# Gemini Gems File Limit: 10-File Upload Cap and Knowledge Base Solutions

The Gemini Gems file limit caps custom knowledge bases at 10 reference files and 100 MB per document. While attached knowledge persists across Gem interactions, active chat sessions also face rolling upload caps and context window constraints. When your reference library exceeds 10 files, external retrieval and workspace indexing provide an alternative to manual document merging.

Source: https://fast.io/resources/gemini-gems-file-limit/
Author: [Derek Labian](https://fast.io/authors/derek-labian/)
Last reviewed: 2026-10-08

## What Is the Gemini Gems File Limit and Upload Cap?

The Gemini Gems file limit restricts custom Gems to a maximum of 10 uploaded knowledge files at up to 100 MB per document as of October 2026, according to Google's official Gemini documentation. The Gemini Gems file limit defines the maximum number and size of reference documents that can be attached to custom Gemini Gems for instruction grounding.

Custom Gems allow users to configure customized system instructions alongside grounding files. When configuring a Gem in the web interface, the builder provides a Knowledge section where you can attach reference documents. These files ground the assistant's replies, allowing it to cite internal guidelines, technical specifications, policy handbooks, or operational procedures.

Supported document types include plain text files, Markdown documents, PDFs, Microsoft Word files, and CSV spreadsheets. Code files across common programming languages can also be uploaded. Video recordings and standalone audio tracks cannot be uploaded into a Gem's persistent Knowledge section, which is restricted to text and document formats.

Gemini maintains an internal activity storage quota that is completely separate from your standard Google Drive or Google Workspace storage allocation. When previous chats, uploaded media, and custom Gems consume this internal space, Gemini blocks further uploads and returns the error message: "Delete data to upload file". Resolving this error requires deleting older chat conversations or removing unused Gems to free storage capacity.

## How Do Knowledge Base Files Differ from Active Chat Uploads?

Understanding Gemini's file constraints requires distinguishing between persistent files attached to a Gem definition and dynamic files uploaded during an ongoing chat conversation. Many users confuse these two upload surfaces, leading to unexpected errors during multi-turn document review.

Files added to a Gem's Knowledge section remain attached across every conversation. Every user who chats with that Gem receives answers grounded in those reference documents. Google Gemini restricts knowledge files to 10 files per prompt with a maximum file size of 100 MB for documents. You can upload them from local disk storage or select them directly from Google Drive. When linked through Google Drive, Gemini reads the latest file revision, provided Workspace extensions remain active in your Google account settings.

In contrast, files uploaded during an active chat session with a Gem are ephemeral. Users can attach up to 10 files in a single prompt turn, with each document capped at 100 MB for standard file uploads. However, active chat sessions face rolling usage restrictions. When analyzing multiple files across successive prompts, users encounter the rate-limit warning: "You've reached your limit for chats with files". This rolling window throttles automated document review until the temporary cooldown expires.

Context window dynamics also influence session uploads. While Gemini 1.5 Pro and Gemini 2.0 Flash feature large context windows, attaching dense documents rapidly consumes active context memory. If attachments approach context thresholds, Gemini produces the warning: "Your uploads may be too large for the best results". In long conversations, the assistant may miss fine connections or lose track of earlier instructions as the context window fills with raw text.

## How Do Custom Assistant Knowledge Limits Compare Across Platforms?

Every major AI assistant platform imposes distinct boundaries on reference documents, file sizes, and context grounding. Choosing the right platform depends on your corpus size, document update cadence, and team collaboration requirements.

| Assistant Platform | Knowledge File Cap | Maximum File Size | Grounding Ceiling | Verified Date |
|---|---|---|---|---|
| Google Gemini Gems | 10 files | 100 MB per file | 10 files (bounded by context window) | 2026-10-08 |
| Anthropic Claude Projects | Unlimited files | 30 MB per file | Bounded by Claude context window | 2026-10-08 |
| OpenAI Custom GPTs | 10 files | 512 MB per file | 2 million tokens per document | 2026-10-08 |

Anthropic Claude supports up to 20 files at up to 500 MB per chat, while Project files allow up to 30 MB per file with an unlimited file count bounded by the context window. This makes Claude Projects suitable for collections of smaller text files, provided the total content does not exceed the model's active context memory.

