Skip to main content
We support --dangerously-set-permissions (see Permissions at bottom of page)
Slate can be configured using a slate.json (or slate.jsonc) file placed in your project root or global config directory. 
{
  "$schema": "https://randomlabs.ai/config.json",
  "permission": {
    "*": "allow",
    "bash": "ask"
  }
}

​
File locations

Configuration files are merged together, not replaced. Settings from all sources are combined — later sources override only conflicting keys.
1

Global config

User-wide preferences at ~/.config/slate/slate.json. Applies to all projects.
2

Project config

slate.json in your project root (or any parent up to the git root). Safe to check into git.
3

Inline override

SLATE_CONFIG_CONTENT env var as a JSON string. Highest precedence.
Project config applies only to that project. Global config applies everywhere.

​
Options

General

FieldTypeDescription
privacybooleanDisables telemetry and logging when true.
Control which actions require your approval before Slate proceeds.
{
  "$schema": "https://randomlabs.ai/config.json",
  "permission": {
    "*": "allow",
    "bash": "ask",
    "read": {
      "*.env": "ask",
      "*": "allow"
    }
  }
}
Each permission key maps to an action ("allow", "ask", or "deny"), or a pattern object for fine-grained control.
PermissionDescription
*Wildcard — sets the default for all tools
readReading files
editWriting / editing files
bashRunning shell commands
grepSearching file contents
webfetchFetching web pages
websearchWeb searches
Set preferred default models for each slot.
{
  "$schema": "https://randomlabs.ai/config.json",
  "default_agent": "build",
  "models": {
    "main": { "default": "claude-sonnet-4-5" },
    "subagent": { "default": "claude-haiku-4-5" }
  }
}
FieldDescription
default_agentDefault agent. Must be a primary agent ("build" or "plan").
models.mainMain model for complex tasks.
models.subagentModel for subagent tasks.
models.searchModel for search tasks.
models.reasoningModel for reasoning tasks.
Connect Model Context Protocol servers for additional tools.
{
  "$schema": "https://randomlabs.ai/config.json",
  "mcp": {
    "my-tool": {
      "type": "local",
      "command": ["npx", "-y", "@my-org/mcp-server"],
      "environment": { "API_KEY": "{env:MY_API_KEY}" },
      "enabled": true
    }
  }
}
Define custom /commands for repetitive tasks.
{
  "$schema": "https://randomlabs.ai/config.json",
  "command": {
    "commit": {
      "description": "Write a commit message for staged changes",
      "template": "Write a concise git commit message for these changes: $ARGUMENTS",
      "agent": "build"
    },
    "review": {
      "description": "Review the code in a file",
      "template": "Review @$ARGUMENTS for correctness, clarity, and edge cases."
    }
  }
}
Invoke with /commit or /review in the TUI. $ARGUMENTS captures text typed after the command name.

​
Environment variables

VariableDescription
SLATE_API_KEYSkip the authentication dialog on launch
SLATE_CONFIGPath to a custom config file
SLATE_CONFIG_DIRPath to a custom config directory
SLATE_CONFIG_CONTENTInline JSON config (highest precedence)
SLATE_PERMISSIONJSON permission config override
SLATE_DISABLE_PROJECT_CONFIGDisable project-level slate.json discovery
SLATE_DANGEROUS_SKIP_PERMISSIONSSet to 1 to bypass all permission checks. Set automatically by --dangerously-set-permissions. Intended for CI/automation. ⚠️ Skips all safeguards including destructive operations.

​
CLI flags

Run slate --help to see all available flags.

Sessions

FlagAliasDescription
--continue-cResume the most recent session instead of starting a new one. Useful for picking up where you left off.
--resume <id>Resume a specific session by its session ID.
--recent-rOpen the session picker on launch so you can choose which past session to load.
FlagAliasDescription
<question>Pass a prompt directly as a positional argument. Slate runs the prompt and exits unless --prompt-interactive is also set.
--question <text>-qEquivalent to passing a positional prompt.
--prompt-interactive-iWhen combined with --question, sends the prompt but keeps the TUI open for follow-up rather than exiting.
FlagAliasDescription
--workspace <path>-wAdd a directory to Slate’s workspace. Can be specified multiple times to add several directories at once.
Headless mode lets you use Slate non-interactively — pipe in prompts, pipe out structured results — useful for CI or scripting.
FlagDescription
--stream-jsonRun Slate headlessly with JSONL (newline-delimited JSON) output. Compatible with the Anthropic Claude Code SDK. Each agent event is emitted as a JSON object on stdout.
--output-format <fmt>Set the output format. text prints plain text, stream-json is equivalent to --stream-json.
--input-format <fmt>Set how stdin is interpreted. stream-json treats stdin as JSONL messages; text treats it as a plain text prompt.
# Run a one-shot prompt and get JSONL output
slate --stream-json "Summarize the changes in the last commit"

# Pipe a prompt in as plain text
echo "What does this codebase do?" | slate --stream-json --input-format text
Server mode starts Slate as a standalone HTTP server without the TUI. The TUI (or another client) can then connect to it separately.
FlagDefaultDescription
--serverStart Slate as an HTTP API server instead of launching the TUI.
--port <number>7777Port to listen on in server mode.
--host <string>0.0.0.0Host to bind to in server mode.
--url <url>Connect the TUI to an already-running Slate server (e.g. http://localhost:7777). Useful for running the server and TUI on separate machines or in a container.
--directory <path>local cwdWhen using --url, sets the working directory on the remote server for the session.
# Terminal 1 — start the server
slate --server --port 7777

# Terminal 2 — connect the TUI to it
slate --url http://localhost:7777
FlagDefaultDescription
--dangerously-set-permissionsfalseBypass all permission checks including config plan rules. Sets SLATE_DANGEROUS_SKIP_PERMISSIONS=1 internally. Intended for CI pipelines and automation where interactive prompts are not possible. ⚠️ Auto-allows all tool calls including destructive shell commands and file writes.
# Run non-interactively with all permissions bypassed
slate --dangerously-set-permissions --stream-json "Run the test suite and fix any failures"