Frequently Asked Questions

Docker one-liner, single binary, or Helm chart - pick what fits your stack. See the Getting Started guide for a step-by-step walkthrough from install to first proxied request.

OpenAI, Anthropic, Azure OpenAI, Ollama, vLLM, and any OpenAI-compatible endpoint. If it speaks the OpenAI API format, VoidLLM can proxy it.

Yes. Cursor and Windsurf work as a standard LLM proxy - just change the base URL. Claude Code connects via the built-in MCP server. See the Connect guide for exact setup steps.

No. SQLite is the default and works well for most deployments. PostgreSQL is optional and recommended only for multi-instance setups with shared state.

No - and not as a configurable setting. There is no content logging code to enable. Prompts and responses pass through memory only and are never written to disk, logs, or the database. Read more about the zero-knowledge architecture.

The zero-knowledge architecture means no personal data from prompts is processed or stored by the proxy. Only metadata is tracked: who made the request, which model, how many tokens. This significantly reduces the compliance surface area compared to proxies that log content.

VoidLLM is infrastructure, not an AI system. The built-in usage tracking provides traceability metadata (who, when, which model, how many tokens) that supports the logging requirements the AI Act defines for deployers. Read our full analysis in the EU AI Act and LLM Proxies post.

Under 500 microseconds at 2000 RPS. In-memory auth lookup and async usage logging keep the hot path lean. See our proxy overhead benchmarks.

Yes. Four strategies: round-robin, least-latency, weighted, and priority. Automatic failover with circuit breakers included in the free tier. More on load balancing and failover.

AI agents write JavaScript to orchestrate multiple MCP tool calls in a single WASM-sandboxed execution. Reduces token usage and round-trips. Read about Code Mode and WASM sandboxing.

Yes. vLLM, Ollama, or any OpenAI-compatible endpoint works. Point VoidLLM at your local or on-prem deployment and it routes requests identically to managed providers.

Yes. A built-in MCP server exposes 6 management tools and a gateway for external MCP servers with scoped access control per org and team.

The Community tier is free with no time limit. It includes load balancing, MCP gateway, RBAC, rate limiting, circuit breakers, and the full admin UI. No feature is artificially time-limited or crippled.

Unlimited organizations and teams, cost reports with model breakdown, cross-org analytics, and priority email support. $49/mo or $490/yr.

Everything in Pro, plus SSO/OIDC (any provider), audit logs, OpenTelemetry tracing, and multi-instance support with Redis. $149/mo or $1,490/yr.

Business Source License. Source available, free to self-host for any purpose - including commercial use within your organization. The one restriction: you cannot take the code and sell it as your own hosted LLM proxy service. Every release converts to Apache 2.0 four years after its release date.

A one-time $999 payment for lifetime Enterprise access - all current and future features, no recurring fees. It also includes Product Advisory Board membership and direct founder access via Slack. Spots are limited.