Skip to content

xt8086/wmcli

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
scripts
scripts
 
 
src
src
 
 
.gitignore
.gitignore
 
 
README.md
README.md
 
 
package-lock.json
package-lock.json
 
 
package.json
package.json
 
 
tsconfig.json
tsconfig.json
 
 

Repository files navigation

wmcli

CLI task runner with expert sub-agent delegation. Models from any provider.

A single AI coordinator orchestrates specialist sub-agents for coding, vision, and web research — all from your terminal.

wmcli > analyze this error log and find a fix
wmcli > what does this screenshot show?
wmcli > what's the latest on Rust 2.0?

Features

  • 972 models across 32 providers — OpenRouter, OpenAI, Anthropic, Google, DeepSeek, xAI, Groq, Mistral, and more
  • Expert delegation — main model spawns coding, vision, and research sub-agents automatically
  • Per-provider API keys — start with one OpenRouter key, add direct provider keys later
  • Hot-swap models — switch main or vision model at runtime, no restart
  • Thinking/reasoning support — steerable thinking levels for reasoning models
  • Session persistence — conversations survive restarts
  • Semantic memory — search past sessions by meaning, not just keywords
  • Markdown terminal rendering — tables, bold, code blocks render properly
  • Live web research — Perplexity Sonar for current data

Prerequisites

  • Node.js ≥ 20 (install)
  • C++ build tools — required for the SQLite native module: 
    # Ubuntu/Debian
sudo apt install build-essential python3

# Fedora
sudo dnf install gcc-c++ make python3

# macOS
xcode-select --install

# Windows — install Visual Studio Build Tools
# (or run: npm install -g windows-build-tools)

Install

npm install -g @wondermotor/wmcli

The postinstall script automatically rebuilds the SQLite native module for your platform.

On first run, it prompts for an OpenRouter API key. Get one free at openrouter.ai/keys.

You can also set the key via environment variable:

export OPENROUTER_API_KEY=sk-or-v1-...
wmcli

Quick Start

wmcli
  ┌──────────────────────────────────────────────────┐
  │  API key needed for OpenRouter                   │
  ├──────────────────────────────────────────────────┤
  │  Get a key at: https://openrouter.ai/keys        │
  │  Or set env: OPENROUTER_API_KEY                  │
  └──────────────────────────────────────────────────┘

  Enter OpenRouter API key (or press Enter to skip): sk-or-v1-...
  ✅ API key saved. You can change it anytime with /key.

  Main:   openrouter/z-ai/glm-5.1 (reasoning)
  Vision: openrouter/google/gemini-2.0-flash-001 (image description)
  Tools: web_fetch, recall_sessions
  Experts: delegate_coding, delegate_vision, delegate_research
  Type /help for commands, /quit to exit.

wmcli >

One key covers everything — main model, vision, and research all route through OpenRouter.

Commands

Command Action
/model [search] List/search 972 models across all providers
/model <number> Select main model from search results (e.g. /model 5)
/model <provider/id> Switch directly (e.g. /model anthropic/claude-sonnet-4)
/vision [search] List/search vision-capable models
/vision <number> Select vision model by number (hot-swaps at runtime)
/vision <provider/id> Switch vision model directly (e.g. /vision google/gemini-2.0-flash)
/key Show all configured API keys (masked)
/key <provider> Show key for a specific provider
/key set <provider> Add an API key for a provider
/key clear <provider> Remove a provider's stored key
/thinking or /t Cycle thinking level (off → minimal → low → medium → high)
/thinking <level> Set directly (e.g. /t high, /t off)
/status Show model, vision, thinking, and key status
/memory or /m Show past indexed sessions
/quit or /q Exit (saves session to memory)

Supported Providers

wmcli works with any provider in pi's ModelRegistry. Common providers with API key support:

Provider Env Var Key URL
OpenRouter OPENROUTER_API_KEY openrouter.ai/keys
OpenAI OPENAI_API_KEY platform.openai.com/api-keys
Anthropic ANTHROPIC_API_KEY console.anthropic.com
Google AI GEMINI_API_KEY aistudio.google.com/apikey
DeepSeek DEEPSEEK_API_KEY platform.deepseek.com
xAI XAI_API_KEY console.x.ai
Groq GROQ_API_KEY console.groq.com/keys
Mistral MISTRAL_API_KEY console.mistral.ai
Cerebras CEREBRAS_API_KEY cloud.cerebras.ai
Fireworks FIREWORKS_API_KEY app.fireworks.ai
Hugging Face HF_TOKEN huggingface.co/settings/tokens

All provider keys can also be set at runtime with /key set <provider>.

