Skip to main content
Chat with Mux is a built-in workspace where you can ask questions about Mux and manage your installation through conversation. Think of it as a help desk that can also make changes for you — it knows the docs, can update your settings, install skills, and show you usage analytics.

Opening Chat with Mux

MethodDetails
SidebarClick the help icon at the top of the project sidebar
Landing pageClick the Chat with Mux card
KeyboardPress F1

What you can do

  • Get help — ask questions about any Mux feature (workspaces, runtimes, agents, models, hooks, keybinds, etc.)
  • Edit settings — read and update ~/.mux/config.json and ~/.mux/providers.jsonc
  • Install skills — search, preview, and install community skills from skills.sh
  • Edit global instructions — update ~/.mux/AGENTS.md (your global agent instruction file)
  • Explore analytics — query your usage data (costs, tokens, latency) with charts and tables
  • Set model parameters — configure per-model defaults like temperature or max output tokens
Every write operation follows a confirm-before-write flow: the agent reads the current state, proposes changes, asks for your explicit approval, and only writes after you confirm.

Get help

Chat with Mux has access to a built-in snapshot of the Mux documentation, so answers are accurate to your installed version. Ask it anything about how Mux works — it will look up the relevant docs pages on demand rather than relying on training data alone.

Example prompts

  • “How do workspaces work?”
  • “What’s the difference between Plan and Exec mode?”
  • “How do I set up SSH runtimes?”
  • “What keyboard shortcuts are available?”
  • “Help me troubleshoot why my Anthropic key isn’t working”
If you’re using Auto mode in a project workspace, you don’t need to navigate to Chat with Mux manually. The Auto agent can route Mux-related questions there automatically.

Managing settings

Chat with Mux can read and write two configuration files:
FilePurpose
~/.mux/config.jsonApp settings (default model, update channel, runtimes, feature flags, etc.)
~/.mux/providers.jsoncProvider settings (API keys, base URLs, model parameters, etc.)
Secrets are always redacted. API keys, tokens, and passwords are replaced with [REDACTED] before the model sees them. The agent also refuses to write [REDACTED] back, so your secrets can’t be accidentally overwritten. Changes take effect immediately — no restart needed.

Example prompts

  • “Show me my provider config”
  • “Set my default model to claude-sonnet-4-5”
  • “Switch to the nightly update channel”
  • “Enable the Mux Gateway”
  • “Change my Anthropic base URL to a custom endpoint”
For the full list of provider settings and environment variables, see Providers. For model selection and thinking levels, see Models.

Model parameter overrides

Chat with Mux can set per-model defaults for parameters like temperature, max_output_tokens, and top_p. Try asking “set temperature to 0.7 for claude-sonnet-4-5” or “set max output tokens to 16384 for all Anthropic models”.
For the full list of supported parameters, resolution order, and configuration format, see Model Parameter Overrides.

Installing skills

Chat with Mux can search the skills.sh community catalog, preview skills, and install them — all through conversation.

Install flow

  1. Search“Find me a skill for Docker”
  2. Preview — the agent clones the skill repo, reads the SKILL.md, and shows you the full content
  3. Confirm — you review and approve the install
  4. Install — the skill is written to ~/.mux/skills/<name>/SKILL.md
Installed skills are available globally across all your workspaces.

Managing skills

Beyond installing, Chat with Mux can:
  • List your installed global skills
  • Update a skill by re-writing its SKILL.md
  • Delete a skill file or entire skill directory (with confirmation)
Built-in skills that ship with Mux cannot be deleted.
For details on skill discovery, the SKILL.md format, and authoring your own skills, see Agent Skills.

Editing global instructions

Chat with Mux can read and write ~/.mux/AGENTS.md — the global instruction file that applies to all agents across all your projects. The agent always reads the current file, shows you a diff of proposed changes, and asks for confirmation before writing.

Example prompts

  • “What does my AGENTS.md say?”
  • “Add a rule: always write tests for new functions”
  • “Help me write global instructions for code reviews”
For more on instruction files (global, project-level, and workspace-level), see Instruction Files.

Usage analytics

Chat with Mux can query your local usage data — costs, token counts, latency, model usage — and render the results as charts or tables.

Available data

TableContains
eventsPer-LLM-call telemetry: model, tokens (input/output/reasoning/cached), costs, latency (TTFT, streaming, tool execution), project, workspace
delegation_rollupsAggregated sub-agent cost summaries per workspace

Visualization

The agent can render results as bar, line, area, pie, stacked bar charts, or plain tables.

Example prompts

  • “How much have I spent this week?”
  • “Show me a chart of token usage by model over the last 30 days”
  • “Which model has the fastest time-to-first-token?”
  • “Break down my costs by project”

Security model

Chat with Mux is intentionally sandboxed:
  • No shell access — cannot run arbitrary commands
  • No file system access — cannot read or edit arbitrary files
  • No arbitrary web browsing — cannot fetch arbitrary URLs or browse the internet
  • No sub-agents — cannot spawn or delegate to other agents
  • Secrets redacted — API keys and tokens are never visible to the model
  • Confirm-before-write — every write requires explicit user approval
  • Schema validation — writes are validated against the config schema before being applied

Next steps