Configuration¶

All configuration is via environment variables, loaded from e2e/.env (or your shell). No hardcoded secrets anywhere in the codebase.

Unified Config System (v0.3.0)¶

Starting with v0.3.0, configuration is unified across a config file, environment variables, and CLI flags — resolved through a deterministic priority chain.

Config File Format¶

Quark Agent reads .quark-agent.json from the project root (or the path specified by --config):

json { "llm": { "provider": "openai", "model": "gpt-4o-mini", "apiKey": "", "baseUrl": "https://api.openai.com/v1" }, "sandbox": { "timeoutMs": 10000, "maxMemoryMb": 128, "policy": "balanced" }, "moa": { "enabled": true, "defaultTier": "medium" }, "checkpoint": { "enabled": true, "autoSaveIntervalMs": 30000, "dir": ".checkpoints" }, "channels": { "http": { "port": 3456 }, "feishu": { "appId": "", "appSecret": "", "verificationToken": "" } }, "logging": { "level": "info" } }

Generate a starter config interactively:

bash quark-agent setup

Environment Variables¶

Variable	Description	Default
LLM Providers
`MICRO_LLM_PROVIDER`	LLM provider (`openai` \| `anthropic` \| `gemini` \| `ark`)	`openai`
`MICRO_LLM_MODEL`	Model name	`gpt-4o-mini`
`QUARK_API_KEY`	Primary API key (any provider)	—
`ARK_API_KEY`	Volcengine Ark API key	—
`ARK_BASE_URL`	Ark base URL	`https://ark.cn-beijing.volces.com/api/coding/v3`
`ARK_MODEL`	Ark model name	`doubao-seed-code`
`MODEL_PROXY_URL`	Anthropic proxy URL	—
`MODEL_PROXY_KEY`	Anthropic proxy key	—
`OPENAI_API_KEY`	OpenAI-compatible API key	—
`OPENAI_BASE_URL`	OpenAI-compatible base URL	`https://api.openai.com/v1`
`OPENAI_MODEL`	OpenAI-compatible model name	`gpt-4o-mini`
`ANTHROPIC_API_KEY`	Anthropic API key (direct)	—
`ANTHROPIC_MODEL`	Anthropic model name	`claude-sonnet-4-20250514`
`GEMINI_API_KEY`	Google Gemini API key	—
`GEMINI_MODEL`	Gemini model name	`gemini-2.0-flash`
Channels
`PORT`	HTTP channel port	`3456`
`FEISHU_APP_ID`	Feishu / Lark app ID	—
`FEISHU_APP_SECRET`	Feishu / Lark app secret	—
`FEISHU_VERIFICATION_TOKEN`	Feishu verification token	—
`WECOM_CORP_ID`	WeCom corp ID	—
`WECOM_AGENT_ID`	WeCom agent ID	—
`WECOM_SECRET`	WeCom secret	—
`TELEGRAM_BOT_TOKEN`	Telegram bot token	—
`DISCORD_BOT_TOKEN`	Discord bot token	—
`DISCORD_GUILD_ID`	Discord guild (server) ID	—
`SLACK_BOT_TOKEN`	Slack bot OAuth token (`xoxb-...`)	—
`SLACK_APP_TOKEN`	Slack app-level token (`xapp-...`)	—
`GITHUB_WEBHOOK_SECRET`	GitHub webhook secret	—
`GITHUB_TOKEN`	GitHub API token	—
Image Generation
`IMAGE_API_URL`	Image generation API URL	—
`IMAGE_API_KEY`	Image generation API key	—
`IMAGE_MODEL`	Image model name	`flux-1.1-pro`
Memory
`MEMORY_DB_PATH`	SQLite database path	`./.data/memory.sqlite`
`MEMORY_COMPRESS_THRESHOLD`	Compression trigger threshold	`50`
Sandbox
`SANDBOX_TIMEOUT_MS`	Code execution timeout	`10000`
`SANDBOX_MAX_MEMORY_MB`	Max memory for sandbox	`128`
`SANDBOX_POLICY`	Security policy preset	`balanced`
`SANDBOX_AUTO_APPROVE`	Tool auto-approve mode (`all` \| `safe` \| `none`)	`safe`
Checkpoint
`CHECKPOINT_DIR`	Checkpoint storage directory	`.checkpoints`
`CHECKPOINT_AUTO_SAVE_MS`	Auto-save interval in ms	`30000`
MoA
`MOA_ENABLED`	Enable Mixture of Agents routing	`false`
`MOA_DEFAULT_TIER`	Default model tier	`medium`
Logging
`LOG_LEVEL`	Log level (`debug` \| `info` \| `warn` \| `error`)	`info`