Going Direct: OpenRouter → Provider API Keys

Start with OpenRouter. Add direct keys when you want lower latency or direct billing:

wmcli > /model anthropic/claude-sonnet-4
  ⚠️  No API key for Anthropic. Use /key set anthropic

wmcli > /key set anthropic
  Set API key for Anthropic
  Get a key at: https://console.anthropic.com/settings/keys
  Enter Anthropic API key: sk-ant-...
  ✅ Anthropic API key saved.

wmcli > /vision google/gemini-2.0-flash
  ✅ Vision model switched to google/gemini-2.0-flash
  ⚠️  No API key for Google AI. Use /key set google

Models from providers without a configured key show ⚠️ in /model and /vision search results.

CLI Flags

# Set runtime-only keys (not persisted to disk)
wmcli --api-key openrouter=sk-or-v1-... --api-key anthropic=sk-ant-...

# Backward compat: bare key defaults to openrouter
wmcli --api-key sk-or-v1-...

Session Persistence & Memory

  • Session resume: restarting wmcli continues your last conversation — no context lost
  • Semantic memory: on exit, session content is embedded and stored in ~/.wmcli/memory.db. The main model can recall past sessions via the recall_sessions tool using cosine similarity search
  • Embedding cost: ~$0.02 per 1M tokens (OpenRouter text-embedding-3-small, 1536 dimensions)

Architecture

User → [user-selected model] (Main Coordinator)
         │
         ├── delegate_coding   → [same as main] (isolated session)
         ├── delegate_vision   → vision model (image description)
         ├── delegate_research → Perplexity Sonar via OpenRouter (live web search)
         ├── recall_sessions   → SQLite + OpenRouter embeddings
         └── web_fetch          → Direct HTTP fetch
Role Default Model Purpose
Main openrouter/z-ai/glm-5.1 Reasoning, tools, coordination
Vision openrouter/google/gemini-2.0-flash-001 Image description, OCR
Coding Same as main Code, math, science (isolated session)
Research perplexity/sonar Live web search + synthesis
Embedding openai/text-embedding-3-small Session memory (1536 dims)

The system prompt includes the current date/time, timezone, locale, OS, shell, and working directory so the model has full context.

Config Files

File Purpose
~/.wmcli.json Model choices ({"model": "openrouter/z-ai/glm-5.1", "visionModel": "openrouter/google/gemini-2.0-flash-001"})
~/.pi/agent/auth.json API keys per provider (shared with pi)
~/.pi/agent/sessions/ Session history (shared with pi)
~/.wmcli/memory.db SQLite embedding store for session recall

Uninstall

npm uninstall -g @wondermotor/wmcli

# Config is automatically removed. Memory data is kept by default:
rm ~/.wmcli.json          # (already removed by uninstall)
rm -rf ~/.wmcli/          # remove memory database (optional)

# To remove everything including memory during uninstall:
WMCLI_UNINSTALL=full npm uninstall -g @wondermotor/wmcli

# Note: API keys in ~/.pi/agent/auth.json are shared with pi and NOT removed.

Build from Source

git clone https://github.com/xt8086/wmcli.git
cd wmcli
npm install          # installs deps + rebuilds native modules + patches pi SDK
npm run build        # compile TypeScript
npm link             # makes `wmcli` available globally

License

MIT

About

CLI task runner with expert sub-agent delegation — models from any provider

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

2 tags

Packages

 
 
 

Contributors

Languages