← Back to Discover

BlackVault

Self-hostable gateway that lets AI agents call OpenAI, Anthropic, Google or Nebius with revocable tokens, so the real API key stays server-side, with spend caps, rate limits and model allowlists.

AI security & guardrailsMCP servers
Visit website ↗

What it does

BlackVault stores your provider API keys in an encrypted vault and issues proxy tokens, prefixed bvt_, for your agents to use instead. An agent sends its request to BlackVault, which adds the real key on the server and forwards the call to OpenAI, Anthropic, Google AI or Nebius. Killing a token cuts off that one agent without rotating the underlying key or breaking others that share it. It offers an OpenAI-compatible chat completions endpoint that picks the provider from the model name and translates request formats, so tools such as LangChain, the OpenAI SDKs, Cursor or Cline can be pointed at it by changing the base URL and key. Per-provider proxy paths with streaming are available too, and BlackVault can also act as a remote MCP server with list_models and chat tools. Each token can carry limits: requests per minute and per day, a dollar budget, a model allowlist and an IP allowlist. Responses include rate-limit and cost headers, and the dashboard shows cost per session and an audit trail of actions such as keys created, sessions killed and budgets exceeded. The project is MIT licensed and built with Next.js, tRPC, Postgres and Redis, with Clerk handling sign-in. It can be deployed on Vercel or self-hosted on any Postgres; Redis is optional, but without it rate limits are off and budgets use the database total.

Capabilities

  • Proxy tokens keep the real provider key away from agents
  • OpenAI-compatible gateway routing to OpenAI, Anthropic, Google AI and Nebius
  • Per-token request limits, budget caps, model allowlists and IP allowlists
  • Kill switch to revoke a session, a key or everything
  • Remote MCP server exposing list_models and chat tools
  • Cost tracking per session and an activity audit trail

When to use it

  • You share one provider key between several agents and want to revoke a single agent without rotating it
  • You want hard spending or rate limits on an automated or experimental agent
  • You want one endpoint and one token for models from several providers
  • You need short-lived tokens for CI jobs or shared team access