Skip to main content
Current limitations, known issues, and planned improvements for Guardway Gateway.

​
Provider limitations

​
Supported providers

Guardway’s gateway ships with presets for 20+ providers across chat, embedding, image, and audio. For the list shown in the dashboard’s provider picker, see Connect a provider. Beyond the picker, the gateway also carries adapters for specialty services — Voyage (embeddings), AssemblyAI (transcription), ElevenLabs (text-to-speech), and Fal.ai (images) — configured via environment variables. Anything else: use the Custom (OpenAI-compatible) preset for any endpoint that speaks the OpenAI API — self-hosted vLLM, partner gateways, private model servers, etc.

​
Missing Features

The following features are not yet implemented or only partially available. Workarounds are provided where possible.

​
1. Batch Processing API

Status: Not implemented No support for batch API. Process requests individually in parallel client-side as a workaround.

​
2. Vector Store Support

Status: Not implemented No built-in vector database integration. Use external vector databases (Pinecone, Weaviate, Qdrant, etc.) as a workaround.

​
3. Assistants API

Status: Partial implementation Thread management, code interpreter, and file search are not yet available. Implement assistant logic client-side as a workaround.

​
4. Fine-tuning API

Status: Not implemented Fine-tune directly with provider APIs as a workaround.

​
5. Multi-Modal Support

Supported: Text (chat, completions), Images (vision, generation), Audio (transcription, TTS) Not yet supported: Video processing, multi-modal embeddings, image editing.

​
Performance Considerations

Guardrail latency is the primary performance consideration. Disable unused guardrails and choose appropriate detection methods to minimize impact.
  • SLM guardrails add some latency per request. Use selective guardrails and adjust inspection direction for latency-sensitive workloads.
  • Streaming responses are not cacheable. Use non-streaming requests when caching is important.
  • Cache effectiveness depends on request similarity since caching uses exact request matching.

​
MCP Limitations

  • Only HTTP and stdio transports are supported (no WebSocket or gRPC)
  • Tool response size is configurable but limited by default
  • Sessions are not shared across gateway instances; use session affinity in load balancers

​
Caching Limitations

  • Cache uses exact request hash matching (no semantic similarity)
  • Streaming responses are not cached
  • In-memory cache size is limited; use Redis cache for larger capacity

​
Browser Compatibility

​
Admin UI Browser Support

Fully Supported:
  • Chrome 90+
  • Firefox 88+
  • Safari 14+
  • Edge 90+
Not Supported:
  • Internet Explorer (all versions)
Recommendation: Use latest Chrome or Firefox for best experience.

​
Future Improvements

Short-Term

  • Expand provider support
  • Optimize guardrail performance
  • Add batch processing API

Medium-Term

  • Vector store integration
  • Advanced routing strategies
  • WebSocket support

Long-Term

  • Semantic caching
  • Fine-tuning API
  • Multi-modal expansion

​
Reporting Issues

Found a bug or limitation not listed here?
  • Email: support@guardway.ai
  • Include: description, steps to reproduce, expected vs actual behavior, environment details, and logs/screenshots
Security Issues: Email security@guardway.ai (do not create public issues)