Environment variables

claude-multi runtime variables

These control claude-multi itself. Set them in your shell before launching.

Variable	Default	Description
CLAUDE_MULTI_INK	true	Set to false to use the simpler prompts-based UI instead of the Ink TUI
CLAUDE_MULTI_HOME	~	Override the base directory for config storage. claude-multi looks for config.json at $CLAUDE_MULTI_HOME/.claude-multi/
CLAUDE_MULTI_UPDATE_CHECK	false	Set to true to check for claude-multi and Claude Code updates on launch

Example:

# Use the simpler UI
CLAUDE_MULTI_INK=false claude-multi

# Store config in a custom location
CLAUDE_MULTI_HOME=/tmp/test-env claude-multi add test --provider deepseek --api-key sk-...

---

Claude Code instance variables

These live in each instance’s settings.json under the env key.

Core API variables

Variable	Description	Example
ANTHROPIC_AUTH_TOKEN	API key for the provider	sk-your-key-here
ANTHROPIC_BASE_URL	Provider’s Anthropic-compatible API endpoint	https://api.deepseek.com/anthropic

Model mapping variables

Variable	Description	Maps to
ANTHROPIC_MODEL	Primary model	Claude Code’s internal opus slot
ANTHROPIC_SMALL_FAST_MODEL	Fast/cheap model for quick tasks	Claude Code’s internal haiku slot
ANTHROPIC_DEFAULT_SONNET_MODEL	Sonnet-tier model	Used when Claude Code requests a “sonnet” class model
ANTHROPIC_DEFAULT_OPUS_MODEL	Opus-tier model	Used when Claude Code requests an “opus” class model
ANTHROPIC_DEFAULT_HAIKU_MODEL	Haiku-tier model	Used when Claude Code requests a “haiku” class model

Sub-agent variables

Variable	Description	Example
CLAUDE_CODE_SUBAGENT_MODEL	Model used by sub-agents for background tasks, exploration, and code review	deepseek-v4-flash
CLAUDE_CODE_EFFORT_LEVEL	Reasoning effort level for the main model	low, medium, high, max

Timeout

Variable	Description	Example
API_TIMEOUT_MS	Request timeout in milliseconds	600000

Context window / auto-compaction variables

Claude Code compresses older conversation history once usage crosses a share of the context window, instead of failing when the model hits its token limit. It assumes a 200K window by default, so a provider with a smaller real window needs these overrides. Without them, compaction runs too late and the API call fails mid-session.

Variable	Description	Example
CLAUDE_CODE_AUTO_COMPACT_WINDOW	The model’s actual context window size, in tokens	131072 (Qwen’s 128K)
CLAUDE_AUTOCOMPACT_PCT_OVERRIDE	Percentage of the window that triggers compaction	75

Models that match or exceed the 200K default, like GLM-5.3 with its 1M window, use a [1m] suffix on the model name instead, so you do not need an override.

---

Per-provider defaults

Each provider template sets these variables to its own values.

GLM (glm)

{
  "ANTHROPIC_AUTH_TOKEN": "",
  "ANTHROPIC_BASE_URL": "https://api.z.ai/api/anthropic",
  "API_TIMEOUT_MS": "3000000",
  "ANTHROPIC_DEFAULT_HAIKU_MODEL": "glm-5-turbo",
  "ANTHROPIC_DEFAULT_SONNET_MODEL": "glm-5.3-flash[1m]",
  "ANTHROPIC_DEFAULT_OPUS_MODEL": "glm-5.3[1m]",
  "ANTHROPIC_MODEL": "glm-5.3[1m]",
  "ANTHROPIC_SMALL_FAST_MODEL": "glm-5-turbo",
  "ENABLE_THINKING": "true",
  "REASONING_EFFORT": "high",
  "MAX_THINKING_TOKENS": "8000",
  "ENABLE_STREAMING": "true",
  "MAX_OUTPUT_TOKENS": "128000"
}

The glm template maps opus to glm-5.3[1m], sonnet to glm-5.3-flash[1m], and haiku/small-fast to glm-5-turbo (200K). Both glm-5.3[1m] and the flash variant use the [1m] suffix for their 1M context window; GLM-5.3-Flash is natively multimodal with 3x the GLM-5.3 quota on the Coding Plan. Requests for glm-5.2 or glm-5.1 are automatically routed to glm-5.3 on the Coding Plan endpoint, so existing configs keep working.

