@awareness.market/setup

@awareness.market/setup

One command to give your AI agent persistent memory. 13+ IDEs supported.

npx @awareness.market/setup

Auto-detects your IDE, starts a local memory daemon, injects workflow rules, and configures MCP — all in one command. No account needed. Works offline.

How It Works

npx @awareness.market/setup
    │
    ├── Detects IDE (Cursor? Claude Code? Windsurf? ...)
    ├── Starts local daemon (localhost:37800)
    ├── Writes workflow rules (teaches agent when to recall/record)
    ├── Writes MCP config (connects agent to memory daemon)
    └── Done ✅ — your agent now has persistent memory

Modes

Local Mode (Default)

npx @awareness.market/setup

• Starts a local daemon on localhost:37800
• Memories stored as Markdown files in .awareness/
• No account, no cloud, no API key needed
• Works offline

Cloud Mode

npx @awareness.market/setup --cloud

• Opens browser for device-auth login
• Select or create a memory entity
• Enables cloud sync (semantic search, team collaboration, memory marketplace)
• Local daemon still runs — cloud is a sync layer, not a replacement

Squad Mode (one lead + N workers)

npx @awareness.market/setup --squad 3

Generates ready-to-paste prompts and installs the squad skill:

• squad/lead.md — lead prompt (decompose → write tasks → monitor, never execute)
• squad/worker-1.md … worker-N.md — worker prompts (claim with CAS → work → hand off / finish)
• squad/run.md — full walkthrough (plugin config, per-OS role commands, progress)
• .opencode/skills/squad/SKILL.md — the squad skill

Cloud mode is required for cross-machine squads; local mode works on one machine. Open one terminal per role, set AWARENESS_AGENT_ROLE=lead|worker-1..N, and paste the matching prompt. Progress is visible via awareness_task_board and on the memory's Task board tab in the browser.

Supported IDEs

IDE	Rules File	MCP Config
Cursor	.cursor/rules/awareness.mdc	.cursor/mcp.json
Claude Code	CLAUDE.md	.mcp.json
Windsurf	.windsurfrules	~/.codeium/windsurf/mcp_config.json
Cline	.clinerules	UI-based (copy-paste)
GitHub Copilot	.github/copilot-instructions.md	.vscode/mcp.json
Codex CLI	AGENTS.md	.codex/config.toml
Kiro	.kiro/steering/awareness.md	.kiro/settings/mcp.json
Trae	.trae/rules/awareness.md	.trae/mcp.json
Zed	.rules	~/.config/zed/settings.json
JetBrains (Junie)	.junie/guidelines.md	.junie/mcp/mcp.json
Augment	.augment/rules/awareness.md	UI-based (copy-paste)
AntiGravity (Jules)	.antigravity/rules.md	~/.gemini/antigravity/mcp_config.json
OpenClaw	Plugin-based	~/.openclaw/openclaw.json

All Options

npx @awareness.market/setup                 Local mode (default)
npx @awareness.market/setup --cloud         Cloud mode (login + sync)
npx @awareness.market/setup --ide cursor    Force specific IDE
npx @awareness.market/setup --no-auth       Rules only, no MCP config
npx @awareness.market/setup --dry-run       Preview without writing
npx @awareness.market/setup --force         Overwrite managed files
npx @awareness.market/setup --list          Show supported IDEs
npx @awareness.market/setup --logout        Clear saved credentials
npx @awareness.market/setup --squad <n>     Generate squad prompts (1 lead + n workers, 2-8)
npx @awareness.market/setup --api-base <url>  Custom API URL

Manual Cloud Config

npx @awareness.market/setup --mcp-url <url> --api-key <key> --memory-id <id>

Non-Interactive Mode

When run without a TTY (e.g., from an AI agent via Bash), the CLI auto-selects all detected IDEs and defaults to English embedding model. No prompts.

Embedding Language

During interactive setup, you can choose:

Search language:
  1. English only  (23 MB, faster)      ← default
  2. All languages (118 MB, 中文/日本語/한국어...)

What It Writes

Workflow Rules

Injected into your IDE's rules file (e.g., CLAUDE.md, .cursor/rules/awareness.mdc). Teaches your AI agent to:

• Call awareness_init at session start
• Call awareness_recall before making decisions
• Call awareness_record after every code change

MCP Config

Connects your IDE to the memory daemon:

{
  "mcpServers": {
    "awareness-memory": {
      "url": "http://localhost:37800/mcp"
    }
  }
}

Related Packages

Package	Description
@awareness.market/local	Local daemon that this CLI starts
@awareness.market/memory-cloud	TypeScript SDK with interceptors
awareness-memory-cloud	Python SDK with interceptors
@awareness.market/openclaw-memory	OpenClaw plugin

What makes Awareness different

Most memory systems pick one extraction strategy. Awareness combines them:

• Hybrid retrieval by default — BM25 full-text + vector cosine + knowledge-graph 1-hop expansion, fused with Reciprocal Rank Fusion. 96.0% R@5 on LongMemEval, zero LLM calls on the retrieval side.
• Salience-aware extraction (v0.4.7+) — the client's LLM self-scores every card on novelty / durability / specificity; cards below 0.4 on novelty or durability are dropped server-side. Framework metadata is filtered before extraction runs, so raw logs never leak into your knowledge base.
• Project isolation — X-Awareness-Project-Dir header scopes memory per project.
• Zero-LLM backend — all extraction runs on the client's LLM. No inference costs pass through to you.
• One memory, many clients — same daemon reachable via Claude Code / Cursor / Windsurf / OpenClaw / plain MCP.

See docs/analysis/MEMPALACE_COMPARISON_2026-04-17.md for the honest side-by-side against MemPalace (96.6% R@5 via raw verbatim storage).

License

Apache 2.0