2. Adopt Your Existing Agent Setup¶
Your Agent setup already contains useful decisions. This step creates the Aikito workspace, discovers the instructions, MCP servers, and subagents you already use, and imports them into one canonical source. It assumes the CLI is installed.
Ask your agent¶
Initialize my Aikito workspace at ~/aikito, or inspect it if it already exists. Inspect the Agent configuration I already use and run
aikito adoptto bring supported resources into the workspace. Adoption preflights everything before writing and leaves the original Agent configuration unchanged. If it stops, explain each finding and ask before skipping a resource or resolving a conflict. Do not synchronize or register a project yet.
Create and connect manually¶
The explicit path becomes the remembered workspace location. Initialization creates the canonical workspace and detects installed supported agents. If you already have a workspace, use its path instead. For a cloned workspace on another machine, follow Connect a workspace on another machine.
Adopt the configuration those Agents already use:
Adoption validates every detected resource before creating a backup or writing.
It imports supported instructions, MCP servers, and subagents into the workspace
but leaves the original Agent configuration unchanged. If sources disagree or a
resource is invalid, the entire operation stops and reports the source, reason,
and available action. Repair resources you want to keep; use the displayed
resource-level --skip only when you intentionally do not want to adopt one.
Agent-native default servers configured in agents/<name>.toml under builtin_mcps
(such as openaiDeveloperDocs for Codex) are automatically excluded from adoption.
Sources follow registered Agent paths and supported import adapters. Bundled
definitions also allow discovery of unregistered built-in Agents. When a new
MCP or subagent resource requires one, adoption registers its bundled
agents/<name>.toml first. Existing definitions are preserved; importing only
instructions does not register an Agent.
Existing skill directories are not imported automatically. After this onboarding
flow, review them separately, use aikito add skill to create a canonical
destination, and move in the content you want Aikito to govern.
| Workspace path | What it holds |
|---|---|
global/AGENTS.md |
Instructions shared across projects |
agents/<name>.toml |
Agent integration paths and capabilities |
skills/ and skills.toml |
Reusable skills and global selections |
subagents/<name>.md |
Subagent personas configured across agents |
mcps/ |
Model Context Protocol configurations |
projects/ |
Each registered project's configuration and memory |
Upgrade an existing workspace¶
Workspaces created before the per-resource layout require an explicit migration. Installing the new Aikito version does not edit the workspace. Normal commands show the migration command until the layout is upgraded:
For a workspace shared through Git, upgrade Aikito to version 1.53.0 or newer
on every machine before migrating it. After every machine is upgraded,
preview and migrate on one machine, commit and push the workspace changes,
then pull them on the others. An older CLI such as 1.52.1 cannot use the
migrated workspace: its commands report Agents config not found, and init
cannot restore the old layout.
Review any collisions or unsupported entries reported by the preview before
applying. The migration writes agents/<name>.toml and frontmatter in
subagents/<name>.md, removes the two legacy registry files, and records
version = 2 in layout.toml after the resource changes succeed. Repeating
the command on a migrated workspace is safe.
Inspect what became canonical:
If nothing is detected, adoption completes as a no-op. Continue with an empty workspace and create resources later. See the adoption safety model for source, backup, credential, and write boundaries.
Verify¶
aikito path workspace should print your selected location. The show commands
should display the adopted canonical resources without changing the original
Agent-native files.
Optionally run aikito web to browse resources and status in the local
Web Console.