openmake_llm: Coordinate AI Models, Agents, and Tools

Overview

OpenMake is a self-hosted workspace for coordinating models, agents, and tools rather than just chatting with a single model. It routes work to models assigned to different capabilities, and can combine agent tasks, research, MCP tools, and sandboxed execution into a response.

It is aimed at developers and teams who want to run AI workflows on infrastructure they control, using local or open-weight models, hosted providers, or both. It makes most sense when you need several capabilities or persistent agent work; for simple model hosting or a basic chat interface, a narrower tool may be easier to operate.

Key Features

• Routes requests across models assigned to roles and capabilities such as planning, vision, speech, and image generation.
• Runs persistent agent tasks with pause, resume, checkpoints, approvals, and parallel sub-agents.
• Provides deep research with parallel searches and cited reports.
• Connects built-in tools and external services through MCP servers, skills, and extensions.
• Runs agent code and MCP servers in Docker sandboxes, with configurable isolation and approval policies.
• Supports local model endpoints, LiteLLM, BYOK providers, and an OpenAI-compatible API.
• Produces artifacts and exports reports to PDF or DOCX.

Use Cases

• A homelab operator can connect vLLM or Ollama and use a self-hosted workspace for chat, research, and agent tasks.
• A developer can delegate multi-step work involving web browsing, file edits, and code execution, with approval gates for sensitive actions.
• A research or knowledge-work team can gather information from several sources and produce cited, exportable reports.
• A team using several model providers can assign different models to specific capabilities and compare responses without relying on one model for every task.

Project Facts

• Language: TypeScript
• License: MIT
• Stars: 43
• Forks: 7
• Topics: ai-agent, ai-assistant, ai-workspace, deep-research, discord-bot, docker, homelab, litellm, local-llm, mcp, model-context-protocol, multi-agent, nextjs, open-source, open-weight, openai-compatible, self-hosted-ai, typescript, vllm
• Archived: false

Getting Started

On Linux or macOS, run the installer:

curl -fsSL https://raw.githubusercontent.com/openmake/openmake_llm/main/install.sh | bash

Then open the URL it prints, sign in, and connect a model endpoint. See the README and self-hosting documentation for prerequisites, setup options, and operations.

Alternatives

• lobehub: LobeHub centers on building and scheduling agent teams in a web app, while OpenMake emphasizes a self-hosted workspace for models, agents, and tools.
• swarmclaw: SwarmClaw focuses on persistent agent runtime features such as memory, schedules, and delegation; OpenMake brings chat models, agents, and tools together in a workspace.
• deer-flow: DeerFlow targets long-running research, coding, and content tasks using subagents and sandboxes, while OpenMake is a broader AI workspace for model and tool choices.
• TrueForge: TrueForge packages runtime components for operating an LLM agent, while OpenMake coordinates chat models, agents, and tools in a self-hosted workspace.

Considerations

• This is a substantial self-hosted stack, not a hosted service. The installer sets up the application and supporting services, including PostgreSQL and Redis; Docker and PM2 are among the stated requirements.
• The documented installer targets Linux and macOS. Windows use is through WSL2.
• A GPU is not required to run the application, but local inference depends on a model server such as vLLM or Ollama; hosted OpenAI-compatible endpoints are also supported.
• Some agent execution capabilities are opt-in and require building the relevant sandbox images. The README also marks features such as an execution graph and scoped memory as planned rather than shipped.
• The repository reports 43 stars and 7 forks, so prospective operators should assess its fit and operational requirements before adopting it for critical workloads.