Both Google Gemini Gems and OpenAI Custom GPTs enforce a fixed 10-file ceiling on persistent knowledge. For OpenAI Custom GPTs, reference files are also capped at 10 knowledge files, though OpenAI has scheduled the retirement of Custom GPTs across ChatGPT plans in favor of workspace plugins, with retirement dates running from December 2026 to February 2027 depending on workspace tier. Google has likewise announced that Gems on personal accounts will transition toward skills in late 2026.

For teams managing tens or hundreds of reference documents, a 10-file ceiling quickly becomes an operational barrier. Legal teams reviewing contract portfolios, engineering teams maintaining multi-service documentation, and operations groups with extensive standard operating procedures cannot fit their knowledge assets into 10 discrete files without significant restructuring.

## Why Does Concatenating Files and Context Stuffing Fail?

When encountering the 10-file ceiling in Gemini Gems, users frequently attempt workarounds such as merging multiple documents into single large files, pasting raw text into instructions, or uploading ZIP archives. While these approaches may offer temporary relief, they introduce severe technical and operational drawbacks.

Google Gemini inspects the internal contents of uploaded ZIP archives. Packaging documents into an archive does not bypass system boundaries, because Google Gemini inspects archive contents and counts each document toward the 10-file cap. Packing dozens of technical documents into a single archive will fail ingestion when the internal count exceeds the platform limit.

Pasting documentation directly into the Gem instruction box creates a different failure mode. System instructions exist to define persona, response format, tone, and operational boundaries. Filling the instruction prompt with raw reference text dilutes the model's attention, causing it to ignore critical behavioural directives while struggling to parse unstructured data.

Concatenating multiple documents into a single consolidated file creates three primary complications:

1. Attention degradation. When a model scans hundreds of pages concatenated together, retrieval accuracy drops on specific factual lookups. Key clauses in the middle of a massive file can be overlooked during synthesis.
2. High update friction. If a single procedure or API endpoint changes, the entire concatenated document must be regenerated, re-exported, and re-uploaded to the Gem. When multiple team members make updates, keeping merged files synchronized becomes impossible.
3. Token consumption and latency. Stuffing an entire corpus into the active prompt forces the model to process thousands of unnecessary tokens on every query, driving up processing latency and exhausting session memory.

## What Architectural Patterns Support Large Knowledge Corpora?

Teams requiring access to large document collections need a structured architecture rather than manual file stitching. Three distinct patterns exist for connecting assistants to extensive knowledge bases.

The first approach involves self-hosted vector databases, such as Pinecone, Qdrant, or Chroma. This pattern gives developers full control over chunking strategies, embedding algorithms, and similarity thresholds. However, it requires writing custom ingestion scripts, maintaining infrastructure, and building dedicated retrieval pipelines. For non-technical teams, this overhead is difficult to maintain.

The second approach uses Google Drive folder integration through Workspace extensions. While this allows Gemini to read files stored in Drive, it relies on Gemini's general search heuristics rather than structured workspace indexing. It does not provide granular team access controls, audit logs, or structured data extraction across heterogeneous files.

The third approach decouples knowledge storage from the AI assistant by using an external intelligent workspace. Instead of forcing files into a 10-file assistant definition, your documents reside in a shared, organization-owned workspace on Fastio. You can upload files directly or import them from Google Drive, Microsoft OneDrive, Dropbox, or Box.

When you turn on Intelligence Mode on a Fastio workspace, files are indexed automatically for hybrid search, combining full-text indexing, semantic retrieval, and metadata value matching. Rather than stuffing 200 documents into an assistant's static context, the assistant queries the workspace on demand through the Model Context Protocol (MCP) and retrieves only the specific passages required for the query.

For teams working with structured document sets like contracts, invoices, or specifications, [Metadata Views](/product/document-data-extraction/) turn unstructured documents into a live, queryable database. Users describe target fields in natural language, and Fastio extracts typed data into a sortable table without requiring manual OCR configuration.

## How Do You Connect AI Assistants to External Workspaces via MCP?

The Model Context Protocol provides an open standard for connecting AI assistants directly to external workspaces. Instead of managing file uploads inside individual chat apps, assistants query remote storage securely over Streamable HTTP.

