> ## Documentation Index
> Fetch the complete documentation index at: https://docs.decodalabs.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Security & Privacy

> How DECODA isolates malware, protects your data, and keeps analysis secure.

Security is the foundation of DECODA's architecture. Every design decision prioritises safe handling of malicious files and protection of user data.

## Sandbox Isolation

Every analysis runs inside an isolated **Firecracker microVM** powered by E2B. Each sandbox is:

* **Fresh** - A new VM is created for every analysis session. No data persists between sessions.
* **Network-disabled** - No outbound or inbound network access. Malware cannot phone home, download payloads, or exfiltrate data.
* **Non-root** - Analysis runs as an unprivileged user inside the VM.
* **Time-limited** - Tools have enforced timeouts (1-3 minutes by default for most tools, up to 13 minutes for heavier ones like Ghidra).
* **Concurrency-limited** - Maximum 10 simultaneous sandboxes to prevent resource exhaustion.

### Command Blocking

The sandbox blocks dangerous shell commands:

* `rm -rf /` and variants
* `curl | sh`, `wget | sh` (download-and-execute)
* `nc`, `netcat` (network tools)
* `sudo`, `su` (privilege escalation)
* `chmod 777` (permission weakening)

### Python Module Blocking

Python scripts cannot import modules that would break isolation:

* `subprocess` - No shell command execution from Python
* `socket` - No network access
* `requests`, `urllib` - No HTTP clients
* `ctypes` - No foreign function interface

<Info>DECODA performs both **static** and **dynamic** analysis. Dynamic tools such as `runSpeakeasy` (PE emulation), `runStrace` (ELF syscall tracing), and `runFakeNet` (network intent capture) execute samples - but only ever inside the isolated, network-disabled Firecracker microVM. Because the sandbox has no network access and is destroyed after each session, malware cannot phone home, persist, or escape.</Info>

## Data Protection

### File Storage

* Uploaded samples are stored securely and associated with your account
* Samples are deduplicated by SHA256 hash within your account
* File size limits are enforced per tier (30-200 MB)

### File Validation

Before any processing, uploads are validated for:

* **Path traversal** - Filenames are sanitised to prevent directory escape attacks
* **Zip bomb detection** - Archives are checked for decompression bombs before extraction
* **Size limits** - Enforced before and after extraction

### Encryption

* **VirusTotal API keys** - Encrypted with AES-256-GCM before storage
* **All traffic** - HTTPS/TLS in transit
* **Security headers** - All API responses include `X-Content-Type-Options: nosniff`, `X-Frame-Options: DENY`, HSTS, and `Cache-Control: no-store`

## Authentication & Access Control

### Authentication

DECODA uses **WorkOS AuthKit** for authentication, supporting:

* OAuth (Google, GitHub, Microsoft)
* SAML SSO for enterprise organisations
* Session management with device tracking

### User Isolation

Every database query enforces user isolation. You can only access your own:

* Samples and analysis results
* Chat conversations and messages
* Reports and artifacts
* Settings and preferences

There is no shared state between users. A database query that doesn't verify `userId` ownership will not return results.

## Rate Limiting

All API endpoints are protected by rate limiting via Upstash Redis:

| Endpoint                   | Limit               |
| -------------------------- | ------------------- |
| Ask mode                   | 60 requests/minute  |
| Agent mode                 | 20 requests/minute  |
| Triage (upload processing) | 30 requests/minute  |
| Report generation          | 10 requests/minute  |
| Other API routes           | Varies per endpoint |

Rate limiting degrades gracefully - non-critical endpoints fail open if the Redis connection is unavailable, while critical endpoints (agent, triage, chat, billing) fail secure and deny requests.

## Billing Security

DECODA uses a **fail-closed** billing model:

* Usage checks deny access on error (never fail-open)
* Credit deductions are atomic - if deduction fails, the request is denied
* Tier normalisation prevents privilege escalation - unknown tier values default to `free`
* Development mode requires both `NODE_ENV=development` AND `DEV_BILLING_BYPASS=true`

## What DECODA Does Not Do

* **No live network access** - Malware is only ever run inside the network-disabled sandbox, so it cannot phone home, download payloads, or exfiltrate data.
* **No data sharing** - Your samples and results are not shared with other users or third parties.
* **No sample retention beyond your account** - Deleting your account permanently removes all associated data.
* **No network access from sandboxes** - There is no configuration option to enable network access in analysis VMs.