Repository Issue Activity (beta)

#2267 [BUG] Path traversal in `DELETE /download/{node_id}/{model_id}` allows recursive deletion outside the model directory

#2272 [BUG] KV prefix cache replays a previous image's features for token-identical vision requests (different images → identical answers)

#2271 Two Ubuntu nodes on same LAN fail to discover each other — no errors, network confirmed working

#2258 Is this repository even actively being watched and maintained in the last 4 months?

#2270 Tensor-parallel (MlxJaccl) output corrupts on prompts >~4k tokens — 2× M3 Ultra 512GB, EXO 1.0.71. Is there a newer build with the JACCL/prefill fixes?

#1847 jaccl RDMA crashes on Mac Studio M3 Ultra (errno=2, errno=60, errno=22) — recurring in production

#2262 [BUG] Memory leak in macOS App requiring periodic restarts

#2220 [Release] ETA for the next release with DeepSeek V4 Pro and GLM-5.2 support

#2236 [Request] Support Kimi K3 on Apple Silicon clusters with RDMA

#2248 Node IDs regenerate on every restart, causing transient placement/eviction churn on multi-node clusters

#2156 [BUG] On linux No module named mlx

#2259 [BUG] MacOS 27.0 The Exo is not visible in local network permission list of privacy settings in macOS

#2115 [BUG] When attaching an image, we get no answer and EXO hang

#2257 [BUG] Zenoh peer discovery succeeds (Hello exchange visible via tcpdump) but nodes never establish a session over USB tethering link (multi-homed hosts)

#957 [MEDIUM] Better placement algorithm for pipeline parallelism using memory bandwidth + latency

#2165 [BUG] Critical Feature Request: Fix Tool Calling Pipeline for Agentic Coding!!!

#1764 [BUG] Kimi-K2.5 fails on 4 node cluster

#1458 MLX GPU check_error SIGABRT crashes all nodes during inference (Kimi-K2.5 4-node RDMA)

#1881 [BUG] GLM-5-8bit-MXFP8 can not run rdma on 4 nodes

#1641 [BUG] KIMI-K2.5 on both Mac Studio 512 Go RAM - TENSOR + MLX RDMA FAILED

#1939 [CRITICAL] MiniMax-M2.7 causes kernel panic (watchdog timeout) on M3 Ultra 512 GB

#2108 Pipeline-parallel deadlock at 32+ concurrent requests — all_gather collective blocks indefinitely

#2173 Pipeline + JACCL RDMA deadlocks at Fence::wait (cv.wait) on 2× M3 Ultra + macOS 26.5 — fence PR#1 not sufficient

#2176 [BUG] Runner SIGSEGV on first inference after idle (4-node Tensor-Parallel RDMA, Kimi-K2.6); cleanup-path race dictionary changed size during iteration then escalates one runner death into a full-node EXO shutdown + cluster re-election

#2208 Long-context requests to 2-node Tensor/MlxJaccl instances hang or produce garbled output (mlx Fence::wait GPU-timeout race); candidate fix e9835615 introduces its own deadlock