All notable changes to Alter Ego are listed here. The format follows Keep a Changelog and the project uses Semantic Versioning.
[1.3.0] - 2026-11-16
Changed
- Directory review fixes.
open_context,list_filesandread_fileno longer write to the journal, so their
read only annotation is true. The turn argument is gone from them. start takes an optional request.
- The server instructions and the text
startandopen_contextreturn describe what the tools do instead of
telling the client how to behave. The user's own rules are returned unchanged.
[1.2.0] - 2026-11-13
Added
- New users get a starter rule and a
token-economyplaybook for clients that can start subagents: the main
chat plans and answers, execution goes to background agents on Sonnet (big simple work to Haiku, never the largest model), each with a complete block of work. Clients without subagents ignore it. Existing users' data is not changed.
[1.1.0] - 2026-11-13
Added
- New users get two more starter playbooks,
when-it-failsandtimesheet-proposal, and starter rules
about finishing the job, looking for another route before saying "cannot", staying within the request, checking the real result and never using em or en dashes.
CHANGELOG.md,RELEASING.mdandserver.jsonfor the MCP Registry.
Changed
- The server version comes from
pyproject.toml(one source of truth) and is reported in the MCP
initialize response.
- Errors in the AI chat now explain what went wrong and how to fix it, and opening
/mcpin a browser
shows a friendly page instead of a protocol error.
Fixed
- Two refreshes with the same refresh token within 60 seconds return the same new token pair instead of
invalid_grant, so Claude no longer drops the connection.
[1.0.0] - 2026-11-12
First public release, published in the MCP Registry as com.cafeboon/alter-ego.
Added
- A remote MCP server that gives AI chats a memory in the user's own storage: rules, profile, contexts with
journals and playbooks as plain Markdown files. Tools: start, open_context, new_context, log, remember, list_files, read_file, write_file.
- Storage in OneDrive, Google Drive, Dropbox, GitHub or Azure DevOps, one address for all of them with a page
to pick the storage before signing in. The server keeps no user content.
- The public website with docs, privacy policy and terms.
- Every tool call is logged with its outcome, and every chat writes a journal line at the start.
- Core agent-context files and journals are protected from
write_file, parent paths are rejected. - Funnel log written to a file, error pages with a way back, sign-in flows of an hour that survive a
restart, GitHub and Azure DevOps setup inside the flow and a storage picker without a preselection.
Changed
- Journal lines are kept when chats write at the same time and the opened storage is reused.
- The
startdescription no longer tells the model when to call it, for the Connectors Directory policy.