Repository navigation
Biorouter v1.94.1
This patch improves local Ollama setup and recovery, makes the Max turns setting clearer, and stops Copilot status checks from launching hostname processes on Windows.
Ollama setup and recovery
- Check the configured Ollama service, selected installed model, tool support and usable context before starting inference. Connection and setup failures include more useful guidance.
- Check a proposed Ollama address without replacing a working saved configuration. Local loopback connections bypass proxies, while remote connections retain proxy support.
- Check readiness before Meditation creates a session. Connection and server failures retain the previous successful progress marker, and failed attempts record the affected provider, model and session when one exists.
- Stop retrying again at the agent level after the provider has exhausted its retry budget.
- Preserve a selected context window during ordinary model discovery; explicit readiness checks can still refresh it.
Max turns
- Choose Unlimited or Set a limit with aligned, keyboard-accessible controls. The Turns field appears when a limit is selected and accepts a positive whole number.
- Remember the previous finite limit when switching modes. Invalid or blank entries do not overwrite a valid saved limit.
- Keep settings saves in order when switching panels or reopening settings, so an older response cannot replace a newer choice.
Windows
- Read the computer's hostname through the native operating-system API and reuse it while the app runs. Copilot's repeated status checks no longer launch
hostname.exe, removing that source of flashing console windows.
Validation
The fixes were checked separately and together, including native Rust tests on Windows, macOS and Linux, frontend regression tests, actual desktop interaction, controlled Ollama outages and recovery, and Windows child-process observation. Live Meditation and CLI acceptance passed with Qwen. Readiness checks do not guarantee that every model will produce valid tool calls or knowledge output.