Priority Chain¶

Configuration values are resolved in this order (later wins):

Defaults < Config file (.quark-agent.json) < Environment variables < CLI flags

Example — the model is resolved as:

Default: gpt-4o-mini
Config file: .quark-agent.json → llm.model: "gpt-4o"
Env var: OPENAI_MODEL=gpt-4o-mini
CLI flag: --model gpt-4

Final value: gpt-4 (CLI flag wins).

Setup Wizard¶

Run the interactive setup wizard to generate a config file and validate API keys:

bash quark-agent setup

The wizard will:

Prompt for LLM provider and API key
Validate the key with a test request
Ask which channels to enable
Set sandbox policy and auto-approve mode
Optionally enable MoA and checkpointing
Write .quark-agent.json to the current directory

Non-interactive mode (CI-friendly):

bash quark-agent setup --provider openai --api-key $OPENAI_API_KEY --policy balanced --yes

LLM Providers¶

Three priorities, tried in order:

Priority 1 — Volcengine Ark (OpenAI-compatible)¶

env ARK_API_KEY= ARK_BASE_URL=https://ark.cn-beijing.volces.com/api/coding/v3 ARK_MODEL=doubao-seed-code

Priority 2 — Model Proxy (Anthropic format)¶

Useful if you run anthropic-proxy locally to bridge Claude API.

env MODEL_PROXY_URL=http://127.0.0.1:43191 MODEL_PROXY_KEY=

Priority 3 — Direct OpenAI-compatible¶

Works with OpenAI, Ollama, vLLM, Together, Groq, OpenRouter — anything that speaks OpenAI's chat completions API.

env OPENAI_API_KEY= OPENAI_BASE_URL=https://api.openai.com/v1 # or http://localhost:11434/v1 for Ollama OPENAI_MODEL=gpt-4o-mini

Channels¶

Env	Channel	Purpose
`PORT`	HTTP	Port for `HttpChannel` (default 3456)
`FEISHU_APP_ID` `FEISHU_APP_SECRET` `FEISHU_VERIFICATION_TOKEN`	Feishu	Feishu / Lark bot
`WECOM_CORP_ID` `WECOM_AGENT_ID` `WECOM_SECRET`	WeCom	企业微信 bot
`TELEGRAM_BOT_TOKEN`	Telegram	Telegram bot
`GITHUB_WEBHOOK_SECRET` `GITHUB_TOKEN`	GitHub	GitHub webhook + reply

Image Generation¶

env IMAGE_API_URL= IMAGE_API_KEY= IMAGE_MODEL=flux-1.1-pro

Memory¶

env MEMORY_DB_PATH=./.data/memory.sqlite MEMORY_COMPRESS_THRESHOLD=50 # entries above this trigger compression

Sandbox¶

env SANDBOX_TIMEOUT_MS=10000 # run_code timeout SANDBOX_MAX_MEMORY_MB=128

Logging¶

env LOG_LEVEL=info # debug | info | warn | error

Profile Defaults¶

The 6 built-in profiles:

Profile	Tools Included
`minimal`	run_code only
`coding`	file, search, shell, run_code
`research`	web, run_code, todo, memory
`full`	all 16 tools
`coding-with-web`	coding + web
`research-with-code`	research + file/search/shell

You can always add more categories on top:

ts createAgent({ profile: "coding", extraCategories: ["web", "image"], });