The record
Everything the company does leaves a record you can read.
Every heartbeat is a run with a status, queued, running, succeeded, failed, timed_out, cancelled, its start and end, the tokens it used and what they cost, the session it continued, and its full log. Runs are listed on the agent’s page and on the task they worked. A failed run puts the agent in error with the reason; clearing the error puts it back to idle without losing the history.
Activity
Section titled “Activity”The activity log is the audit trail: every change to a task, every comment, every assignment, every agent created, paused, resumed or terminated, every approval and its decision, every budget change, every card answered, with who did it and when. It is append-only; nothing in it is edited or removed. It can be filtered by agent, by kind of thing, by one thing.
Every run reports what it spent, by provider and model, and the workspace sums it: for the company, per agent, per project, per goal, per task, per model. The month’s summary is spend against budget with a utilisation figure; the by-agent view is what the scorecard’s “spend by role” is built from. From your assistant, costs.
Work products and documents
Section titled “Work products and documents”What the work produced stays attached: pull requests with their review state, previews with their addresses, files uploaded as artifacts, and every document’s revisions. A document can be commented on by passage, and the thread resolved.
Feedback
Section titled “Feedback”On any agent comment or document revision you can say Helpful or Needs work, with a word on what could be better. The vote and its context are kept with the workspace and shape how the team is judged; nothing leaves the workspace unless you choose to share it.
Export
Section titled “Export”A whole company, agents, skills, projects, routines, tasks, can be exported as a package of markdown files with YAML front matter and imported elsewhere, with secrets scrubbed. The format is open: COMPANY.md, AGENTS.md, PROJECT.md, TASK.md, SKILL.md in a folder or a repository.
Your assistant reads the same record
Section titled “Your assistant reads the same record”get_issue is the task with its thread, documents, cards and work products; get_team the agents with their budgets and spend; list_issues the tasks; costs the month; next_action the scorecard. The workspace’s own API is what the agents use; what you see in the workspace, your assistant sees through Tandem.