Business Configuration

FinClaw business configuration is done via .env and the YAML files under deploy/config/. This page is grouped by business-side items (what the customer's ops/admin changes); for ops-side items (what the deployment engineer/SRE changes) see Operations Configuration; for the first-deploy minimum see Basic Configuration.

LLM and Multi-Model

Supported Provider List

Native providers (built-in default base URLs):

provider key	Default base_url
openai	https://api.openai.com/v1
deepseek	https://api.deepseek.com/v1
anthropic / claude	https://api.anthropic.com
litellm	http://localhost:4000/v1
ollama	http://localhost:11434/v1
volcengine / doubao	https://ark.cn-beijing.volces.com/api/v3
kimi / moonshot	https://api.moonshot.cn/v1
qwen / dashscope / alibaba	https://dashscope.aliyuncs.com/compatible-mode/v1
minimax	https://api.minimax.io/v1
glm / zhipu / bigmodel	https://open.bigmodel.cn/api/paas/v4
openrouter	https://openrouter.ai/api/v1

OpenAI-compatible aliases: gemini, lmstudio (default http://localhost:1234/v1), custom (falls back to the OpenAI-compatible default); unknown providers fall back to the OpenAI-compatible protocol. mock is for testing only.

Multi-Model / Multi-Tenant

Model selection requires middleware, chatkit-web, and ai-infra-rs version alignment; the user model list API is GET /users/me/llm-models. In multi-tenant scenarios each tenant can configure models and quotas on the Admin -> Tenant -> Models page.

Channel Connectors

FinClaw provides three channel connectors:

Service	Port	Credential source
feishu-connector	26460	Admin -> Tenant -> Channels page / .env fallback
wecom-connector	26550	Same as above
dingtalk-connector	26462	Same as above

Credential Priority

1. Admin -> Tenant -> Channels page configuration (persisted to identity-service, recommended)
2. .env fallback (initialization or testing only)

Channel binding URLs are configured only via the pages; do not use FEISHU_BIND_* / WECOM_BIND_* / DINGTALK_BIND_* env variables.

env Fallback Variables

Variable	Service
FEISHU_APP_ID / FEISHU_APP_SECRET / FEISHU_DOMAIN / FEISHU_ACCOUNTS	feishu-connector
WECOM_BOT_ID / WECOM_SECRET / WECOM_CORP_ID / WECOM_WEBSOCKET_URL	wecom-connector
DINGTALK_*	dingtalk-connector

Warning: do not use FEISHU_BIND_* / WECOM_BIND_* / DINGTALK_BIND_* env variables. INTERNAL_SERVICE_TOKEN is only used by connectors to verify internal calls; it does not participate in channel authentication.

Channel External Network Addresses

Channel	External domains
Feishu	open.feishu.cn, open.larksuite.com, attachment CDN
WeCom	qyapi.weixin.qq.com, *.work.weixin.qq.com, Tencent CDN
DingTalk	oapi.dingtalk.com, *.dingtalk.com

Persona Templates

config/persona-templates/ provides multiple role templates:

• finclaw-lobster, research-partner, wiki-agent, writing-partner, etc.
• Each template contains definition files such as SOUL.md / IDENTITY.md / AGENT.md
• persona-templates.json is the index file

Templates can be bound to specific tenants on the Admin -> Tenant -> Personas page; new templates require restarting persona-service to load.

Skill Service

config/skill-service.yaml configures the skill service; config/mcp-servers.yaml.example is the MCP server configuration template.

Variable	Default	Description
AI_INFRA_RS_SKILLS_ENABLED	true	Whether to enable the skill runtime
AI_INFRA_RS_SKILLS_CUSTOM_DIRS	-	Custom skill directories (comma-separated)
AI_INFRA_RS_SKILLS_USE_STANDARD_LOCATIONS	true	Whether to use standard skill directories
AI_INFRA_RS_SKILLS_INCLUDE_WORKSPACE_SKILLS	true	Whether to include workspace skills

The MCP server configuration path is specified via AI_INFRA_RS_MCP_SERVERS_CONFIG_PATH; it reads config/mcp-servers.yaml by default.

Administrator and Session

Variable	Description
ADMIN_EMAILS	Administrator emails (comma-separated); determines administrator privileges; restart required after change
AUTH_MODE	Auth mode: mixed / local / oidc
IDP_PUBLIC_KEY	IdP public key (required in OIDC mode)
LOGINSERVER_URL	Login service URL
APP_URL	Application URL (e.g. http://localhost:26100)
OIDC_STATE_COOKIE_SECRET	OIDC state cookie secret
SESSION_COOKIE_NAME / SESSION_COOKIE_SECRET	Session cookie name / secret
SESSION_IDLE_TIMEOUT_DAYS	Session idle timeout (default 7)
SESSION_ABSOLUTE_TIMEOUT_DAYS	Session absolute timeout (default 30)
IFRAME_PARENT_ORIGINS	iframe parent origins (CSP frame-ancestors; required when embedding in an iframe)

Storage and Upload

Variable	Default	Description
MAX_FILE_SIZE	52428800 (50MiB)	Upload file size upper limit
WORKSPACE_FILES_BASE_PATH	-	Workspace files base path (typically /app/data/workspaces inside the container)
MEMORY_FILES_BASE_PATH	-	Memory files base path
AI_INFRA_RS_MEMORY_FILES_BASE_PATH	-	claw-side memory files base path
AI_INFRA_RS_MEMORY_SESSIONS_BASE_PATH	-	claw-side memory sessions base path
AI_INFRA_RS_MEMORY_INDEX_PATH	-	SQLite index path (vector retrieval)
AI_INFRA_RS_MEMORY_VECTOR_ENABLED	-	Whether to enable vector retrieval
AI_INFRA_RS_AUDIT_PATH	-	Audit log path

AG-UI

Variable	Default	Description
RUN_RESUME_ENABLED	true	Whether AG-UI resume is enabled (set 0 / false to disable)

Reference Documentation

• Operations Configuration
• Basic Configuration