Taming enterprise AI sprawl with a private marketplace
HiMarket wraps LLMs, MCP servers, and agents into a branded, metered internal marketplace so engineering teams stop flying blind on AI usage and costs.

What it does
HiMarket is an enterprise-focused platform built on the Higress AI Gateway that lets companies run a private AI marketplace. Administrators package models, MCP servers, agents, and agent skills into standardized API products with authentication, rate limiting, and versioning. Developers browse an internal portal to subscribe, test tools in a chat interface called HiChat, or write code in a sandboxed online IDE called HiCoding. The stack includes metering, billing, and observability so finance and ops can see where the tokens are going.
The interesting bit
Instead of treating AI integration as a free-for-all of API keys and shadow IT, HiMarket builds an app-store model inside your firewall—complete with gray-scale releases, approval workflows, and OIDC integration. The README even recommends deploying itself via AI coding assistants like Cursor or Claude Code, which is either brilliantly meta or a quiet admission that the setup is nontrivial.
Key highlights
- Supports five marketplace types: Models, MCP Servers, Agents, Agent Skills, and Agent Workers.
- Built-in developer experience tools:
HiChatfor model comparison and MCP testing,HiCodingfor sandboxed “vibe coding” and human-AI pair programming. - Enterprise controls: token-based metering, full-chain observability with tracing and heatmaps, custom approval workflows, and portal branding.
- Multi-version support with gray-scale releases and quick rollback for AI products.
- Underlying infrastructure relies on
HigressandNacos; deployment options include local, Docker Compose, Helm, and Alibaba Cloud ComputeNest.
Caveats
- The project requires JDK 17, Maven, MySQL, and multiple frontend builds—this is not a lightweight single-binary install.
- The README explicitly recommends using AI coding tools to troubleshoot deployment, suggesting environment compatibility can be tricky.
- It is tightly coupled to the Higress ecosystem; if you are not already using or willing to adopt Higress and Nacos, this is a significant infrastructure commitment.
Verdict
This is for mid-to-large engineering organizations that need governance and cost visibility over a growing pile of AI services, not for hobbyists who just want to call an LLM. If your company is already running Higress, HiMarket is a natural extension; if not, weigh the infrastructure tax carefully.
Frequently asked
- What is higress-group/himarket?
- HiMarket wraps LLMs, MCP servers, and agents into a branded, metered internal marketplace so engineering teams stop flying blind on AI usage and costs.
- Is himarket open source?
- Yes — higress-group/himarket is an open-source project tracked on heatdrop.
- What language is himarket written in?
- higress-group/himarket is primarily written in Java.
- How popular is himarket?
- higress-group/himarket has 1.2k stars on GitHub.
- Where can I find himarket?
- higress-group/himarket is on GitHub at https://github.com/higress-group/himarket.