bug: Debian 13 and Flatpack - libnccl2 missing
Author: canonexCreated Sep 12, 2026Updated Sep 17, 2026
Version: v0.8.4
Describe the Bug
In Debian 13 Flatpack it does not find libraries
Steps to Reproduce
- Be in Debian, Nvidia and CUDA from Debian package
- Install Jan from Flathub
- Try to run it: you get:
Generation failed
Failed to create model: Failed to start model: "LlamacppError { code: LlamaCppProcessError, message: \"The model process encountered an unexpected error.\" }"- read the log you find that libnccl2 is missing and cannot be found from the flatpak:
[2026-09-11][22:15:19][tauri_plugin_llamacpp::router][INFO] [llamacpp-router] /home/user/.var/app/ai.jan.Jan/data/Jan/data/llamacpp/backends/b9967/linux-avx-cuda-cu12.0-x64/build/bin/llama-server: error while loading shared libraries: libnccl.so.2: cannot open shared object file: No such file or directory
[2026-09-11][22:15:19][tauri_plugin_llamacpp::router][ERROR] llama-server router exited before readiness: ExitStatus(unix_wait_status(32512))
[2026-09-11][22:15:19][webview:l@tauri://localhost/assets/index-qC5avcdD.js:53:5579][ERROR] Failed to start model Jan-v3.5-4B-Q4_K_XL for provider llamacpp: LlamacppError { code: LlamaCppProcessError, message: "The model process encountered an unexpected error." }
[2026-09-11][22:15:19][webview:l@tauri://localhost/assets/index-qC5avcdD.js:53:5579][ERROR] Failed to start llamacpp model: LlamacppError { code: LlamaCppProcessError, message: "The model process encountered an unexpected error." }
[2026-09-11][22:15:19][webview:l@tauri://localhost/assets/index-qC5avcdD.js:53:5579][ERROR] Failed to create model: Failed to start model: "LlamacppError { code: LlamaCppProcessError, message: \"The model process encountered an unexpected error.\" }"- if you analyze the dependencies:
$ flatpak run --command=sh ai.jan.Jan -c ' BACKEND="/home/user/.var/app/ai.jan.Jan/data/Jan/data/llamacpp/backends/b9967/linux-avx-cuda-cu12.0-x64/build/bin"
echo "=== NCCL ==="
ls -l "$BACKEND/libnccl.so.2"
echo
echo "=== dependencies ==="
ldd "$BACKEND/llama-server" 2>&1 | grep -E "not found|nccl|cuda|cudart"
echo
echo "=== version ==="
"$BACKEND/llama-server" --version
'
F: Not sharing "/usr/lib/x86_64-linux-gnu" with sandbox: Path "/usr" is reserved by Flatpak
=== NCCL ===
-rw-r--r-- 1 user user 241002200 12 set 00.38 /home/user/.var/app/ai.jan.Jan/data/Jan/data/llamacpp/backends/b9967/linux-avx-cuda-cu12.0-x64/build/bin/libnccl.so.2
=== dependencies ===
libllama-server-impl.so => /home/user/.var/app/ai.jan.Jan/data/Jan/data/llamacpp/backends/b9967/linux-avx-cuda-cu12.0-x64/build/bin/libllama-server-impl.so (0x00007f2a22e00000)
libllama-common.so.0 => not found
libmtmd.so.0 => not found
libllama.so.0 => not found
libggml.so.0 => not found
libggml-base.so.0 => not found
=== version ===
/home/user/.var/app/ai.jan.Jan/data/Jan/data/llamacpp/backends/b9967/linux-avx-cuda-cu12.0-x64/build/bin/llama-server: error while loading shared libraries: libllama-common.so.0: cannot open shared object file: No such file or directory- With the help of other LLM I got the following fix:
# 1. Install host NCCL
sudo apt install libnccl2
# 2. Find Jan's backend
find ~/.var/app/ai.jan.Jan/data/Jan/data/llamacpp/backends \
-path '*/linux-avx-cuda-*/build/bin/llama-server' -print
# 3. Set BACKEND to the containing build/bin directory
BACKEND="..."
# 4. Copy NCCL into it
cp -L /usr/lib/x86_64-linux-gnu/libnccl.so.2 "$BACKEND/"
# 5. Make Jan's bundled libraries look beside themselves
sudo apt install patchelf
for f in \
"$BACKEND/llama-server" \
"$BACKEND"/libllama-server-impl.so \
"$BACKEND"/libllama-common.so.* \
"$BACKEND"/libllama.so.* \
"$BACKEND"/libmtmd.so.* \
"$BACKEND"/libggml-cuda.so.* \
"$BACKEND"/libggml-cpu.so.* \
"$BACKEND"/libggml.so.*
do
patchelf --set-rpath '$ORIGIN' "$f"
done
# 6. Verify
ldd "$BACKEND/llama-server" | grep 'not found'
"$BACKEND/llama-server" --version
Screenshots / Logs
Operating System
- MacOS
- Windows
- Linux
Source: janhq/jan