - Secure by Design. Security built into every architecture layer, not bolted on. Security-first principles from code to infrastructure. Data protection (defense in depth) - Encryption at Rest. AES-256. Encryption keys are managed by the cloud platform. - Encryption in Transit. TLS 1.2 or higher. HTTPS enforced. Regularly tested against industry standards. - Secure Infrastructure. Enterprise-grade cloud, automated security patching, network isolation, 24/7 threat monitoring. - Access Controls. API key management and IP allowlisting. You control who can access your data and from where. Privacy & control - No Training on Your Data. Documents never used to train AI models. Strict separation between customer data and model development. - Data Residency Options. Choose where your data is stored, to meet regulatory and compliance requirements. You pick the area when you set up the account; documents are processed and stored there and nowhere else, and AI processing happens in the same place as storage. The published matrix, as area → location: - North America → United States. Single region. - Europe → Frankfurt and Berlin. Dual-region; data is replicated across both. - Asia → Asia. Single region. - Middle East → Israel. Single region. - Configurable Retention. Set retention period; data automatically purged when it expires. Every document has a built-in expiry date. - Instant Deletion. Delete any document instantly via API or admin console. No waiting, no support tickets. Sub-processors (last reviewed August 2026) All document processing, storage, and AI inference happen inside Google Cloud. - Google Cloud — hosting, storage, and all AI processing; documents are processed and stored exclusively here. Regions per your data-residency setting. - Cloudflare — serves the website and console; traffic passes through its network in transit (TLS). Network and delivery only. - Paddle — billing and payments. Account and payment data only, never documents. - Sentry — error monitoring. Technical event data only; document content and personal data are excluded. - Google Analytics — website usage analytics. Site visitors only, governed by cookie consent. Transparency - Admin Console. Everything controlled from one dashboard. Retention policies, data residency, user management, real-time extraction view. - Real-Time Visibility. View all extractions as they happen. Track processing status, results, detailed reports. --- ## Contact - Email: info@gemina.co - LinkedIn: https://linkedin.com/company/gemina-ai - Console (sign-up, billing, API keys): https://console.gemina.co - Free trial: https://console.gemina.co/registration/create-account?planId=trial - Free FileTag tier (no credit card): https://console.gemina.co/registration/create-account - Contact form: https://www.gemina.co/contact (reCAPTCHA-protected — prefer info@gemina.co for programmatic outreach) --- ## Key URLs at a glance - Marketing site: https://www.gemina.co/ - Authenticated console: https://console.gemina.co - REST API base: https://api.gemina.co - API errors: every failure returns the same envelope as a success — status "failed", data null, and one entry in errors carrying a stable error_code (branch on that, not on the human-readable description). Sending no credential is a 401; sending an unrecognized API key is a 403. Rate limits and quota exhaustion are 429 with a Retry-After response header. Full reference at https://www.gemina.co/docs#errors - MCP endpoint: https://api.gemina.co/api/v1/mcp/ (Streamable HTTP, auth via OAuth 2.1 sign-in or X-API-Key / Authorization: Bearer) - MCP manifest: https://www.gemina.co/.well-known/mcp.json - Site index for LLMs: https://www.gemina.co/llms.txt - Markdown docs: https://www.gemina.co/docs.md (production harness API), https://www.gemina.co/docs/mcp.md (MCP server) and https://www.gemina.co/docs/filetag.md (FileTag) --- Last updated: 2026-08-25. Maintained at public/llms-full.txt. When pages above change materially, regenerate from the canonical source pages.