The Fastio MCP server is hosted at `mcp.fast.io`. Different types of AI clients connect through dedicated endpoints:

* Claude desktop and general MCP assistants connect to `https://mcp.fast.io/mcp/tools`.
* Coding assistants in terminal and IDE environments connect to `https://mcp.fast.io/mcp/code`.
* ChatGPT and Codex connect through the Fastio plugin in the plugin directory, or via custom MCP configuration at `https://mcp.fast.io/mcp/operations`.

In standard MCP client configurations, adding the connection requires defining the remote HTTP server:

```json
{
  "mcpServers": {
    "fastio": {
      "url": "https://mcp.fast.io/mcp/tools"
    }
  }
}
```

When the client connects, users authenticate via OAuth 2.0 in the browser. A Review Permissions screen allows you to select Read Only or Read & Write permissions and choose exactly which organizations and workspaces the assistant can access. Complete setup instructions and client guides are available at [Fastio MCP Documentation](https://mcp.fast.io/docs).

Once connected, the assistant queries the workspace using the `storage` tool with the `search` action. When a user asks a question, the assistant searches the indexed workspace, receives relevant excerpts with exact file references, and synthesizes an accurate answer.

Shared workspaces also support collaborative workflows between humans and AI agents. Files maintain complete per-file version history, and all workspace actions are recorded in an append-only audit log. An agent can create a workspace, organize reference files, and transfer ownership to a human team member while retaining administrative access.

Monthly plans start with a 30-day free trial (credit card required). Creating an account is free; doing real work requires an organization on a paid subscription. Plans on [Fast.io pricing](/pricing/) are Starter at `$9.99/mo`, Business at `$49.99/mo`, and Enterprise at `$199.99/mo`. By pairing model intelligence with external workspaces, teams eliminate artificial upload barriers and keep reference documents continuously updated.

## Frequently asked questions

### How many files can you upload to a Gemini Gem?

Google Gemini allows up to 10 files per prompt with a maximum file size of 100 MB for documents and standard files. If you need a Gemini Gem to reference more than 10 files, you must either combine documents manually or connect an external workspace via the Model Context Protocol.

### What file types does Gemini Gems support?

Gemini Gems support standard document and text formats, including PDF, DOCX, TXT, Markdown, CSV, and common source code formats. While general Gemini chats support video and audio files, these media formats cannot be added to a Gem's persistent Knowledge section.

### How do you give a Gemini Gem access to more files?

To expand access beyond the 10-file cap, you can link files from Google Drive using Workspace extensions, concatenate multiple text documents into consolidated files, or connect your assistant to an external intelligent workspace using MCP. An intelligent workspace indexes hundreds of files and provides semantic search on demand.

### What is the maximum file size limit for Gemini Gems?

Google Gemini enforces a maximum file size of 100 MB for documents and standard files uploaded to a Gem. This limit applies to both direct local file uploads and files linked from Google Drive.

### What does the Delete data to upload file error mean in Gemini?

This error indicates that you have reached the internal storage quota for saved Gemini Apps activity, which is separate from your Google Drive storage limit. To clear space and resume file uploads, delete previous chat conversations with large file attachments or remove unused custom Gems.

### Can Gemini Gems access Google Drive folders directly?

Gemini Gems can reference individual files selected from Google Drive when Workspace extensions are active. When you attach a Drive file, Gemini queries the latest saved version. However, Gems cannot automatically index entire nested folder structures without individual file selection.

## Sources

- [Google Gemini Apps Help: Upload & analyze files in Gemini Apps](https://support.google.com/gemini/answer/14903178): Google Gemini allows up to 10 files per prompt with a maximum file size of 100 MB for documents and standard files.
- [Anthropic Help Center: Upload files to Claude](https://support.claude.com/en/articles/8241126-upload-files-to-claude): Anthropic Claude supports up to 20 files at up to 500 MB per chat, while Project files allow up to 30 MB per file with an unlimited file count bounded by the context window.

## About Fast.io

Fast.io provides shared workspaces where people and AI agents work on the same files, with built-in semantic search and citation-backed chat over what they hold. Agents reach it through a remote MCP server, a REST API at https://api.fast.io/current/, and a command line client published on npm as @vividengine/fastio-cli. MCP setup is at https://mcp.fast.io/docs: Claude and most MCP clients connect to https://mcp.fast.io/mcp/tools, ChatGPT to https://mcp.fast.io/mcp/operations, and coding agents to https://mcp.fast.io/mcp/code.