Run your teams' Claude and Codex agents where IT can see them

Spaces are the access boundary. Anyone in the workspace can find and join a public Space, a private Space is limited to the people who have been added, and a private thread can be limited to its participants even inside a public Space. Channels organize work inside a Space; they aren't a separate permission boundary.

Connections are added once, as personal or Team connections, and adding one to the workspace doesn't give every Space access to it. A new Team connection starts as available to its owner and admins only. Sharing it gives the whole workspace, a Space, or a person Full or Read-only access, and the highest grant wins. Read-only limits AI to the tools an app marks as read-only, and to GET requests for a Custom API.

AI acts with the access of the person who started the work, and it doesn't gain broader access because a tool is connected somewhere else in the workspace. The connected service still enforces its own permissions, so for data that must never change, also connect an account that is read-only in the service itself.

type.com stores OAuth grants and encrypted API keys on its servers. Code automation runs get a short-lived token instead of a provider credential, custom API credentials go only to the approved API host, and MCP clients receive tool results, not tokens.

Each automation has its own isolated Linux sandbox for shell commands, Git, the GitHub CLI, and scripts, and commands in it get no type.com credential. Code automations, in early access, run in a new QuickJS VM for each run, with no fetch, file system, or shell.

type.com doesn't offer SAML single sign-on or SCIM provisioning on any plan yet. Sign-in is built on WorkOS, an identity platform that supports both standards, so if your policy requires them, ask type.com about timing. When an admin removes a member, that person's sessions and access tokens stop working immediately.

type.com holds a CASA Tier 2 certification from the App Defense Alliance, lab tested and verified by TAC Security, issued May 20, 2026 and valid until May 21, 2027. It is working toward SOC 2 with Vanta, with the audit planned to begin November 16, 2026, and has completed an independent third-party penetration test. The certificate, controls, and subprocessors are on the trust center.

There isn't a dedicated security audit log yet. type.com records each run's agent activity and tool calls, automation run history, skill version history, and document revisions in the product, and admins see usage and cost analytics, but that history can't be exported as a log yet.

Settings → Analytics puts Overview, Health, Usage, and Cost in one dashboard for admins and owners. Health shows execution outcomes, a Needs attention list of failed and blocked automations, including ones paused after repeated failures, each Space's activity, and the status of each connection. Select a Space to see its automations, connections, and activity.

Admins and owners can also ask a type.com agent to query the same data through read-only analytics tools. Admins can't open people's Sidekick conversations, artifacts, or personal automation details through Analytics; they see each Sidekick's owner-labeled activity, automation counts, and cost.

Health doesn't measure application uptime, missed schedules, queue delay, or integration sync freshness, and its coverage notes say so.

Scheduled runs use the automation creator's identity, model credentials, and billing context. That's why team automations should use organization connections, such as the CRM account a sales team shares, rather than one person's login. Admins can reassign an automation's owner when someone changes roles or leaves. The new owner needs access to the automation's Spaces, and reassigning doesn't copy the previous owner's personal credentials.

A schedule pauses after three failed runs in a row, and paused and failing automations show up under Needs attention in the Health tab, so someone can fix them before the team notices a missing report.

type.com meters AI usage at the model provider's published API rates. Requests that run through a person's connected Claude or ChatGPT subscription aren't metered against the workspace at all, and automations use their creator's subscription even when a teammate or an integration starts the run.

In the Cost tab, admins see AI spend, which runs used a personal Claude or Codex subscription, and which teammates' eligible runs fell back to workspace credits. Model choice is the biggest lever, so set cheaper default models and lower reasoning effort for Spaces that do routine work.

The workspace has one monthly allowance. Admins get a notice at about 80%, and at 100% new AI work pauses unless auto-upgrade is on, which moves the plan to its next price point after a short notice. The $750 Enterprise plan is the top price point, so above its allowance work continues and admins keep getting warnings. There are no per-person or per-team spending limits yet.

Self-serve plans cover 2 to 60 team members, from $50 a month for 2 to $750 a month for 60, and every plan includes every feature. Companies with more than 60 people, or with volume, security, or procurement requirements beyond these plans, use a custom contract.

Most companies start with the teams that already run recurring work, such as engineering, support, or revenue operations: one Space per team, organization connections for the systems they share, and the automations people already run in personal accounts moved into those Spaces.

Worth knowing before a pilot, especially if your security review requires a specific identity setup.