Skip to content

OmniDev Platform ​

OmniDev — AI knowledge & coding workbench: a team-oriented workbench for chat, retrieval-augmented knowledge, and background coding agents.

Working Features ​

Chat ​

  • OpenAI, DeepSeek, Qwen, and Anthropic models, with streaming and non-streaming responses
  • Conversation management, multimodal images, and image generation
  • Attachments stored in MinIO and parsed with Apache Tika
  • Per-user AI provider configurations (BYOK); keys are encrypted when ENCRYPTION_KEY is set
  • External MCP tools over streamable HTTP

Knowledge ​

  • Knowledge bases, document upload and Tika parsing, and configurable chunk size/overlap
  • OpenAI, Gemini, or Ollama embeddings with pgvector similarity search and Redis caching
  • RAG context injected into chat
  • Embedding dimensions are stored per knowledge base and follow its embedding model

The gateway embeds the knowledge-engine Go module. The same engine is also available separately as an HTTP API and MCP server.

Background Agents ​

  • User-defined agents with a system prompt, maximum steps, and selected tools
  • Asynchronous, persistent runs with a plan → think/tool call/code execution → response loop
  • Workspace file operations and search, calculator, and code execution tools
  • Python, JavaScript/Node.js, Go, and shell execution in isolated, resource-limited Docker containers
  • Run polling and cancellation through the real HTTP API

Pull-request delivery is planned, not implemented.

Tech Stack ​

LayerTechnology
BackendOne Go 1.22+ binary in apps/gateway (Gin)
FrontendNext.js 15 / React 19 in apps/web
DataPostgreSQL 16 + pgvector / Redis 7
Files and parsingMinIO / Apache Tika
Agent runtimeThrowaway Docker containers

Scope ​

OmniDev is a modular monolith, not a 16-microservice platform. It does not include an online IDE, visual workflow canvas, billing or subscriptions, one-click deployment, a notification service, or a bundled Prometheus/Grafana/Loki/Jaeger stack. Kafka, Elasticsearch, Temporal, Kubernetes, and Helm are not part of the current architecture.

See Getting Started, Architecture, and API Reference.

Released under the MIT License.