🐧 Pengy
New Chat
hi
Settings
🐧 Pengy
Settings
Settings
Base URL
OpenAI-compatible API endpoint — the default is a local Ollama server (no API key needed)
API Key
Current: set
Model
Fetch
Proactive context compaction
Selected-model policy
Off
Auto-discover
Explicit window
Selected-model window (tokens)
Threshold (all models)
Policy applies to the selected model. Off keeps reactive recovery and existing checkpoints. Auto needs endpoint metadata; an unknown window stays reactive. Typical windows: 131072, 262144, 1048576. Threshold is shared.
Theme
Follow system
Light
Dark
Shared with the desktop app via
theme_mode
Tool Confirmation
Confirm every tool call
Auto-approve read-only tools (Safe)
YOLO — approve everything automatically
LLM Timeout (seconds)
HTTP timeout for each LLM API request
Tool Timeout (seconds)
Max Tool Output (chars, 0=no limit)
Snipped (head+tail) when tool output exceeds this limit
Reasoning Effort
Provider default (do not send)
Off / none
Low
Medium
High
Extra high
Max
Optional best-effort parameter; unsupported proxies/providers may reject non-default values.
Preserve returned reasoning fields
Keeps reasoning_content/reasoning/reasoning_details in message history when providers return them.
Context Keep Turns
Elide tool results older than N turns (0 = keep all)
User Agent
System Message Template
You are a helpful assistant named Pengy. The current date is {date} and the user is {username} on host {hostname} which is {osinfo}.
Placeholders:
{date}
,
{username}
,
{hostname}
,
{osinfo}
Save Settings
Cancel
About Pengy