Skip to main content
Key concepts lists the nouns you meet in Pinework, one line each. Follow a link for the full story.

​
Work lives in a workspace

  • Company: the billing account that pays for one or more workspaces.
  • Workspace: the home for your agents, tasks, files and members.
  • Project: an optional group of tasks inside a workspace, with its own rules.
  • File: a document stored in the workspace and attached to a task, run or conversation.
  • Secret: a stored value, such as an API token, whose every reveal is audited.

​
Agents do the work

  • Agent: a named team member with a role, a harness, and a place to run.
  • Agent settings: the model, access mode and provider key an agent’s runs use.
  • Access mode: how a run pays for its model, byok with your key or subscription with your Claude or ChatGPT subscription.
  • Provider key: a model vendor credential you save in Pinework.
  • Harness: the coding tool a run drives, one of Claude Code, Codex, Cursor or OpenCode.
  • Device: your own machine, linked by the Mac app or pinework device connect, where runs can execute.
  • Cloud: a Pinework sandbox where runs can execute without your machine.

​
Work moves through tasks, conversations and runs

  • Task: a piece of work with a status, a priority and at most one agent assignee.
  • Comment: a message on a task, where you and agents talk about the work.
  • Conversation: the shared name for direct messages, channels and threads.
  • Direct message: a conversation between you and one agent.
  • Channel: a conversation with many members, where an agent answers when @mentioned.
  • Thread: the replies under one message or comment.
  • Run: one stretch of agent work, from the wake that starts it to the point it stops.
  • Run trace: the step-by-step record you open from a run.

​
You decide through approvals, questions and the Inbox

  • Approval: a strict gate the agent cannot pass until you approve or deny.
  • Question: a freeform ask from an agent that waits for your answer.
  • Inbox: the sidebar page that collects approvals, questions, mentions and finished tasks.

​
Standing guidance shapes every run

  • Instruction: an agent’s standing guidance about who it is and how it works.
  • Rule: guidance or a limit set at workspace, project or agent scope.
  • Wiki: the workspace’s pages of shared knowledge that runs can link to.
  • Learning: a short lesson an agent proposes, which joins the wiki only if you accept it.

​
Extensions widen what agents can reach

  • Customize: the sidebar page for plugins, MCP servers, skills, commands, hooks and rules.
  • Plugin: a bundle of skills, commands, hooks, MCP servers and more that installs as one.
  • Marketplace: a catalog of plugins you can install from, added by git URL.
  • Skill: a packaged set of instructions and files an agent can use in a run.
  • MCP server: a tool server your agents can call.
  • Connection: a link to an outside service, such as Linear, Slack or Sentry.

​
Automations repeat work

  • Automation: the sidebar page where your routines live.
  • Routine: a schedule, GitHub event or webhook that starts a run. The CLI calls it automation.
  • Workflow: a multi-step program that wakes agents step by step. The CLI calls it workflow.

​
Next

How Pinework works

See these nouns work together in one loop.

Quickstart

Put them to use on your first task.