Helmdeck is a self-hosted, containerized platform designed to enable 7B–30B-class open-weight AI agents to reliably perform complex tasks through schema-validated Capability Packs and native Model Context Protocol (MCP) integration.

How does Helmdeck ensure reliability for smaller AI models?

It uses 'Capability Packs' — schema-validated, one-shot JSON tools that encapsulate complex multi-step workflows (like browser or Git actions) into single, typed calls, making them reliably executable even by smaller LLMs.

What are 'Capability Packs'?

Capability Packs are the core of Helmdeck, offering schema-validated, one-shot JSON tools. They abstract complex operations across browser, desktop, Git, and code domains, allowing AI agents to perform tasks reliably and securely with simple, typed inputs.

What kind of AI models does Helmdeck support?

Helmdeck is optimized for 7B–30B-class open-weight models, ensuring high success rates. Its OpenAI-Compatible AI Gateway also supports various large language models including Anthropic, Gemini, OpenAI, Ollama, and Deepseek.

Is Helmdeck suitable for secure or air-gapped environments?

Yes, Helmdeck is specifically designed for local and air-gapped deployments. It features a Credential Vault to securely manage and inject sensitive credentials, protecting them from AI agents and ensuring data privacy.

Helmdeck

Name: Helmdeck
Author: tosin2013

0•

Provides a self-hosted, containerized platform designed to enable AI agents, particularly 7B–30B-class open-weight models, to reliably perform complex tasks through schema-validated Capability Packs and native Model Context Protocol (MCP) integration.

Helmdeck

0•

Helmdeck serves as a robust, self-hosted platform for AI agents, specifically optimized for local, air-gapped, and security-sensitive environments where reliance on frontier models is not feasible. It transforms intricate multi-step workflows across browsers, desktops, Git, and code into simplified, typed JSON calls known as Capability Packs, ensuring high reliability for smaller language models. The platform includes a credential vault for secure access to SaaS applications without exposing secrets to agents, an OpenAI-compatible AI gateway, and a management UI, making it ideal for self-hosted AI teams, the LocalLLaMA/Ollama community, and organizations prioritizing data privacy and control.

主要功能

01Capability Packs: Schema-validated, one-shot JSON tools for reliable agent interaction across various domains.

020 GitHub stars

03OpenAI-Compatible AI Gateway: Supports multiple large language models including Anthropic, Gemini, OpenAI, Ollama, and Deepseek with encrypted keys and fallback routing.

04Code Edit Loop & Repo Packs: Facilitates agent-driven code modifications, repository management, and version control operations.

05Self-Hosted & Containerized: Designed for local, air-gapped deployments with Docker Compose for development and Helm charts for Kubernetes production.

06Credential Vault: Securely manages and injects placeholder tokens for sensitive credentials, protecting them from AI agents.

使用案例

01Enabling security-sensitive organizations to deploy agents that interact with SaaS applications without exposing credentials.

02Empowering self-hosted AI teams with MCP-native infrastructure for private, VPC-bound agent deployments.

03Enhancing reliability of AI agents powered by 7B–30B-class open-weight models (e.g., Ollama) in local environments.