GitHub identity — link your GitHub identity to MemHub so pull requests can be attributed to you, including merged PRs that have no linked coding session.
Memory reading — agents can open a fact, episode, artifact, or ingested document by ID with read_memory. For long artifacts, they can browse an outline and read individual sections instead of loading the whole document.
Rulebook — rules can block file edits and writes before they happen, with an explicit override reason when an exception is needed.
Rulebook — read rules can check a file before it enters the agent's context, including reads through shell commands. Available in newer Codex/Cursor plugin releases; the Claude marketplace update is still pending.
Sessions Insights — compare agent effort by PR type to see how work on fixes differs from work on features.
Brain overview — an Index lists the brain's artifacts and memory alongside its overview, with authors and dates. Agents can narrow searches by content group or author.
Managed billing — customers on managed contracts can still manage credits, auto-reload, payment cards, invoices, and usage themselves, while the account team handles contract terms.
Improvements15
Sessions — linked PRs appear across List, Lanes, and session details, with a popover for sessions that contributed to multiple PRs. A PR opened by the agent links automatically; use /memhub:link-pr to link a PR manually.
PR linking — the Linked PR column explains how linking works. Use /memhub:find-contributing-sessions when several sessions contributed to the same PR.
Sessions performance — changing filters avoids duplicate requests and background reads for hidden tabs. Narrower date windows can reuse a complete cached wider window, reducing the wait for results.
Rulebook — click the rules icon in the Sessions list to open that session's rule activity history.
Transcripts — rule markers show the turns where rules fired. Click a marker to open the rule's details, or use the transcript's jump control to move between rule events.
Brain overview — refreshed overviews link topics to the artifacts behind them. The Index keeps exact counts and a compact list of recent items, reducing the context an agent needs before opening a document.
Credits — enter a custom credit-purchase amount and preview the credits before continuing to checkout.
Auto top-up — review the trigger balance, refill target, and spending limit before enabling it. Auto top-up stays off until you save, with clearer validation and cost explanations.
Session outputs — organization admins can open deliverables produced by a teammate's session.
Rulebook — fire history supports grouping and ordering, and Summary shares the same period, engineer, repository, and rulebook filters.
Rulebook — statistics distinguish rules presented at session start from rules that matched an action, so exposure no longer inflates fire counts.
Rulebook — running sessions refresh their cached rules more frequently, and paused or superseded rules explain when cached copies may still apply.
Artifacts — HTML deliverables are saved as file bundles with searchable text. Long artifacts can be read in pages, and saving identical content in the same brain reuses the existing artifact.
Capture — markdown deliverables written through Bash can be captured as draft artifacts too. Available in newer Codex/Cursor plugin releases; the Claude marketplace update is still pending.
APIMemory billing — new memory usage is billed from measured model and embedding costs instead of fixed per-operation estimates. Historical usage retains its existing calculation.
Fixes11
Sessions storage — the session cache no longer fills the browser storage shared with other Studio settings. Cached sessions are stored once, with a size limit, and the old cache is cleared on load.
Rulebook — the activity count on a session matches the history opened from its rules icon, so the badge and ledger no longer show different totals.
Sign-in — a failed sign-in callback now shows an error and asks you to try again instead of failing silently.
Billing — subscriptions set not to renew show when access ends, rather than suggesting another invoice is coming.
Routines — goals referring to "me" retain the identity of the person who requested them when the routine runs later.
Documents — reading copied documents preserves page boundaries and avoids repeating text where adjacent chunks overlap.
Sessions — date windows include sessions active during the selected period, even if they started earlier, keeping List, Lanes, and Summary aligned.
Transcripts — authors can open their post-migration transcripts and see the tools used in those sessions again.
Rulebook — result rules inspect both the beginning and end of long outputs, so errors near the top are no longer missed.
Rulebook — empty filters, grouped counts, pagination, and "show all" actions behave consistently. A rule with no fires in the selected period no longer misleadingly says "never."
Capture — captured facts reconnect to their source episodes when processing finishes later.
Security3
GitHub identity — organization admins can revoke an identity link, and leaving the organization clears it so rejoining requires a fresh link.
Session outputs — admin access stays bounded to the selected session; consolidated memory drawing on the author's other sessions remains protected.
Billing — paid access is checked against verified payment and subscription state, and stale payment events cannot restore access or grant credits twice.