Skip to content

Spaces and Settings

A space is the workspace where intents run. It represents a product, service, or feature area and groups together:

  • One or more git repositories (the code host backing the agents' workspace)
  • Members with space-level roles
  • Tracker bindings (GitHub Issues, GitLab Issues, Jira Cloud)
  • A published managed environment for new intents
  • Runtime settings — which agent CLI runs, which models, how parallel construction behaves
  • The space's intents and their history

Earlier name

Spaces were called projects in earlier releases. API routes and stored identifiers keep that name, for example /projects/{projectId}/… and the projectId field, so you will still see it in API calls, logs, and the data model. Old /project/… links in the application redirect to the matching /space/… page.

Creating a space

From the dashboard, choose New Space. Enter a name, pick the code host (GitHub, GitLab, Bitbucket or AWS CodeCommit), connect your account if prompted, and select the repository. CodeCommit needs no connection step: once a platform admin has set the CodeCommit connection, you pick repositories from the usual list; if it is not configured yet, you can still select it, but a note asks you to contact a platform admin and Next stays disabled. You can optionally enable the provider's issue tracker in the same step. See Git and Tracker Integration for details on connections and auth modes.

Roles and permissions

Two independent layers govern access:

Space roles

Every space has members with one of two roles, managed in Space Settings → Members:

Role Can do
Owner Everything, including managing all members, all settings, and deleting the space
Member Create and start intents, answer gates, steer runs, participate in discussions

Non-members have no access to a space's intents, discussions, or real-time channels.

Existing space Admins become Owners during update. Space Owner and Platform Admin remain separate roles. Several Owners are allowed; removing or demoting an Owner requires another Owner in the space.

The platform-admin role

Platform-wide administration is gated by the Cognito platform-admin group — separate from space roles. Platform admins see two extra entries in the sidebar:

  • Workflows — the block library and workflow composer. Reading workflows is open to everyone; creating, forking, and editing requires platform-admin.
  • Admin — the Platform Admin page: user management, agent credentials and default models, source-control configuration, and tracker OAuth apps.

The first platform admin is bootstrapped via the CLI during setup; after that, admins grant or revoke the role in Admin → Users (changes apply at the user's next sign-in, and self-demotion is blocked).

Space settings

Space Settings (gear icon on the space page) is a tabbed page. Viewing is open to members; editing requires the Owner space role.

General

  • Space Name — rename the space.
  • Runtime — the space's execution profile. A badge shows the pinned workflow (default aidlc-v2). Settings:
  • Park release (seconds) — how long an agent session stays warm after parking on a human question before it is stopped (0–900, default 300). Lower values save compute; higher values make near-instant answers resume faster.
  • Max parallel units — concurrency cap for parallel construction lanes (0 = unbounded, limited only by the unit dependency graph).
  • PR strategy — Platform default, Intent PR, or PR per unit. Existing spaces remain on explicit Intent PR until changed; new spaces inherit the platform default.

Members

Add or remove members and assign their space access (Owner / Member). The picker includes enabled, confirmed Cognito accounts and enabled enterprise identities that pass their provider's access gate. An enterprise user appears after their first successful sign-in, when Cognito creates the broker identity. Entries are labeled Cognito account or with the enterprise provider name so separate identities that share an email address remain distinguishable.

Removal revokes REST access and new realtime tokens immediately; an existing realtime session expires within ten minutes. Delegated integration access is removed and must be configured again if the person is re-added.

Business roles

Each member can hold one or more business roles in a space: Product Owner / Product Manager, Business Analyst, Architect, UX/UI Designer, Developer, QA / Tester, Security / Compliance, Operations / DevOps, and Delivery Manager. Roles are assigned per space, so the same person can be Architect in one space and Developer in another.

Every member can see the role chips in the member list. Only Owners can change them: select a member's roles to open the editor, add or remove roles, and save. Business roles are separate from space access. Product Owner does not make someone a space Owner, and a business role never grants access to a space. Removing a member clears their roles; a re-added member starts with none.

Business roles currently describe responsibilities only. Gates and execution do not check them yet.

Agent

  • Space Agent Credentials — optional shared Bedrock and Kiro credentials for members without personal credentials. Personal credentials take precedence; platform credentials remain the fallback.
  • Recommended CLI — marks Kiro, Claude Code, OpenCode, or Codex as the space recommendation on the intent compose page. Each user still explicitly selects an available CLI for every new intent. CLIs a platform admin disabled in Admin → Agents → Agent CLIs cannot be chosen. A new space recommends Kiro, or the first enabled CLI when Kiro is disabled.
  • Model Override — pin a specific model per CLI for this space. When unset, the platform-wide default model from Admin → Agents → Default Models applies. Space overrides take precedence over the stage/agent-level model hints in the workflow.

Environment

  • Environment — select the published toolchain used by newly created intents. The panel shows the exact published revision, image digest, runtime compatibility version, and included tools.
  • Compatibility warnings — detected repository stacks are compared with the selected tools. Warnings are advisory and do not block assignment.
  • Standard fallback — spaces without an explicit assignment use the protected Standard Node.js/Python environment.

Choose Assign Environment before creating the intent that should use it. Changing the assignment does not move an existing intent to another runtime. See Managed tools and environments for publication, selection, and snapshot behavior.

Source Control

  • Repositories — add or remove the repositories that agents check out. Multi-repository spaces are supported; at intent creation you can pick a base branch per repository. For CodeCommit, repositories come from the shared platform connection; a space created before the connection existed (or before a connection change) keeps working on its original role and moves to the current connection the next time an owner runs Rebind and verify.
  • PR strategy — select whether the space inherits the platform delivery default or explicitly uses Intent PR / PR per unit.

Trackers

  • Tracker bindings — bind GitHub Issues, GitLab Issues, or one or more Jira Cloud projects so intents can be started from issues. See Binding a tracker to a space.

Settings snapshots

Runtime-relevant settings (CLI, models, environment revision, park release, max parallel units, and PR strategy) are snapshotted onto each intent when it is created. Changing a setting affects the next intent, never a run already in flight.

Deleting

  • Intents can be deleted by space owners when they are not running. Deletion cascades through the whole footprint: graph artifacts, process state, and collaboration threads.
  • Spaces support deep delete — the space and all of its intents and their data are removed. This cannot be undone.