Glossary

2 min read

The terms you will encounter in Kujira, in plain language.

• Agent — the AI worker that operates independently.
• Employee / person — the real people on your team with a Kujira account.
• Organization — your company within Kujira; everything belongs to it.
• Role — the set of actions a person can perform (owner, admin, operator, reader).
• Memory — what an agent knows and remembers.
• Skill — an instruction manual an agent can follow.
• External tool (MCP) — a connection to another system that you give an agent.
• Connector — one of your Google or GitHub accounts that an agent uses to act.
• Credential — a key stored in the vault that an agent uses without seeing it.
• Cron — a task an agent repeats on a schedule, or runs once.
• Channel — an agent's communication route (Telegram, email, webhook, and so on).
• Agent Desk — the shared place for handling external conversations: the agent replies and a person can take over when needed.
• Contact — a person who writes to your agents through a channel.
• Session — an independent conversation with an agent; several can run in parallel.
• Approval — when the agent stops and asks for permission to do something significant.
• Level — the importance an agent estimates (0–100) for each action, which determines whether it acts, asks, or waits.
• Compaction — the automatic summary of a long conversation, so each reply does not cost more than necessary.
• Maximum usage rate — the brake: tokens per hour an agent must not exceed.
• Service token — an access pass that lets another program talk to Kujira.
• Two-factor authentication — the six-digit code from your phone that protects your account.
• Server — the machine where agents run; you provide it.
• Supervisor — the program you install on that machine to connect it to Kujira.