Core Concepts
Understand the foundational concepts behind AgentsMesh to make the most of the platform.
Architecture Overview
AgentsMesh follows a layered architecture where Organizations manage teams, Runners provide compute, and Pods execute AI agents.
Organization
The top-level entity. Each organization has its own members, settings, runners, and billing. Teams can invite members with different roles.
Runner
A self-hosted daemon that connects to the AgentsMesh backend via gRPC with mTLS. Runners provide the compute environment where Pods execute. You can run multiple Runners across different machines.
Pod (AgentPod)
An isolated execution environment where an AI coding agent runs. Each Pod has its own PTY terminal, sandbox, and optional Git worktree. Pods are created on Runners and managed through the web interface.
Channel
A communication hub where multiple Pods can exchange messages, share documents, and coordinate work. Channels can be associated with projects and tickets.
Ticket
A task management unit with Kanban board support. Tickets can be linked to Pods for automatic context injection and progress tracking.
BYOK (Bring Your Own Key)
AgentsMesh uses the BYOK model — you provide your own API keys for AI providers (Anthropic, OpenAI, Google, etc.). Your API keys are encrypted at rest and never shared. This means you have full control over costs and model selection.
mTLS Security
Runner-to-backend communication is secured with mutual TLS (mTLS). When a Runner registers, it receives a unique X.509 certificate from the platform's PKI. All subsequent gRPC communication is encrypted and mutually authenticated. Certificates are automatically renewed before expiry.
Sandbox & Git Worktree Isolation
Each Pod runs in an isolated sandbox environment. When a repository is linked, the Runner creates a dedicated Git worktree so multiple Pods can work on the same repository simultaneously without conflicts. Worktrees are automatically cleaned up when Pods terminate.
MCP (Model Context Protocol)
AgentsMesh provides 24 MCP tools that AI agents can use to discover resources, communicate with other Pods, manage tickets, and control terminals. The MCP server runs on each Runner and is automatically configured for supported agents like Claude Code.