llmswap

Package Overview

Dependencies

Maintainers

Versions

Alerts

File Explorer

Advanced tools

License

Install Socket

Detect and block malicious and high-risk dependencies

Install

llmswap

Universal LLM SDK with Tool Calling: Enable ANY LLM to access YOUR data/APIs. Multi-provider (Anthropic, OpenAI, Gemini, Groq, xAI) with automatic format conversion. Features: universal tool calling, workspace memory, cost optimization, provider switching, zero vendor lock-in. Build AI agents, chatbots, data assistants with one SDK.

PyPI

Version: 5.2.2

Maintainers: 1

llmswap: Universal LLM SDK That Just Works

Ship AI Features in Minutes, Not Weeks

One SDK. 11 providers. Zero vendor lock-in.

Stop wrestling with complex frameworks. llmswap gives you tool calling, caching, and multi-provider support in 10 lines of code. Used by developers shipping real products.

📚 Documentation: llmswap.org | ⚡ CLI Reference: CLI Docs | 🐍 SDK Guide: SDK Docs

🆕 NEW in v5.2.0: Universal Tool Calling

Enable LLMs to access YOUR data and systems - Define tools once, works across ALL providers.

from llmswap import LLMClient, Tool

# Define tool to access YOUR weather API
weather = Tool(
    name="get_weather",
    description="Get real-time weather data",
    parameters={"city": {"type": "string"}},
    required=["city"]
)

# Works with ANY provider - Anthropic, OpenAI, Gemini, Groq, xAI
client = LLMClient(provider="anthropic")
response = client.chat("What's the weather in Tokyo?", tools=[weather])

# LLM calls YOUR function → you return data → LLM gives natural response

Real-World Use Cases:

🌦️ Give LLM access to YOUR weather API for real-time data
💾 Let LLM query YOUR database for customer information
🛒 Enable LLM to search YOUR product catalog for shopping assistance
🔧 Connect LLM to YOUR systems and APIs

Works with: Anthropic, OpenAI, Groq, Gemini, xAI | Quick Start Guide → | Full Docs →

⚡ Quick Start (30 seconds)

# Install
pip install llmswap

# or Homebrew
brew tap llmswap/tap && brew install llmswap

# Create your first workspace
cd ~/my-project
llmswap workspace init

# Chat with AI that remembers everything
llmswap chat "Help me with Flask routing"
# AI has full project context + all past learnings!

# 🆕 Compare models visually (optional)
pip install llmswap[web]
llmswap web  # Opens browser - compare GPT-4 vs Claude vs Gemini

🆕 Use Any Model from Any Provider! New model just launched? Use it immediately. llmswap's pass-through architecture means GPT-5, Claude Opus 4, Gemini 2.5 Pro work the day they release. Currently supports 10 providers (OpenAI, Anthropic, Gemini, Cohere, Perplexity, IBM watsonx, Groq, Ollama, xAI Grok, Sarvam AI).

✅ Battle-Tested with LMArena Top Models: All 10 providers tested and validated with top-rated models from LMArena leaderboard. From Grok-4 (xAI's flagship) to Claude Sonnet 4.5 (best coding model) to Gemini 2.0 Flash Exp - every model in our defaults is production-validated and arena-tested for real-world use.

The First AI Tool with Project Memory & Learning Journals - LLMSwap v5.1.0 introduces revolutionary workspace system that remembers your learning journey across projects. Build apps without vendor lock-in (SDK) or use from terminal (CLI). Works with your existing subscriptions: Claude, OpenAI, Gemini, Cohere, Perplexity, IBM watsonx, Groq, Ollama, xAI Grok, Sarvam AI (10 providers). Use any model from your provider - even ones released tomorrow. Pass-through architecture means GPT-5, Gemini 2.5 Pro, Claude Opus 4? They work the day they launch.

🎯 Solve These Common Problems:

❌ "I need multiple second brains for different aspects of my life" 🆕
❌ "AI strays over time, I need to re-steer it constantly" 🆕
❌ "I keep explaining the same context to AI over and over"
❌ "AI forgets what I learned yesterday"
❌ "I lose track of architecture decisions across projects"
❌ "Context switching between projects is exhausting"
❌ "I want AI to understand my specific codebase, not generic answers"

✅ llmswap v5.1.0 Solves All These:

✅ Multiple independent "second brains" per project/life aspect 🆕
✅ Persistent context prevents AI from straying 🆕
✅ Per-project workspaces that persist context across sessions
✅ Auto-tracked learning journals - never forget what you learned
✅ Architecture decision logs - all your technical decisions documented
✅ Zero context switching - AI loads the right project automatically
✅ Project-aware AI - mentor understands YOUR specific tech stack

Why Developers Choose llmswap

✅ 10 Lines to Production - Not 1000 like LangChain ✅ Automatic Fallback - Never down. Switches providers if one fails ✅ 50-90% Cost Savings - Built-in caching. Same query = FREE ✅ Workspace Memory - Your AI remembers your project context ✅ Universal Tool Calling - Define once, works everywhere (NEW v5.2.0) ✅ CLI + SDK - Code AND terminal. Your choice ✅ Zero Lock-in - Switch from OpenAI to Claude in 1 line

Built for Speed:

🚀 Hackathons - Ship in hours
💡 MVPs - Validate ideas fast
📱 Production Apps - Scale as you grow
🎯 Real Projects - Trusted by developers worldwide

v5.1.0: Revolutionary AI mentorship with project memory, workspace-aware context, auto-tracked learning journals, and persistent mentor relationships. The first AI tool that truly remembers your learning journey across projects.

