I Ran PI-Desktop's Plan, Goal and Subagent Modes on a 1.7B Local Model. Here's What Broke.

Description and use

Issues running PI-Desktop with local Qwen3:1.7b: Plan mode creates invalid paths, Goal mode silently changes permissions, subagents function correctly.

Problem

Running PI-Desktop with a local 1.7B Qwen model on a plain Linux box exposed several failures in Plan, Goal and subagent modes, revealing model‑dependent issues and app‑level bugs such as silent permission changes and incorrect path handling.

Outcome

Plan mode generated a plan file but with non‑existent paths, leading to execution failures; Goal mode changed permission to Auto and did not create a goal file; subagent mode successfully delegated test execution. The 1.7B model’s limited context caused hallucinated paths and slow processing.

Workflow

• Install PI-Desktop AppImage and extract
• Configure local Ollama endpoint for qwen3:1.7b
• Run smoke test to confirm connectivity
• Open clsx project folder
• Set mode to Plan and request a test plan
• Observe plan file creation and execution log
• Set mode to Goal and observe silent permission change
• Run Goal mode and monitor for goal file creation
• Set mode to Agent and invoke test‑runner subagent
• Approve delegation and run npm test
• Review test results

Evidence

The plan was saved as a real file, .pi/plan/add-nested-array-falsy-values-tests-for-clsx-20261007-1108.md (315 bytes).

Source

The permission chip next to it had silently switched from "Ask every time" to "Auto".

Source

The app reported "Processed for 5m 19s", but wall-clock time from sending the prompt to seeing the card was about 17 minutes.

Source

The test‑runner subagent asks for approval before running npm test.

Source