FlowMind turns common developer operations into skills: load local knowledge, query logs, review code, check design rules, and reuse what it learns from your corrections.

中文 | Integration | Skills | Security | Changelog

Install

npm install -g flowmind
flowmind doctor

Try it without any provider or MCP configuration:

flowmind demo log-audit
flowmind demo code-review
flowmind skills

Runtime supports Node.js 18 and newer.

One-Step Setup

Initialize FlowMind's AI, resource, component, and skill config from a portable Markdown package:

flowmind config --import config/flowmind-config-package.example.md --dry-run --json
flowmind config --import config/flowmind-config-package.example.md

This creates the local files FlowMind reads at runtime:

~/.flowmind/config.json
~/.flowmind/ai-config.json
~/.flowmind/resource-config.json
~/.flowmind/component-config.json
~/.flowmind/skill-config.json

Use your own package by copying config/flowmind-config-package.example.md and replacing MCP server names, project IDs, namespaces, and provider settings.

Use From Model Tools

Codex

Use the JSON-friendly wrapper:

flowmind-codex skills
flowmind-codex --skill log-audit "query traceId abc123"
flowmind-codex doctor

By default, Codex mode shares learning and resource state with MCP and the regular CLI under ~/.flowmind/. Set FLOWMIND_SHARED_HOME=/path/to/home to choose the shared parent directory. Set FLOWMIND_CODEX_HOME=/path/to/state only when an isolated Codex store is required.

Claude

Install FlowMind as a Claude MCP server:

flowmind install claude

Install Claude and initialize FlowMind config in one command:

flowmind install claude \
  --flowmind-home /absolute/path/to/flowmind-home \
  --flowmind-config config/flowmind-config-package.example.md

Restart Claude after changing MCP configuration.

WorkBuddy

Install FlowMind as a WorkBuddy MCP server:

flowmind install workbuddy

Install WorkBuddy and initialize FlowMind config in one command:

flowmind install workbuddy \
  --flowmind-home /absolute/path/to/flowmind-home \
  --flowmind-config config/flowmind-config-package.example.md

If WorkBuddy uses a custom MCP config path:

flowmind install workbuddy --config /absolute/path/to/workbuddy-mcp.json

Manual config example: config/workbuddy-mcp-config.example.json.

Common Workflows

Goal	Command
Load local knowledge and inspect learned assets	`flowmind learn --assets --json`
Query logs or trace chains	`flowmind process --skill log-audit "query traceId abc123"`
Review code with fixed rules	`flowmind process --skill code-review "review the staged changes"`
Analyze design and implementation flow locally	`flowmind analyze "summarize local design and code flow" --offline`
Validate data logic	`flowmind process --skill data-logic-validation "check order data consistency"`
Run guarded workflows	`flowmind workflow --chain follow-up-chain.json`

FlowMind can run without an AI provider for deterministic local analysis and demos. Real log, database, document, and deployment skills need the matching MCP resources or adapters configured.

Decision Provider Routing

FlowMind 2.3.1 adds explicit decision-provider routing for skill selection. Decision providers such as jev answer typed choice questions and never take over chat or completion calls. They remain dormant until you enable them and provide credentials:

{
  "ai": {
    "providers": {
      "jev": {
        "enabled": true,
        "apiKey": "${JEV_API_KEY}",
        "model": "jev-latest"
      }
    },
    "featureProviders": {
      "skillSelection": "jev"
    }
  }
}

If the decision provider is disabled, fails, or has low confidence, FlowMind falls back to the default LLM provider. See the provider adapter guide.

Automatic Learning

FlowMind stores explicit corrections as local reusable state:

flowmind process --skill log-audit "query traceId abc123"
flowmind "next time use a table"

The next similar request can reuse that formatting preference. Skill choices are learned too:

flowmind process --skill offline-design-analysis "analyze the contract limit validation flow"

Later similar analysis requests prefer the learned route.

Local State

FlowMind reads and writes local state under:

FLOWMIND_HOME/.flowmind

If FLOWMIND_HOME is not set, it uses your user home directory. Files that may contain credentials or workflow state are written with private permissions on POSIX systems.