Settings

Move a foreground `Bash` command to a background task on timeout instead of killing it; set to `false` to kill timed-out foreground commands instead

Default timeout (seconds) for background `Bash` tasks when the call omits `timeout`; `0` means no timeout. Explicit per-call `timeout` values are unaffected; print mode defaults to `0`

Whether to keep still-running background tasks when the session closes; in print mode only a fallback when `print_background_mode` is unset (`true` = `drain`)

Grace period in milliseconds after a task is asked to terminate; still-running tasks are force-stopped when it elapses

Maximum number of background tasks running concurrently

Print mode only: how pending background tasks are handled when the main agent's turn ends; `"exit"` exits immediately, `"drain"` waits for terminal states without feeding results back, `"steer"` injects completions as synthetic user messages steering new turns until none are pending

Maximum number of new turns triggered by background-task completions in `"steer"` mode; keeps the steering loop bounded

Wall-clock ceiling (seconds) for the print-mode wait/steer loop; no effect outside print mode or with `"exit"`

Whether the built-in skills that document Kimi Code itself are offered to the model

Default model alias; must be defined in `models`

Default permission mode for new sessions: `manual`, `yolo`, or `auto`. See the three permission modes

Whether new sessions start in Plan mode by default

Extra custom agent search directories, layered on top of the default directories

Extra skill search directories, layered on top of the default directories

Display name the agent calls itself in the system prompt (fills the `${product_name}` slot, including in your own `SYSTEM.md` and agent files)

Machine identifier in protocol fields (`User-Agent` product token, MCP client name); derived from `name` when omitted: lowercased, non-alphanumeric runs folded to `-`

Longest-edge ceiling in pixels; larger images scale down proportionally. Raising it preserves more detail at the cost of larger request bodies

Per-image byte budget for images the model reads for itself (`ReadMediaFile` default reads); `region` and `full_resolution` read-backs are exempt

Maximum total attempts for a failing step, including the initial attempt

Maximum steps per turn; unset or `0` means unlimited

Number of tokens reserved for model output; automatic compaction is triggered when the remaining context window falls below this value

Global default connection (startup + tool discovery) timeout in milliseconds for all MCP servers; a per-server `startupTimeoutMs` in `mcp.json` wins

Global default single tool-call timeout in milliseconds for all MCP servers; a per-server `toolTimeoutMs` in `mcp.json` wins

Whether to merge Agent Skills from all available directories

`anthropic` provider only; force adaptive thinking on or off, omit to infer from the model name (Claude ≥ 4.6 uses adaptive)

Per-model endpoint override (written by catalog imports); takes precedence over the provider's `base_url`, only effective together with `protocol`

Capability tags added explicitly: `thinking`, `always_thinking`, `image_in`, `video_in`, `audio_in`, `tool_use`; only ever added, never removed

Default thinking effort for the model; managed and open-platform refreshes may rewrite it. Pin via model overrides

Name shown in the UI; falls back to `model` when unset

Maximum context length in tokens; must be at least 1

Declared per-request input limit; compaction, context-overflow checks, and usage ratios prefer it, completion budgeting keeps the total window

Per-request output token cap (maps to `max_tokens`); currently only the `anthropic` provider reads it

Model identifier sent to the server when calling the API

Effort value sent on the wire to disable thinking (e.g. `none` for xai grok); the only way to actually stop reasoning on models that reason by default

Name of the provider to use; must be defined in `providers`

`openai` provider only; set when the gateway returns reasoning content under a non-standard field name (`reasoning_content` and friends are auto-detected)

Thinking effort levels the model accepts; unsupported values fall back to `default_effort`, out-of-list values fail; managed refreshes may rewrite it (pin via overrides)

Action on match: `allow` (permit immediately), `deny` (reject immediately), `ask` (prompt each time)

Match pattern in the form `ToolName` or `ToolName(arg-pattern)`, e.g. `Read` or `Bash(rm -rf*)`

Rule description for debugging and auditing

Rule scope: `turn-override`, `session-runtime`, `project`, `user`; defaults to `user`

API key, written in plain text in the config file

API base URL

Custom HTTP headers attached to each request

Fallback source for provider credentials; see the `env` sub-table

OAuth credential reference (`storage` and `key` fields); injected automatically by the login flow, so you normally never write this by hand

Provider type: `kimi`, `anthropic`, `openai`, `openai_responses`, `google-genai`, `vertexai`

API key

Service API URL

Custom HTTP headers attached to each request

OAuth credential reference, same structure as `providers.*.oauth`

Maximum wall-clock time (milliseconds) a single `Agent` subagent may run before it is settled as `timed_out`; `0` means no timeout

Maximum wall-clock time (milliseconds) a single `AgentSwarm` subagent may run; on timeout it is aborted and the aggregated report marks `Subagent timed out.`; `0` means no timeout

Whether anonymous telemetry is enabled; disabled only when explicitly set to `false`

Thinking effort: `low` / `medium` / `high` / `xhigh` / `max`; falls back to the model default when not in its supported list

Whether Thinking is enabled by default for new sessions; set to `false` to force Thinking off

Preserved Thinking passthrough: `kimi` sends it as `thinking.keep`, `anthropic` as a `clear_thinking_20251015` edit (routes to the beta Messages API). An off-value disables it; overridden by `KIMI_MODEL_THINKING_KEEP`; injected only while Thinking is on

`measured+estimated` combines measured usage with an estimate of the unmeasured tail; `measured` reports provider usage alone, updated when a request completes; `estimated` is a pure estimate, for providers that do not report usage

Global denylist, applied after `enabled`

Global allowlist: when non-empty, only the listed tools are available; omitting the field or setting an empty array imposes no constraint

On resume or when submitting after a long idle stretch, warn that the context cache may have expired and offer to compact or start a new session (v2 engine only)

Disable the occasional session rating prompt above the input box

Disable the non-bracketed paste-burst fallback that keeps rapid multi-line pastes from submitting line by line

External editor command for composing long input; empty falls back to `$VISUAL` / `$EDITOR`

Whether desktop notifications are sent

When to notify: `unfocused` (only when the terminal is not focused) or `always`

Render LaTeX math expressions in Markdown messages as Unicode text; `false` keeps the raw source

Custom status line command: its first stdout line replaces the footer, and a JSON snapshot is passed on stdin; capped at 300ms, throttled to once per second, failures fall back to the built-in layout

Built-in slots on the first footer line and their order: `mode`, `goal`, `model`, `tasks`, `cwd`, `git`, `tips`; unknown ids are skipped with a warning

Color theme: `auto`, `dark`, `light`, or the name of a custom theme

Whether new versions are installed automatically

Additional workspace directories (absolute paths); written automatically when you confirm "remember this directory" in `/add-dir`, and available in every session of this project