You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Server self-registration: a running server now writes its own runtime.json and
global registry.json entry (learned from the bound socket) regardless of how it
was launched, and re-asserts every 180s — fixing servers (orphaned, raw-uvicorn,
IDE-launched) being invisible to the dashboard.
Server-side self-heal heartbeat: restarts a dead file watcher or job worker,
rebuilds a corrupt vector index (crash-safe file-only check), and relaunches the
web dashboard if it is down (locked, single-launch).
Dashboard config save now guards against data-incompatible changes: editing the
embedding provider/model, storage backend, or graph store type while data is
indexed is blocked with an explanation and a "Save & reindex now" action
(server exposes a read-only GET /index/fingerprint).
Per-field help across Storage, GraphRAG, Git Indexing, Reranker, and Session
Extraction, plus per-section descriptions distinguishing Session Extraction
(curated summary) from Session Indexing (raw transcript archive/embedding).
Changed
Single locked writer for registry.json (server registry.py); CLI update_registry now delegates to it (no more lockless writes).
One build_server_command builds the uvicorn argv for all CLI/MCP launch paths;
foreground + MCP no longer hand-copy the spawn block.
One render_dashboard_url renders the hot_pink dashboard box for init/start/dashboard.
Fixed
brainpalace config wizard now defaults graphrag store_type to sqlite
(persistent, temporal) instead of the ephemeral simple, matching the server
default and the dashboard.