Soliplex is a self-hosted, AI-powered Retrieval-Augmented Generation (RAG) platform. It combines semantic document retrieval with generative AI to deliver accurate, cited responses grounded in your document collections. The ecosystem spans ingestion, indexing, retrieval, and user interface -- providing a complete pipeline from raw documents to conversational AI.
Full documentation: soliplex.github.io
graph TB
subgraph "User Interfaces"
FE["Frontend<br/>(Flutter / Dart)"]
TUI["Terminal UI<br/>(Textual)"]
end
subgraph "Backend"
SX["Soliplex<br/>(FastAPI)"]
MCP["MCP Server<br/>(FastMCP)"]
end
subgraph "AI & Retrieval"
PAI["Pydantic AI<br/>Agents"]
HR["Haiku RAG<br/>(LanceDB)"]
HS["Haiku Skills"]
end
subgraph "Storage"
LDB["LanceDB<br/>(Vectors)"]
PG["PostgreSQL / SQLite<br/>(State)"]
end
subgraph LLM["LLM Providers"]
OL["Ollama"]
OA["OpenAI"]
GE["Gemini"]
AN["Anthropic"]
VL["VLLM"]
end
subgraph "Ingestion Pipeline"
IA["Ingester Agents"]
SI["Hauku-Rag Ingester"]
DL["Docling"]
end
subgraph "Document Sources"
FS["Filesystem"]
SCM["GitHub / Gitea"]
WD["WebDAV"]
WEB["Web Pages"]
S3["S3"]
end
FE -->|"AG-UI Protocol"| SX
TUI --> SX
SX --> PAI
SX --> MCP
PAI --> HR
PAI --> HS
LDB --> HR
FS --> IA
SCM --> IA
WD --> IA
WEB --> IA
S3 --> IA
IA -->|"Discovers & fetches"| SI
SI -->|"Parse, chunk, embed"| DL
PAI <-->|"Inference"| LLM
SI --> LDB
SI --> PG
Soliplex (Backend)
The core backend service built with FastAPI and Python. Soliplex orchestrates AI agents, RAG retrieval, and real-time chat across isolated room environments.
- Multi-room architecture with independent configurations and knowledge bases
- Multi-provider LLM support (OpenAI, Ollama, Gemini, Anthropic)
- OIDC/JWT authentication with Keycloak integration
- MCP server and client for extended AI tool integration
- AG-UI protocol for structured frontend communication
- Quiz system with LLM-based evaluation
- 100% unit test branch coverage
A cross-platform Flutter application that serves as the primary user interface. Runs on Android, iOS, macOS, Web, Linux, and Windows.
- Multi-server OIDC authentication with secure credential storage
- Room-based chat with streamed AI responses and execution step visibility
- RAG document filtering with source citations and PDF chunk visualization
- File uploads, interactive quizzes, and network diagnostics
- Responsive Material Design 3 layouts for mobile through desktop
- Modular shell architecture with composable feature modules
A multi-source document discovery and delivery system. Ingester Agents connect to diverse origins and feed documents into the ingestion pipeline.
- Filesystem, WebDAV, GitHub, Gitea, web page, and S3 source support
- Declarative YAML manifests with cron scheduling
- SHA-256 content hashing for incremental sync (only changed files processed)
- Async-first architecture with retry logic and exponential backoff
- FastAPI REST API and Typer CLI
Concierge provides an extension for LLM-driven, on-demand room requests: users chat with a self-serve "about" room to request access to a non-public room or the creation of a new room, and the concierge files each request as a tracking issue on a Gitea repository for an administrator to resolve.
| Library | Role |
|---|---|
| Haiku RAG | RAG engine with hybrid vector + full-text search, agentic QA, and multi-provider embeddings |
| Haiku Skills | Modular skill system for Pydantic AI agents with sub-agent execution and AG-UI streaming |
| Docling | Document conversion and structural extraction for PDFs and other formats |
| Pydantic AI | Agent framework with function calling, tool integration, and multi-provider LLM support |
| LanceDB | Embedded vector database for semantic search and retrieval |
| FastMCP | Model Context Protocol server/client for AI tool interoperability |
Full configuration guides, tutorials, and API reference are available at soliplex.github.io.