Configuration

Model providers, MCP servers, and settings.

Alpha Release. Session history is persisted as raw JSON files in ~/.wmind/sessions/. There is no search, no tagging, no smart retrieval, and no way to browse sessions from outside the TUI. These will be improved in future releases.

Configuration File

Working Mind stores its configuration in ~/.wmind/config.json. You can edit it directly or use the --configure wizard.

Model Providers

Working Mind supports 7 primary providers. Any provider with an OpenAI-compatible API also works with custom configuration.

Local Fast (wmind-serve)

npm install -g wmind-serve
wmind-serve start
wmind --model local-fast

No API key needed. 4-5x faster than Ollama on Apple Silicon. All processing on your machine.

Ollama (Local)

ollama pull llama3.1
wmind --model ollama/llama3.1

No API key needed. All processing on your machine.

OpenRouter (Universal)

export OPENROUTER_API_KEY=sk-or-...
wmind --model openrouter/anthropic/claude-sonnet-4-20250514

Access to 200+ models including Anthropic, DeepSeek, and Google via OpenRouter.

Together AI

export TOGETHER_API_KEY=...
wmind --model together/meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8

Google

export GEMINI_API_KEY=...
wmind --model google/gemini-2.5-flash

OpenAI

export OPENAI_API_KEY=sk-...
wmind --model openai/gpt-5.4

Groq

export GROQ_API_KEY=...
wmind --model groq/llama-3.3-70b-versatile

MCP Servers

MCP (Model Context Protocol) servers provide tools for the agent. The starter pack declares two:

Server	Package	Required	What It Does
brave-search	`@brave/brave-search-mcp-server`	No	Web search for current information
firecrawl	`firecrawl-mcp`	No	Web scraping from URLs

Memory is built-in (native SQLite, no MCP server needed). Filesystem access is provided by a third MCP server (@modelcontextprotocol/server-filesystem).

Connecting MCP Servers

Use the /mcp-connect command in the TUI to configure servers interactively. You can also set API keys as environment variables:

export BRAVE_API_KEY=...
export FIRECRAWL_API_KEY=...
wmind

The agent will auto-connect to servers that have their required environment variables set.

Adding Custom MCP Servers

Any MCP server that uses the standard stdio transport works with Working Mind. Add it to your pack's pack.json:

{
  "mcpServers": {
    "my-server": {
      "package": "my-mcp-server-package",
      "required": false,
      "env": {
        "MY_API_KEY": {
          "setting": "MY_API_KEY",
          "sensitive": true,
          "required": true,
          "label": "My API Key",
          "hint": "Get one at https://example.com"
        }
      }
    }
  }
}

Settings

Configure in ~/.wmind/config.json under the agents key:

Setting	Default	Description
`maxTurns`	20	Maximum agent loop iterations per conversation
`autoApprove`	false	Auto-approve all tool calls without confirmation
`noThinking`	false	Disable reasoning/thinking output
`maxTokens`	(model default)	Maximum tokens per LLM response
`thinkingBudget`	(model default)	Reasoning token budget for thinking models

Environment Variables

Variable	Description
`OPENAI_API_KEY`	OpenAI API key
`OPENROUTER_API_KEY`	OpenRouter API key (access to Anthropic, DeepSeek, etc.)
`TOGETHER_API_KEY`	Together AI API key
`GEMINI_API_KEY`	Google AI API key
`GROQ_API_KEY`	Groq API key
`BRAVE_API_KEY`	Brave Search API key
`FIRECRAWL_API_KEY`	Firecrawl API key
`MEMORY_FILE_PATH`	Custom path for memory SQLite database
`WMIND_DEBUG`	Set to `1` to enable debug logging
`WMIND_PORT`	Custom port for wmind-serve (default: 19421)

Custom System Prompts

Override the pack's system prompt:

wmind --prompt "You are a helpful research assistant"

Or save named prompts in your config:

{
  "systemPrompts": {
    "analyst": "You are a thorough research analyst...",
    "writer": "You are a clear technical writer..."
  }
}

Then activate: wmind --prompt analyst

Knowledge Graph & Memory

How Working Mind remembers, compounds, and queries knowledge.

Architecture

How Working Mind works under the hood.