// the find
Haohao-end/openagent
What if OpenAI Deep Research and Dify were one platform? OpenAgent — harness architecture for rapidly building vertical AI agents, with deep reasoning loops, visual workflows, RAG, and A2A delegation.
OpenAgent is a self-hosted AI agent platform in the Dify mold: visual workflow editor, multi-LLM support, RAG with Weaviate/FAISS, agent-to-agent delegation, and a Vue 3 frontend. It targets teams who want to build and publish vertical AI apps without writing everything from scratch. Think of it as an opinionated orchestration layer on top of LangGraph with a UI bolted on.
The provider coverage is genuinely wide — OpenAI, DeepSeek, Google, Grok, Moonshot, Ollama, and several Chinese providers all have proper adapters with YAML model configs, not a single hardcoded string in sight. The workflow node set is practical: LLM, tool call, dataset retrieval, HTTP request, code execution, if/else, template transforms — enough to cover most real pipelines without reaching for custom code. Docker Compose setup is straightforward: clone, copy .env, docker compose up, and you have Postgres, Redis, Weaviate, Celery, and Nginx running together. The security posture is better than average for this type of project — they've had a Host Header poisoning issue and an SSRF vulnerability responsibly disclosed and patched, and they have a SECURITY.md, which suggests someone is actually paying attention.
The README conflates marketing with documentation badly — 'Deep Research' is described at length but maps to a single `enable_deep_thinking` flag, and the actual behavior of that flag is nowhere explained. The skills catalog is a grab-bag of YAML manifests that mostly point at OpenAI-specific agent configs, so if you're running DeepSeek or Ollama you'll be rewriting skills from scratch. Flask with Celery is a reasonable stack but LangGraph already gives you a stateful execution model; using both adds complexity without an obvious payoff, and the architecture diagram doesn't clarify which layer owns what. At 825 stars with sponsor logos prominently placed for Atlas Cloud and Bloome, the incentive to maintain this independently of those sponsors is unclear — if Atlas Cloud's API changes or Bloome pivots, it's uncertain who picks up the maintenance.