Issues

EmbeddinggemmaONNX crashes on <image_soft_token>: Gather index 262144 out of bounds

#2114

· belt opened

on Jul 30, 2026

`mempalace repair` segfaults (SIGSEGV / exit 139) on a large palace, leaving vector search permanently disabled

#2113

· jbhyphy opened

on Jul 30, 2026

sweep: a transcript with a repeated message uuid drops the entire file (duplicate IDs in upsert batch)

#2112

· neo-wanderer opened

on Jul 30, 2026

Many palaces, one machine: config-dir lever, caller-supplied vectors, source-adapter registry

#2110

· joshuafontany opened

on Jul 29, 2026

Single-writer discipline: bound how many daemons exist per palace, and how writers queue when they collide

#2108

· joshuafontany opened

on Jul 29, 2026

HNSW bloat guard (batch_size=2 / sync_threshold=2) corrupts the index it protects on chromadb 1.5.9

#2106

· joshuafontany opened

on Jul 29, 2026

docs/hook-write-routing.md documents unreleased behaviour in the present tense; it does not exist at v3.6.0 and configuring from it produces an inert (and later fail-closed) key

#2105

· HectorBernstorff opened

on Jul 29, 2026

sweep costs 12.4x more per drawer than mine --mode convos (377s vs 16s on the same transcript); a 2.3 MB transcript never finishes

#2104

· HectorBernstorff opened

on Jul 29, 2026

Stop-hook transcript ingest mines the whole project directory (511 files to capture 1), holding the single-writer lock and failing concurrent diary checkpoints

#2102

· HectorBernstorff opened

on Jul 29, 2026

MCP --read-only gates on the palace-write set, so mempalace_hook_settings still rewrites config.json

#2101

· mvalentsev opened

on Jul 29, 2026

Systematic code review — batch of fixes incoming (tracking)

#2100

· KeilerHirsch opened

on Jul 29, 2026

repair --mode from-sqlite silently ignores --dry-run and performs the real rebuild

#2095

· JLujan10 opened

on Jul 28, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

All issues

EmbeddinggemmaONNX crashes on <image_soft_token>: Gather index 262144 out of bounds

`mempalace repair` segfaults (SIGSEGV / exit 139) on a large palace, leaving vector search permanently disabled

sweep: a transcript with a repeated message uuid drops the entire file (duplicate IDs in upsert batch)

Many palaces, one machine: config-dir lever, caller-supplied vectors, source-adapter registry

Single-writer discipline: bound how many daemons exist per palace, and how writers queue when they collide

HNSW bloat guard (batch_size=2 / sync_threshold=2) corrupts the index it protects on chromadb 1.5.9

docs/hook-write-routing.md documents unreleased behaviour in the present tense; it does not exist at v3.6.0 and configuring from it produces an inert (and later fail-closed) key

sweep costs 12.4x more per drawer than mine --mode convos (377s vs 16s on the same transcript); a 2.3 MB transcript never finishes

Stop-hook transcript ingest mines the whole project directory (511 files to capture 1), holding the single-writer lock and failing concurrent diary checkpoints

MCP --read-only gates on the palace-write set, so mempalace_hook_settings still rewrites config.json

Systematic code review — batch of fixes incoming (tracking)

repair --mode from-sqlite silently ignores --dry-run and performs the real rebuild

Uh oh!

Search results