MiniMax (minimax)

{
  "ANTHROPIC_BASE_URL": "https://api.minimax.io/anthropic",
  "ANTHROPIC_MODEL": "MiniMax-M3",
  "ANTHROPIC_SMALL_FAST_MODEL": "MiniMax-M3",
  "ANTHROPIC_DEFAULT_SONNET_MODEL": "MiniMax-M3",
  "ANTHROPIC_DEFAULT_OPUS_MODEL": "MiniMax-M3",
  "ANTHROPIC_DEFAULT_HAIKU_MODEL": "MiniMax-M3",
  "CLAUDE_CODE_SUBAGENT_MODEL": "MiniMax-M3",
  "CLAUDE_CODE_EFFORT_LEVEL": "max"
}

DeepSeek (deepseek)

{
  "ANTHROPIC_BASE_URL": "https://api.deepseek.com/anthropic",
  "ANTHROPIC_MODEL": "deepseek-v4-pro[1m]",
  "ANTHROPIC_SMALL_FAST_MODEL": "deepseek-v4-flash",
  "ANTHROPIC_DEFAULT_SONNET_MODEL": "deepseek-v4-flash",
  "ANTHROPIC_DEFAULT_OPUS_MODEL": "deepseek-v4-pro[1m]",
  "ANTHROPIC_DEFAULT_HAIKU_MODEL": "deepseek-v4-flash",
  "CLAUDE_CODE_SUBAGENT_MODEL": "deepseek-v4-flash",
  "CLAUDE_CODE_EFFORT_LEVEL": "max",
  "API_TIMEOUT_MS": "600000"
}

Xiaomi MiMo (mimo)

{
  "ANTHROPIC_BASE_URL": "https://api.xiaomimimo.com/anthropic",
  "ANTHROPIC_MODEL": "mimo-v2.5-pro",
  "ANTHROPIC_SMALL_FAST_MODEL": "mimo-v2.5",
  "ANTHROPIC_DEFAULT_SONNET_MODEL": "mimo-v2.5",
  "ANTHROPIC_DEFAULT_OPUS_MODEL": "mimo-v2.5-pro",
  "ANTHROPIC_DEFAULT_HAIKU_MODEL": "mimo-v2.5",
  "API_TIMEOUT_MS": "600000"
}

The mimo-token template uses the same models but a different base URL (token-plan-cn.xiaomimimo.com). Replace it with your regional endpoint from the subscription console.

Moonshot Kimi (kimi)

{
  "ANTHROPIC_BASE_URL": "https://api.moonshot.ai/anthropic",
  "ANTHROPIC_MODEL": "kimi-k2.6",
  "ANTHROPIC_SMALL_FAST_MODEL": "kimi-k2.5",
  "ANTHROPIC_DEFAULT_SONNET_MODEL": "kimi-k2.6",
  "ANTHROPIC_DEFAULT_OPUS_MODEL": "kimi-k2.7-code",
  "ANTHROPIC_DEFAULT_HAIKU_MODEL": "kimi-k2.5",
  "API_TIMEOUT_MS": "600000"
}

Alibaba Qwen (qwen)

{
  "ANTHROPIC_BASE_URL": "https://dashscope-intl.aliyuncs.com/apps/anthropic",
  "ANTHROPIC_MODEL": "qwen3-coder-next",
  "ANTHROPIC_SMALL_FAST_MODEL": "qwen3-coder-flash",
  "ANTHROPIC_DEFAULT_SONNET_MODEL": "qwen3-coder-plus",
  "ANTHROPIC_DEFAULT_OPUS_MODEL": "qwen3-coder-next",
  "ANTHROPIC_DEFAULT_HAIKU_MODEL": "qwen3-coder-flash",
  "API_TIMEOUT_MS": "600000"
}

The qwen-coding template uses the same models but a different base URL (coding-intl.dashscope.aliyuncs.com).

---

Where to set variables

claude-multi variables: Set in your shell profile (.zshrc, .bashrc, .profile) or export them before running claude-multi.

Instance variables: Edit ~/.claude-<name>/settings.json directly, or let the provider template fill them in when you create the instance. They apply to that instance only and do not affect other instances or your default ~/.claude setup.