Teams
Export your data
Download what Agentower keeps about your team's agents, routines, computers, keys, plan limits and you, as CSV or JSON.
History → Export lets you take out the data Agentower keeps — as CSV or JSON, from the page, the export API or the MCP server. It's metadata only: never chat text, code or messages, never a routine's brief, and never keys or notification addresses. The one exception is the work record of projects with Work details on: its instructions and the agent's summaries (first 600 characters, secrets masked) are in the two work datasets, for people who may approve the project.
On Solo, the audit log, guardrails and notification channels can't be exported — they're Team features.
Also not in an export:
- Agent chats — never exported, on purpose. Chat text is held only while the agent is open, then deleted (see Agent chats).
#What you can export
| Dataset | What's in it | Who can export it |
|---|---|---|
| Agents and subagents | One row per agent: project, person, tool, model, state and why, start, last signal and end, active time, tokens and ≈ cost | Everyone |
| State changes | Every change of state: when, which agent, from and to, and why | Everyone |
| Work record | What each agent did, step by step: files changed (lines added and removed, new or not), commands by kind, test results, commits, pushes, pull requests and each turn's totals, with the work stream's name. File paths, commit subjects, failed test names, branches, pull request links, instructions and the agent's summaries only for projects with Work details on, and only for people who may approve the project | Everyone (names and text: owners and the project's approvers) |
| Work summary by stream and week | The work record rolled up by work stream and ISO week (Monday to Sunday, UTC), as on History → Work: files created and edited, lines added and removed, commits, the last test result and whether it passed. With Work details on: the commit messages, the week's first instruction and the agent's latest summary (its own words, not verified) | Everyone (messages and text: owners and the project's approvers) |
| Daily usage | Per day, project, person and model: agents started and their hours (by each agent's main model), and tokens and ≈ cost by the model that served each request — a subagent on Haiku counts as Haiku | Everyone |
| Projects | Each project with its lane, settings and totals | Everyone |
| Guardrails | Every guardrail, its limits, and who last changed it | Everyone |
| Notification channels | The service, a hint of the address (never the address itself), events, last sent | Everyone |
| Routines | Each routine: name, project, the computer and person it runs for, schedule and time zone, permission mode, limits, on or off, whether its computer has approved the current version, and who made it and when. Never the brief (the instructions typed for the agent) | Everyone (allowed tools: owners and the project's approvers) |
| Routine runs | Every run of a routine, and every time skipped or blocked: when it was due, how it was started, its state and why, start and end, the agent it became, tokens and ≈ cost | Everyone |
| Computers | Each computer that reports to Agentower: its label, random ID and system, whose it is, first and last heard from, what it allows answers from Agentower to do and the app's switch, the hook's version and setup, when it last sent each kind of signal, and the name of the key it reported with. Never the computer's real name | Everyone (their own computers; owners everyone's) |
| Keys | Each Agentower key: its name, what it's for, whose it is, when it was created, last used, expires and was revoked. Never the key itself or any part of it | Everyone (their own keys; owners everyone's) |
| Plan limits | Each plan account's limits as the Usage page shows them (Claude, ChatGPT (Codex), …): the provider, your label for the account, whose it is, and per window (such as 5 h or Week) the percent used, when it resets and when the numbers last arrived. One row per account and window. Never the account's email or ID | Everyone (their own accounts; owners everyone's) |
| Audit log | Who did what, and when | Owners and approvers |
| Team members | Name, email, role, when they joined, weekly email on or off | Owners |
| My data | What Agentower keeps about you: name, email, when you joined, your role in each workspace, your settings (weekly email, background), and your profile picture's type, when it was last changed and the address to download it from while signed in. Your computers, keys and routines are in their own datasets | Everyone (only their own) |
#Download
On History → Export, pick a dataset, an optional date range (UTC, including the whole of the last day; empty means everything — Today, Week, Month, 6 months and Year fill the dates in for you), and CSV or JSON, then Download.
#Automate it
Create an export key on the same tab and fetch any dataset from a script — see Export API. An export key can only read exports, limited to what the person who made it is allowed to see.
Every export is recorded in the audit log.