key	type	scope	default	sensitive	description
credit.deduction_policy	string	platform	{"tiers":[{"up_to_tokens":2000,"multiplier":0},{"multiplier":1}]}	false	Credit deduction policy (JSON)
credit.initial_grant	int	platform	1000	false	Initial credit grant for new organizations
credit.invite_reward	int	platform	500	false	Reward credits for the inviter
credit.invitee_reward	int	platform	200	false	Reward credits for the invitee
credit.per_usd	int	platform	1000	false	Credit exchange rate: credits per 1 USD
email.from	string	platform		false	SMTP sender address; leave empty to disable email sending
email.smtp_host	string	platform		false	SMTP Host
email.smtp_pass	string	platform		true	SMTP Password
email.smtp_port	int	platform	587	false	SMTP Port
email.smtp_tls_mode	string	platform	starttls	false	SMTP TLS mode: starttls/tls/none
email.smtp_user	string	platform		false	SMTP Username
feature.byok_enabled	bool	both	true	false	Whether to allow org-level credentials (BYOK)
feature.mcp_remote_enabled	bool	both	false	false	Whether to allow remote MCP
gateway.ip_mode	string	platform	direct	false	Gateway IP mode: direct/cloudflare/trusted_proxy
gateway.ratelimit_capacity	number	platform	600	false	Gateway Rate Limit Capacity
gateway.ratelimit_rate_per_minute	number	platform	300	false	Gateway Rate Limit Per Minute
gateway.risk_reject_threshold	int	platform	0	false	Gateway risk rejection threshold (0-100)
gateway.trusted_cidrs	string	platform		false	List of Gateway trusted proxy CIDRs
invite.default_max_uses	int	both	0	false	Default maximum uses for an invitation code; 0 means unlimited
invite.max_codes_per_user	int	both	1	false	Maximum invitation codes a single user can create
limit.agent_reasoning_iterations	int	both	0	false	Maximum reasoning-turn limit for Agent Loop; 0 means unlimited
limit.tool_continuation_budget	int	both	32	false	Maximum continuation budget for long-running tools
limit.concurrent_runs	int	both	0	false	Maximum concurrent run limit; 0 means unlimited
limit.max_input_content_bytes	int	both	32768	false	Maximum byte size for Run input content submission
limit.max_parallel_tasks	int	platform	32	false	Maximum limit for Lua parallel tasks/parallel tool calls
limit.team_members	int	both	0	false	Maximum team member limit; 0 means unlimited
llm.max_response_bytes	int	platform	16384	false	Maximum limit for reading LLM Provider HTTP responses (bytes)
llm.retry.base_delay_ms	int	platform	1000	false	Base delay for LLM retries (milliseconds)
llm.retry.max_attempts	int	platform	3	false	Maximum number of LLM retry attempts
openviking.base_url	string	platform		false	OpenViking Base URL
openviking.cost_per_commit	number	platform	0	false	OpenViking CommitSession Cost (USD)
openviking.root_api_key	string	platform		true	OpenViking Root API Key
quota.runs_per_month	int	both	0	false	Monthly run quota; 0 means unlimited
quota.tokens_per_month	int	both	0	false	Monthly token quota; 0 means unlimited
sandbox.agent_port	int	platform	8080	false	Sandbox Agent listening port
sandbox.base_url	string	platform		false	Sandbox Service address; Worker calls Sandbox via this URL; if empty, sandbox tools are not registered
sandbox.boot_timeout_s	int	platform	30	false	VM/container boot timeout (seconds)
sandbox.credit_base_fee	int	platform	1	false	Fixed credit deduction per sandbox call to cover cold start/scheduling overhead
sandbox.credit_rate_per_second	number	platform	0.5	false	Credit rate per second of sandbox execution duration
sandbox.allow_egress	bool	platform	true	false	Whether Sandbox backends may access the public network
sandbox.docker_image	string	platform	arkloop/sandbox-agent:latest	false	sandbox-agent image used by the Docker backend
sandbox.idle_timeout_lite_s	int	platform	180	false	Idle timeout for Sandbox lite tier (seconds)
sandbox.idle_timeout_pro_s	int	platform	300	false	Idle timeout for Sandbox pro tier (seconds)
sandbox.idle_timeout_ultra_s	int	platform	600	false	Idle timeout for Sandbox ultra tier (seconds)
sandbox.max_lifetime_s	int	platform	1800	false	Maximum lifetime of a Sandbox session (seconds)
sandbox.max_sessions	int	platform	50	false	Maximum concurrent Sandbox sessions
sandbox.provider	string	platform	firecracker	false	Sandbox backend type: firecracker / docker
sandbox.refill_concurrency	int	platform	2	false	Maximum concurrency for pre-warm refill
sandbox.refill_interval_s	int	platform	5	false	Pre-warm refill check interval (seconds)
sandbox.warm_lite	int	platform	3	false	Number of pre-warmed instances for lite tier
sandbox.warm_pro	int	platform	2	false	Number of pre-warmed instances for pro tier
sandbox.warm_ultra	int	platform	1	false	Number of pre-warmed instances for ultra tier
turnstile.allowed_host	string	platform		false	Turnstile Allowed Host
turnstile.secret_key	string	platform		true	Turnstile Secret Key
turnstile.site_key	string	platform		false	Turnstile Site Key
web_fetch.firecrawl_api_key	string	both		true	Firecrawl API Key
web_fetch.firecrawl_base_url	string	both		false	Firecrawl Base URL
web_fetch.jina_api_key	string	both		true	Jina API Key
web_fetch.provider	string	both	basic	false	Web Fetch Provider: basic/firecrawl/jina
web_search.provider	string	both		false	Web Search Provider: searxng/tavily
web_search.searxng_base_url	string	both		false	SearXNG Base URL
web_search.tavily_api_key	string	both		true	Tavily API Key

Configuration Reference