- Category
- AI Agents
- Rank
- No. 71Tools index
- Pricing
- Open Source
- Type
- APP
- Builder
- onyx-dot-app
- GitHub
- 32.0k stars
- Latest release
- cli/v1.4.1
- Date
About
Open source AI chat platform that works with any LLM and includes advanced features like RAG, web search, custom agents, and connectors to 40+ knowledge sources. Can be self-hosted in completely airgapped environments.
What it does
Onyx gives teams a shared workspace for asking models questions, researching across indexed company knowledge, running tools, producing files, and executing code. Users can create agents with their own instructions, sources, and actions. It supports hosted and locally operated models, while voice, image generation, web browsing, and downloadable artifacts broaden the interface beyond text chat.
Why it's ranked here
The strongest case is breadth backed by credible deployment and operations work. Onyx combines more than 50 indexing connectors, hybrid retrieval, multi-step research, external actions, code execution, and several model providers. It also documents authenticated metrics, connector health signals, audit events, load testing, and deployment choices. That makes it more convincing for team infrastructure than a chat interface with a few integrations attached.
What's good
The split between Lite and Standard gives evaluators a sensible adoption path. Lite runs with less than 1GB of memory and focuses on chat and agents. Standard adds keyword and vector retrieval, connector synchronization workers, model inference services, Redis caching, and MinIO storage. Security and operations receive concrete support through structured audit events, authenticated metrics, request tracing, connector-state monitoring, and load-test tooling with deterministic model behavior.
Tradeoffs
The complete product is a substantial distributed system. Standard operation introduces retrieval indexes, job queues, background workers, inference services, Redis, and object storage. Lite avoids that machinery but also omits indexed retrieval and connector synchronization. Community Edition includes the core chat, retrieval, agent, and action features, while capabilities aimed at larger organizations live in Enterprise Edition. The Python environment also requires version 3.13 and carries a large, tightly pinned dependency set.
How to use it well
Start with Lite when a team needs a hosted chat and agent interface without knowledge synchronization. Move to Standard when internal search, recurring connector ingestion, and larger-scale operation justify dedicated infrastructure. It best fits organizations willing to operate or deploy an application platform around several model providers. Lite does not cover the adjacent need for a full indexed knowledge layer, and Standard should not be treated like a small embeddable library.
Technical notes+
pyproject.toml requires Python 3.13, manages dependencies with uv, sets package = false, and separates backend, development, enterprise, load-test, and model-server groups. backend/onyx/main.py builds a FastAPI API server from many feature and administration routers, while backend/model_server/main.py runs a separate FastAPI inference service with Prometheus instrumentation and cgroup-aware Torch thread limits. package.json defines a private Bun workspace for widget, examples/widget, and desktop; widget/src/index.ts registers and exports the onyx-chat-widget custom element. cli/main.go and tools/ods/main.go provide Go command entry points. docs/METRICS.md documents authenticated API and MCP metrics, and docs/AUDIT_LOGGING.md specifies JSON audit events on the onyx.audit logger tree with Redis-backed deduplication.
Observed
- License
- Community Edition uses the MIT license; Enterprise Edition contains additional organization-focused features.
- Implementation languages
- Python powers the backend and model server, TypeScript powers the chat widget, and Go powers command-line entry points.
- Dependency and packaging surface
- The Python project requires Python 3.13 and uses uv for dependency management without building an installable Python package; frontend workspaces use Bun.
- Interfaces
- Onyx exposes FastAPI HTTP services, MCP integration, Go command-line tools, and an exported browser custom element.
- Deployment support
- Documented deployment targets include Docker, Kubernetes, Helm, Terraform, major cloud providers, and a hosted cloud option.
- Deployment structure
- Lite provides chat and agents with a smaller stack; Standard adds retrieval indexes, synchronization workers, inference services, Redis, and MinIO.
Read from README.md, Makefile, package.json, pyproject.toml, docs/METRICS.md, docs/AUDIT_LOGGING.md, cli/main.go, tools/ods/main.go, widget/src/index.ts, backend/onyx/main.py, backend/onyx/__init__.py, loadtest/mock_llm/app.py, backend/model_server/main.py.
What it can do
Chat with any large language model
Text prompts and LLM configuration → AI-generated responses
Search the web for current information
Search queries or prompts requiring web data → Search results and synthesized information
Retrieve and analyze documents using RAG
Questions and uploaded documents/knowledge base → Contextually relevant answers based on document content
Create and deploy custom AI agents
Agent configuration and behavioral parameters → Specialized AI assistants for specific tasks
Connect to external knowledge sources
Connector configurations and access credentials → Integrated data from 40+ knowledge platforms
Conduct deep research on topics
Research topics and parameters → Comprehensive research reports and analysis
Run in completely offline environments
Local LLM models and airgapped infrastructure → Functional AI chat system without internet connectivity
Tags
Tech Stack
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.