Neural Networks and Model Providers
Understand how Hive relates to coding agents, language models, and the providers you configure.
Hive is the runtime around your coding agents. It does not train, host, or choose a neural network for you. Your agent CLI—such as Claude Code, Codex, OpenCode, Gemini CLI, or another supported tool—connects to the model and provider you configure for that tool.
Hive manages the terminal and agent process ↓Agent CLI handles prompts, tools, and provider credentials ↓Model provider runs inferenceThis separation means you can use different agents and providers side by side. Start each agent with Hive’s --kind option, or run its command inside a pane. Configure API keys and model preferences through the agent’s own supported setup; Hive does not store or proxy those provider credentials.
For example:
hive agent start --kind claude --env apphive agent start --kind codex --env app --worktree review/changeHive focuses on the operational layer: keeping terminal sessions alive, showing agent state, directing prompts, and restoring sessions. Read Agents for supported CLIs and Orchestration for coordinating agents.