Skip to content

Security: Deathcharge/unified-llm

SECURITY.md

Security policy

Supported versions

No public release is currently supported. Security fixes target the latest repository revision until the first package release.

Reporting a vulnerability

Use the repository's private GitHub vulnerability-reporting flow when available. If it is unavailable, email support@samsarix.com with [SECURITY] in the subject and a minimal, non-sensitive summary requesting a private reporting channel. Do not include exploit details, credentials, prompt content, provider responses, or customer data in a public issue or initial email.

Include the affected version or commit, the smallest reproduction, impact, preconditions, and whether a provider account was charged. Use synthetic data and revoke any credential that may have been exposed.

Samsarix LLC does not publish a response-time or remediation-time service-level agreement for this prerelease package. Receipt and next steps will be communicated through the private channel.

Security model

  • Endpoint URLs, API keys, model IDs, custom headers, and custom provider adapters are trusted operator configuration.
  • Message and tool content can be untrusted and crosses the configured provider boundary as JSON.
  • Normalized provider text and metadata remain untrusted output; terminal and log consumers must encode control characters.
  • The core stores no prompts, responses, keys, usage records, or telemetry.
  • Remote provider retention, training, authorization, quotas, and billing remain operator responsibilities.
  • Retries are bounded and transient-only, but a lost response can still duplicate provider work.

See docs/ARCHITECTURE.md for detailed trust boundaries and controls.

There aren't any published security advisories