thedotmack/claude-mem v15.2.0

v15.2.0 — Browser install funnel

on GitHub

Highlights

Install funnel in the browser (#5031)

Fresh installs now open a browser setup flow that finds your recent work and turns it into Work, Goals and Rules before setup finishes. Existing installations keep opening My Work as before.

1. Install — npx claude-mem install installs the detected hooks, starts the worker and opens the browser.
2. Scan — reads the last 30 days of supported local agent history. Configured excluded projects and agent-authored prompts are left out, stored prompts are deduplicated, and private-tag cleanup and configured redaction run before anything is shown or extracted.
3. Reveal — uploads the filtered owner prompts (gzip) to CMEM Cloud for hosted intent extraction. Anonymous installs get a private, persistent installation credential that retries and sign-in reuse. Returned goals, rules and request groups fill the local My Work snapshots. Upload and extraction errors are shown and can be retried. The pre-checked product-improvement box only changes improveProduct.
4. Storage choice — At Home or On The Go. On The Go uses the existing pairing, login and checkout flow; delivered memory credentials are activated before checkout is marked paid, and sync-only accounts keep their provider.
5. Finish Setup — installs any other selected local tools and can import existing memory. Extraction or import in progress is kept when a settings change restarts the worker.
6. What's next — search saved memory and prompts, keep a result as a note, then hand off to My Work. The terminal exits successfully.

Ongoing goals and progress refresh uses an installed, logged-in agent: Claude Code gets bounded file tools and memory recall only; Codex runs with a strict read-only workspace/runtime profile (no writes, no network), with native and npm runtime layouts supported; Gemini gets prepared evidence inline with no builtin tools, hooks or YOLO mode. Every saved answer passes the worker's validators.

Setup no longer fails when Cloud leaves conversations ungrouped (#5060)

On real 30-day scans the hosted model sometimes leaves conversations out of every group. The worker used to throw Hosted extraction omitted conversation … and fail setup after a successful hosted run. Now the groups Cloud returned are saved and conversations it did not place stay undecided: they show in My Work as their own request, and the local refresh can group them later. Unknown or repeated conversation ids still fail loudly.

Requirements

• Node ^20.19.0 || >=22.12.0.
• Uses CMEM Cloud hosted history import and intent extraction (claude-mem-pro #353, #355, #358).

Full Changelog: v15.1.0...v15.2.0