Enable a single LightRAG server instance to serve multiple isolated workspaces
via HTTP header-based routing. This allows multi-tenant SaaS deployments where
each tenant's data is completely isolated.
Key features:
- Header-based workspace routing (LIGHTRAG-WORKSPACE, X-Workspace-ID fallback)
- Process-local pool of LightRAG instances with LRU eviction
- FastAPI dependency (get_rag) for workspace resolution per request
- Full backward compatibility - existing deployments work unchanged
- Strict multi-tenant mode option (LIGHTRAG_ALLOW_DEFAULT_WORKSPACE=false)
- Configurable pool size (LIGHTRAG_MAX_WORKSPACES_IN_POOL)
- Graceful shutdown with workspace finalization
Configuration:
- LIGHTRAG_DEFAULT_WORKSPACE: Default workspace (falls back to WORKSPACE)
- LIGHTRAG_ALLOW_DEFAULT_WORKSPACE: Require explicit header when false
- LIGHTRAG_MAX_WORKSPACES_IN_POOL: Max concurrent workspace instances (default: 50)
Files:
- New: lightrag/api/workspace_manager.py (core multi-workspace module)
- New: tests/test_multi_workspace_server.py (17 unit tests)
- New: render.yaml (Render deployment blueprint)
- Modified: All route files to use get_rag dependency
- Updated: README.md, env.example with documentation
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Set EMBEDDING_MODEL default to None
- Pass model param only when provided
- Let providers use their own defaults
- Fix lollms embed function params
- Add ollama embed_model default param
- Add model parameter to jina_embed
- Pass model from API server
- Default to jina-embeddings-v4
- Update function documentation
- Make model selection flexible
- Change build check from error to warning
- Redirect to /docs when WebUI unavailable
- Add webui_available to health endpoint
- Only mount /webui if assets exist
- Return status tuple from build check
Added support for structured output (JSON mode) from the OpenAI API in `openai.py` and `azure_openai.py`.
When `response_format` is used to request structured data, the new logic checks for the `message.parsed` attribute. If it exists, it's serialized into a JSON string as the final content. If not, the code falls back to the existing `message.content` handling, ensuring backward compatibility.
• Always append cell text to maintain columns
• Preserve empty cells in table structure
• Check for any content before adding rows
• Use tab separation for proper alignment
• Improve table formatting consistency
• Include tables in extracted content
• Maintain original document order
• Add spacing around tables
• Use tabs to separate table cells
• Process all body elements sequentially
Previously, configure_vchordrq would fail silently when probes was empty
(the default), preventing epsilon from being configured. Now each parameter
is handled independently with conditional execution, and configuration
errors fail-fast instead of being swallowed.
This fixes the documented epsilon setting being impossible to use in the
default configuration.
• Track if we acquired the pipeline lock
• Auto-acquire pipeline when idle
• Only release if we acquired it
• Prevent concurrent deletion conflicts
• Improve deletion job validation
