Local Coder
Qwen · local projects, persistent sessions and tools
1 · Memory and services
Jetson shares CPU and GPU memory. Process RSS does not account for all CUDA allocations. Available RAM is the start gate.
Last sample: — · stale| Select | Service / core | State | RSS / cgroup | GPU | Dependencies |
|---|
Only listed application services can be selected. Pausing also affects their dependent services; inspect the preview before confirming.
Cache release runs sync and drops reclaimable filesystem caches. It cannot unload active CUDA allocations: stop the owning process first.
2 · Model and agent
Model installation is not complete.
Stopping interrupts running coding work. Service restoration is a separate explicit operation.
3 · Configuration and project
Context: — token · llama.cpp · OpenCode
Imported Windows projects are local copies on Thor. Changes stay in that workspace until an explicit export or sync. Project instructions and repository files are read from the selected local directory.
This environment uses the local Qwen provider. External publication requires connectivity; local builds require dependencies already present. You can stop Qwen and continue from OpenAI or Anthropic using the same exported code.
4 · Coding session
Analysis · lifecycle receipt
The monitor unloads only the coding agent when the reserve is insufficient or memory telemetry is unavailable.
No suspended service receipt.