# voidllm

> Privacy-first LLM proxy and AI gateway — load balancing, multi-provider routing, API key management, usage tracking, rate limiting. Self-hosted. Zero knowledge of your prompts.

- **URL**: https://www.freshcrate.ai/projects/voidllm
- **Author**: voidmind-io
- **Category**: Infrastructure
- **Latest version**: `v0.0.19` (2026-05-20)
- **License**: NOASSERTION
- **Source**: https://github.com/voidmind-io/voidllm
- **Homepage**: https://voidllm.ai
- **Language**: Go
- **GitHub**: 65 stars, 6 forks
- **Registry**: github (`voidmind-io/voidllm`)
- **Tags**: `ai`, `ai-gateway`, `anthropic`, `api-gateway`, `docker`, `go`, `golang`, `helm`, `kubernetes`

## Description

Privacy-first LLM proxy and AI gateway — load balancing, multi-provider routing, API key management, usage tracking, rate limiting. Self-hosted. Zero knowledge of your prompts.

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v0.0.19` | 2026-05-20 | High | ### Fixes - Admin TLS configuration (`server.admin.tls`) is now actually applied in dual-port mode. Previously `tls.enabled: true` was a silent no-op - the schema and validation existed but no listener consumed the cert/key. In single-port mode (admin sharing the proxy port) configuring TLS now emits a WARN since external termination is expected there. Thanks to @martinsotirov for the fix (#92)  Full changelog: https://github.com/voidmind-io/voidllm/blob/main/CHANGELOG.md |
| `v0.0.18` | 2026-05-13 | High | ### Fixes - Startup panic when Code Mode is disabled - default config caused a nil pointer dereference of `SchemaTTL` in `app.go`. Thanks to @kernelb00t for the report (#87) and @SAY-5 for the fix (#90)  Full changelog: https://github.com/voidmind-io/voidllm/blob/main/CHANGELOG.md |
| `v0.0.17` | 2026-04-29 | High | ### Features - Code Mode response unwrapping - MCP ToolResult wrapper is stripped before results reach JS, so scripts work with plain values not protocol envelopes (#73) - Code Mode output schema inference - return types are learned from the first successful tool call and persisted, surfacing as TypeScript via `search_tools` (#73) - Code Mode tool descriptions rewritten with STRONG PREFERENCE / WORKFLOW / PATTERNS guidance to push LLMs toward chained calls and reduce sequential round-trips (#73) |
| `v0.0.16` | 2026-04-12 | Medium | ### Features - Model fallback chains - cross-model failover when all deployments of the primary are unavailable (Enterprise, #45)   - Configurable chain depth via `settings.fallback_max_depth`   - Per-hop access control enforcement   - Cycle detection at config, API, and runtime   - Usage events track both requested and served model name   - UI: Fallback Model dropdown in model create and edit dialogs   - UI: depth-0 warning when fallback is configured but disabled  ### Fixes - Flaky MCP usage d |
| `voidllm-0.0.16` | 2026-04-12 | Medium | Privacy-first LLM proxy and AI gateway with load balancing, RBAC, MCP gateway, and built-in admin UI. Self-hosted, single binary, sub-500us overhead. |
| `v0.0.15` | 2026-04-07 | Medium | ### Features - Configurable data retention for usage events and audit logs (#46)   - Opt-in background cleanup job with per-table retention durations   - Dialect-aware SQL for correct SQLite and PostgreSQL behavior   - Batched deletes with single-column timestamp indexes - Admin UI update notification via GitHub release check - PostgreSQL migration locking via advisory lock prevents concurrent-migration races (#48)  ### Improvements - Batch dependency updates: grpc 1.80.0, OpenTelemetry 1.43.0, |
| `voidllm-0.0.15` | 2026-04-07 | Medium | Privacy-first LLM proxy and AI gateway with load balancing, RBAC, MCP gateway, and built-in admin UI. Self-hosted, single binary, sub-500us overhead. |
| `v0.0.14` | 2026-04-04 | Medium | ### Features - MCP OAuth Client Credentials auth type with token URL auto-discovery (#49) - Google Gemini and Vertex AI provider adapter (8 providers total) - MCP usage dashboard with tabbed layout - Overview, LLM, MCP (#44) - Binary deployment documentation for Linux, macOS, Windows  ### Improvements - Shared credentials warning banner in MCP server dialogs - Windows binary pauses on error to show message before closing - 42 new tests for MCP usage, handlers, and health checker  Full changelog: |
| `voidllm-0.0.14` | 2026-04-04 | Medium | Privacy-first LLM proxy and AI gateway with load balancing, RBAC, MCP gateway, and built-in admin UI. Self-hosted, single binary, sub-500us overhead. |
| `v0.0.13` | 2026-04-04 | Medium | ### Features - MCP server health indicators in UI with auto-refresh (#43) - Standalone binary support for Windows, Linux, macOS (#50) - Cross-platform binaries in GitHub Release pipeline - License instance identification via heartbeat - Bench metrics sampler with realistic streaming scenario  ### Improvements - Comprehensive logging review: audit coverage for MCP, SSO, license, settings - Key cache log noise reduced (INFO to DEBUG) - Rate limit and token budget violations now logged - Migration |

## Citation

- HTML: https://www.freshcrate.ai/projects/voidllm
- Markdown: https://www.freshcrate.ai/projects/voidllm.md
- Dependencies JSON: https://www.freshcrate.ai/api/projects/voidllm/deps

_Generated by freshcrate.ai. Indexes github releases for AI-agent ecosystem packages._