v1.3.3

Added

• Drag a node in the Weave. Grab any knot and the graph follows: its ties
  stretch with it and the rest of the weave resettles when you let go.

v1.3.2

Changed

• The Weave reads more clearly. The same concept seen in two projects is
  now one knot, not two, so cross-project ties pull together instead of
  doubling. Decisions carry a label at rest; files, topics and sessions reveal
  theirs on hover, each on a soft pill so a dense weave stays legible.

Added

• A hover card on the Weave. Point at any node for its kind, confidence and
  how many ties it holds, without leaving the graph.

v1.3.1

Overview redesign, a neural Weave graph, a richer demo, performance fixes, and the em dash removed throughout.

v1.3.0

Confidence scoring, the causal Weave, a self-closing Weaver loop, and a two-week demo. See CHANGELOG.

v1.2.0

Added

• The Weave — the record drawn as a graph instead of a list. Entities and
  sessions are nodes; edges are the relationships already in the tables (a
  session that informs an entity, a note that contests another, a handoff
  lineage). A dependency-free force layout with the same visibility rules as
  recall; hover lights a neighbourhood, click opens a node.
• The Weaver scans for untested changes. Beyond acting on a ruling, it now
  finds fixes and features the record kept but the repo never tested —
  motif weaver scan lists them, motif weaver run <file> queues one. Queueing
  is always a person's choice; the agent is handed the session's own context and
  writes the focused test as a draft PR, never wandering the tree.
• The Review loop. Distilled memory notes now carry a human axis next to the
  machine one: a person can confirm, prefer (resolving a conflict),
  retire or dispute a claim — from motif memory review or the dashboard's
  new Review inbox, which counts what waits on a live badge. Rulings never
  delete and are themselves recorded with who, over what, and why. Recall
  serves the outcome: retired notes disappear, human-verified ones outrank
  machine-only ones, unresolved conflicts are shown with both sides and a
  warning.
• Staleness. A machine-made note whose source files were reworked by later
  sessions, with nothing refreshing the entity, is flagged "possibly stale" —
  computed deterministically when the queue is read, no model call involved.
  Human-verified notes are exempt.
• motif demo. Not a museum — a show: a team's week replays live in five
  acts. Sessions stream in, memory catches two of them contradicting each
  other, YOU rule on the conflict from the terminal, the Weaver aligns a real
  throwaway git repository with your ruling (diff on screen; rule the other way
  and it refuses to invent work), and recall answers with your verdict marked
  verified. No reader runs; nothing real can be touched.
• motif blame <file>. From the code back to the conversation: the
  sessions that touched a file, exact match beating loose and fresh beating
  old, drawn from the team server and this machine's own history. Attribution
  is inferred from what sessions recorded about themselves, and personal
  sessions stay out of a stranger's blame.

Fixed

• Entity names and notes distilled from personal sessions were listed for every
  token holder by the memory endpoints. Notes now inherit the visibility of
  their evidence, the same rule sessions follow.

v1.1.3

Changed

• Node 22 is the minimum. engines said 20, but the SQLite prebuilds now
  start at 22, so a Node 20 install compiled from source without saying so. Node
  20 reached end of life in April 2026.
• The README leads with what the product does rather than with prose: the
  cross-tool handoff is shown above the fold, install appears once instead of
  twice, the solo story sits next to the team story, the retrieval benchmark
  carries its real figures, and the command reference is collapsed so it no
  longer sits between the pitch and the setup.

v1.1.2

Fixed

• Installing on Node 24 compiled SQLite from source. The pinned
  better-sqlite3 line ships prebuilt binaries for Node 18, 20, 22 and 23 — but
  not for 24, the current LTS. Anyone on it fell back to a node-gyp build,
  which needs a full C++ toolchain and fails outright without one, so
  npx getmotif up could break on the most common Node version. Moved to the
  release line that carries prebuilds for Node 22 through 26.

Changed

• CI runs the suite and the real global-install check on Node 22 and 24, across
  Linux, macOS and Windows. Testing a single Node version is what let the gap
  above through.

v1.1.1

Fixed

• motif doctor asked the server who you were, so with the server off it
  reported "member identity (writes enabled)" as missing and told you to re-run
  connect — over a working token already on the machine. It now trusts the
  local member token when the server cannot be reached, and the server stays the
  authority when it answers.
• motif daemon start dropped --claude-dir, so a backgrounded daemon read the
  default directory instead of the one you named.
• A handoff out of a Codex session was annotated as coming from Claude Code. The
  marker now names the tool the session actually came from.

Changed

• motif recall says in the terminal that what it printed is context for an
  agent, not an answer to your question.
• The quickstart is now the real first run — npx getmotif up, on your own
  sessions, local until you connect to a team. The seeded two-member demo script
  it replaced is no longer part of the repository.

v1.1.0 — security release

Fixes an identity takeover through the registration route, a live event stream that leaked personal session titles and paths to any token holder, and an ask path that trusted the session record it was handed. Everyone running a shared server should upgrade. See CHANGELOG.md.

v1.0.6

Team visibility now applies to sessions already synced, teammate handoffs are no longer refused by the already-local guard, and the sender is told whether the delivery landed. New guide: docs/TEAM-SETUP.md.