Settings

Grouped in the Settings UI under four categories (Extensions > LlamaNexus).

Connection

Setting	Default	Description
`llamanexus.baseUrl`	`http://127.0.0.1:11434`	Base URL of the LlamaNexus proxy. Used by the status bar and commands outside a chat session; each session instead uses whichever discovered server you selected for it.
`llamanexus.apiPath`	`/openai/v1`	OpenAI-compatible path prefix.
`llamanexus.discoveryPort`	`50050`	UDP port to listen on for LlamaNexus server-discovery heartbeats; must match the `--discovery-port` a `serve` instance is broadcasting on.
`llamanexus.serverHeartbeatTimeoutSeconds`	`10`	Drop a discovered server from the list if no heartbeat arrives within this long (heartbeats are sent every 1s).

Model & Generation

Setting	Default	Description
`llamanexus.model`	`""`	Model name to request; empty uses the server's current/default model.
`llamanexus.temperature`	`0.2`	Sampling temperature.
`llamanexus.maxTokens`	`2048`	Max tokens per response. Raise this if the model gets cut off mid-response or mid-tool-call.
`llamanexus.systemPrompt`	(coding assistant prompt)	System prompt sent with every request.
`llamanexus.contextSize`	`4096`	Fallback context window size, used only for models LlamaNexus can't report a real ctx-size for.
`llamanexus.requestTimeoutSeconds`	`120`	Idle timeout once generation has started (aborts only if no further data arrives for this long).
`llamanexus.loadModelTimeoutSeconds`	`300`	Idle timeout used only while waiting for the very first token, for a cold model load.
`llamanexus.autoCompactThreshold`	`85`	Percentage of context usage at which the older part of the conversation is auto-summarized. `0` disables automatic compaction.
`llamanexus.autoCompactKeepMessages`	`6`	Number of most recent messages kept verbatim when compacting.

Tool Calling

Setting	Default	Description
`llamanexus.enableFileTools`	`true`	Allow `read_file`/`patch_file`/`write_file`/`list_files`. Writes/patches always ask for confirmation.
`llamanexus.enableBashTool`	`false`	Allow `run_bash` directly in your workspace root. Always asks for confirmation; output truncated, 60s kill timeout.
`llamanexus.enablePlanningTool`	`true`	Allow `update_plan` - never touches disk, works even in Plan Mode.
`llamanexus.enableAgentTool`	`false`	Allow `run_agent` to delegate to the session's configured agent server.
`llamanexus.maxAgentSteps`	`8`	Maximum back-and-forth turns a single `run_agent` dispatch can take.
`llamanexus.maxToolCallRepeats`	`3`	Stop a tool-calling chain if the model repeats the exact same call this many times in a row.
`llamanexus.maxToolChainDepth`	`20`	Maximum chained tool-call round-trips per message, as a backstop.

Chat Appearance

Setting	Default	Description
`llamanexus.timestampFormat`	`HH:mm`	Format for the timestamp next to each message's role title. Tokens: `YYYY`, `MM`, `DD`, `HH`, `mm`, `ss`, `hh`, `A`.
`llamanexus.chatFontSize`	`12`	Font size (px) for chat message content.
`llamanexus.chatCodeFontSize`	`12`	Font size (px) for fenced/inline code blocks and other monospace content.
`llamanexus.chatMetaFontSize`	`12`	Font size (px) for message metadata: role titles, timestamps, system/tool/agent notes, reasoning sections, plan panel.
`llamanexus.uiFontSize`	`12`	Font size (px) for the rest of the chat UI: composer, model/command menus, tuning sidebar, server list, dialogs.