weave

Graph-native memory CLI for AI agents. Chat with AI that actually remembers — across sessions, across agents.

  ◈ weave v0.1.0
  graph-native memory for AI agents

What is Weave?

Weave is a terminal CLI agent (like Claude Code) with a twist: persistent, graph-structured memory. Every conversation is remembered. Memories are connected by semantic similarity, temporal sequence, and shared entities. Your agents build knowledge over time — they never cold-start.

Built on the MemWeave memory architecture:

• Multi-layer memory graph — semantic, temporal, causal, and entity edges
• Tiered memory — working → short-term → long-term → archival with automatic promotion and decay
• Multi-agent — spawn multiple agents with different personas that share a memory fabric
• Hybrid retrieval — vector similarity + graph traversal for smarter recall
• Local-first — works with just hash-based embeddings + SQLite, no cloud required for memory

Install

npm install -g weave-cli

Or from source:

git clone <repo>
cd cliproject
npm install
npm run build
npm link

Quick Start

# Initialize
weave init

# Set your API key
weave config set apiKey sk-your-openai-key

# Start chatting (memories persist automatically)
weave chat

# Or use Anthropic
weave config set provider anthropic
weave config set apiKey sk-ant-your-key
weave config set model claude-sonnet-4-20250514
weave chat

# Or use Ollama (local, no API key required)
weave config set provider ollama
weave config set model llama3.2
weave chat

# Or use LM Studio (local; start a model in LM Studio first)
weave config set provider lmstudio
weave config set model local-model
weave chat

Commands

Chat

weave chat                          # interactive chat with memory
weave chat --agent researcher       # chat as specific agent
weave chat --model gpt-4o-mini      # use a different model
weave chat --provider anthropic     # use Anthropic
weave chat --provider ollama        # use Ollama (local)
weave chat --provider lmstudio      # use LM Studio (local)

In-chat slash commands: | Command | Description | |---------|-------------| | /help | Show available commands | | /memory | Show working memory | | /recall <query> | Search all memories | | /agents | List all agents | | /switch <agent> | Switch to another agent | | /stats | Memory statistics | | /compact | Run memory consolidation | | /save | Force save | | /clear | Clear chat history (keeps memories) | | /exit | Exit weave |

Agents

weave agent spawn researcher --role "ML Researcher"
weave agent spawn engineer --role "Software Engineer" --model gpt-4o-mini
weave agent list
weave agent inspect researcher
weave agent kill researcher

Memory

weave memory add "The deadline is March 15" --agent researcher
weave memory search "project deadline"
weave memory search "tech stack" --agent researcher
weave memory list
weave memory list --agent researcher
weave memory consolidate               # promote, merge, decay, prune

Configuration

weave config set apiKey sk-...
weave config set provider openai       # or anthropic, ollama, lmstudio
weave config set model gpt-4o
weave config set baseURL http://localhost:11434/v1  # for ollama/lmstudio (optional)
weave config set embeddingBackend local # or openai
weave config list
weave config get provider

Workspaces

weave workspace create ml-research
weave workspace list
weave chat --workspace ml-research

Diagnostics

weave doctor                           # system health check

Memory Architecture

Working Memory (L1)   ← In-context, immediate access
       ↓ eviction
Short-Term Memory (L2) ← Recent, fast retrieval
       ↓ consolidation (importance-based)
Long-Term Memory (L3)  ← Consolidated, graph-structured
       ↓ aging
Archival Memory (L4)   ← Old, rarely accessed

Memories are automatically:

• Embedded using local hash projections (zero-latency) or OpenAI embeddings
• Linked by semantic similarity, temporal order, and shared entities
• Scored by importance × recency × access frequency
• Promoted from short-term to long-term when importance ≥ 0.4
• Decayed over time, with archival memories pruned when importance drops below 0.01
• Merged when duplicates are detected (cosine similarity ≥ 0.92)

Architecture

src/
├── core/
│   ├── types.ts        # Core type definitions
│   ├── embedding.ts    # Local hash + OpenAI embedding backends
│   ├── graph.ts        # Multi-layer memory graph with retrieval
│   ├── storage.ts      # SQLite persistence
│   ├── agent.ts        # Per-agent memory + extraction
│   └── fabric.ts       # Memory fabric orchestrator
├── llm/
│   └── provider.ts     # OpenAI + Anthropic with streaming
├── ui/
│   ├── theme.ts        # Terminal colors, icons, components
│   └── chat.ts         # Interactive chat REPL
├── config.ts           # Configuration management
└── index.ts            # CLI entry point

Config Storage

All data lives in ~/.weave/:

~/.weave/
├── config.json          # Global configuration
└── workspaces/
    ├── default.db       # Default workspace (SQLite)
    └── ml-research.db   # Named workspaces

Requirements

• Node.js ≥ 18
• An API key (OpenAI or Anthropic) for chat

License

MIT