Hive documentation

Hive is a local-first command center for AI work. You can begin with one agent, develop a reusable Staff directory, coordinate specialists in a Colony, and turn successful work into pipelines, schedules, or webhooks.

New to the product? Follow Your first hour with Hive for an end-to-end tour.

Choose a starting point

Capability map

People and execution

• Staff — durable role profiles with personality, prompts, skills, tools, memory, model preferences, performance, and improvement suggestions.
• Agents and tools — individual runtime workers for chat and tasks.
• Colony missions — recipe-driven teams with a shared blackboard, explicit handoffs, and evidence.
• Pipelines and automation — repeatable multi-step execution, schedules, retries, and run history.

Models, tools, and connections

• Models and providers — local Ollama, Anthropic, OpenAI, Gemini, and LiteLLM.
• Skills and MCP — reusable operating guidance, built-in tool groups, and external MCP servers.
• Integrations and webhooks — inbound automation, GitHub access, tunnels, and delivery security.
• Discord bridge — route Discord conversations to Staff personas.

Operating Hive

Hive stores its state locally under ~/.hive by default. Read Security and architecture before enabling sandbox network access or remote exposure, and keep Backups and recovery close when upgrading.

Hands-on guide library

Agents

• Create and configure an agent
• Edit, tune, import, and export agents
• Write effective agent instructions
• Configure tools, skills, and memory
• Work with chat and sessions

Teams and workflows

• Build and develop a Staff member
• Run your first Colony
• Build and run a pipeline
• Schedule recurring work
• Trigger work with webhooks
• Connect and assign an MCP server
• Operate automation safely