Skip to main content
This page covers the architecture: how source views fetch data, how the FUSE layer serves files, and when to use local vs remote mode.

​
The core idea

Airstore turns your services—GitHub, Gmail, Drive—into files. This lets AI assistants work with emails, issues, and documents just as naturally as they work with code. Airstore filesystem Source views contain data from your integrations, materialized as files. Tools are executables that let you take actions (create issues, send messages).

​
Source views

When you create a source view in the dashboard:
  1. You describe what you want — either in natural language (smart mode) or with structured filters (query mode)
  2. Airstore translates it to the appropriate API query
  3. Results are fetched and cached
  4. When you mount, the view appears as a folder with matching files
Source views sync in the background, so data stays fresh. See Source Views for query examples and filter details.

​
How reads work

When Claude (or any program) reads a file from ~/airstore/:
  1. The FUSE layer intercepts the read request
  2. Request goes to the Airstore gateway
  3. Gateway returns cached data from the source view
  4. File contents arrive in milliseconds
You’re not making live API calls on every read—you’re reading from a cached, materialized view.

​
Tools

In addition to data, Airstore exposes MCP tools as executables: 
~/airstore/tools/github create-issue --repo=acme/api --title="Bug"
Tools let Claude take actions, not just read data. They output JSON and can be piped together.

​
Local vs Remote

Remote mode (default): Connect to the Airstore gateway at gateway.airstore.ai. Your source views and integrations are managed in the dashboard. 
airstore mount ~/airstore
Local mode: Run the gateway locally for development. Configure MCP servers in a YAML file. 
airstore mount ~/airstore --config config.local.yaml

​
Privacy and security

Your source data is never stored. When source views fetch data from your integrations (GitHub, Gmail, etc.), we query your services and cache results in memory for fast reads. The cache is refreshed periodically and cleared when you unmount. We don’t persist your documents to disk in our system. Uploaded content is isolated and encrypted. Any context you upload—skills, scripts, or other files—is stored in a per-workspace S3 bucket with encryption at rest. Each workspace is fully isolated.

​
Next steps

Ready to try it? Head to the Quickstart.