NEW in v5.2.0:

🛠️ Universal Tool Calling - Enable LLMs to use YOUR custom functions across all providers
🔧 5 Providers Supported - Anthropic, OpenAI, Groq, Gemini, xAI with automatic format conversion
📖 Complete Documentation - Full guides, examples, and real-world use cases
✅ 100% Backward Compatible - All existing features work without changes

v5.1.6:

🌐 Web UI - Compare 20+ models side-by-side in beautiful browser interface & learn prompting techniques
📊 Visual Comparison - Live streaming results with speed badges (⚡🥈🥉), cost charts, efficiency metrics
💰 Cost Optimizer - See exact costs across providers, find cheapest model for your use case
🎨 Markdown + Code Highlighting - Syntax-highlighted code blocks with individual copy buttons
💾 Smart Preferences - Remembers your favorite models via localStorage
📈 Real-time Metrics - Tokens/sec efficiency, response length indicators, actual API token counts

NEW in v5.1.0:

🧠 Workspace Memory - Per-project context that persists across sessions
📚 Auto-Learning Journal - Automatically tracks what you learn in each project
🎯 Context-Aware Mentorship - AI mentor understands your project and past learnings
📖 Architecture Decision Log - Document and remember key technical decisions
🔄 Cross-Project Intelligence - Learn patterns from one project, apply to another
💡 Proactive Learning - AI suggests next topics based on your progress
🗂️ Project Knowledge Base - Custom prompt library per workspace

🧠 Finally: An Elegant Solution for Multiple Second Brains

The Problem Industry Leaders Can't Solve:

"I still haven't found an elegant solution to the fact that I need several second brains for the various aspects of my life, each with different styles and contexts." - Industry feedback

The LLMSwap Solution: Workspace System

Each aspect of your life gets its own "brain" with independent memory:

💼 Work Projects - ~/work/api-platform - Enterprise patterns, team conventions
📚 Learning - ~/learning/rust - Your learning journey, struggles, progress
🚀 Side Projects - ~/personal/automation - Personal preferences, experiments
🌐 Open Source - ~/oss/django - Community patterns, contribution history

What Makes It "Elegant":

✅ Zero configuration - just cd to project directory
✅ Auto-switching - AI loads the right "brain" automatically
✅ No context bleed - work knowledge stays separate from personal
✅ Persistent memory - each brain remembers across sessions
✅ Independent personas - different teaching style per project if you want

Stop Re-Explaining Context. Start Building.

🎯 Transform AI Into Your Personal Mentor with Project Memory

Inspired by Eklavya - the legendary self-taught archer who learned from dedication and the right guidance - LLMSwap transforms any AI provider into a personalized mentor that adapts to your learning style and remembers your journey.

The Challenge: Developers struggle to learn effectively from AI because:

🔴 Responses are generic, lack personality, and don't adapt to individual needs
🔴 AI loses context between sessions - you repeat the same explanations
🔴 No learning history - AI doesn't know what you already learned
🔴 Project context is lost - AI doesn't understand your codebase

LLMSwap's Solution v5.1.0: Choose your mentorship style, initialize a workspace, and ANY AI provider becomes your personalized guide that remembers everything:

# 🆕 v5.1.0: Initialize workspace for your project
cd ~/my-flask-app
llmswap workspace init
# Creates .llmswap/ with context.md, learnings.md, decisions.md

# Now your AI mentor KNOWS your project
llmswap chat --mentor guru --alias "Guruji"
# Mentor has full context: your tech stack, past learnings, decisions made

# 🆕 Auto-tracked learning journal
# Every conversation automatically saves key learnings
llmswap workspace journal
# View everything you've learned in this project

# 🆕 Architecture decision log
llmswap workspace decisions
# See all technical decisions documented automatically

# View all your workspaces
llmswap workspace list

# Get wisdom and deep insights from a patient teacher
llmswap chat --mentor guru --alias "Guruji"

# High-energy motivation when you're stuck
llmswap ask "How do I debug this?" --mentor coach

# Collaborative peer learning for exploring ideas
llmswap chat --mentor friend --alias "CodeBuddy"

# Question-based learning for critical thinking
llmswap ask "Explain REST APIs" --mentor socrates

# 🆕 Use Claude Sonnet 4.5 - Best coding model
llmswap chat --provider anthropic --model claude-sonnet-4-5
# Or set as default in config for all queries

🔄 Rotate Personas to Expose Blind Spots

Industry Insight: "Rotate personas: mentor, skeptic, investor, end-user. Each lens exposes blind spots differently."

Use Case: Reviewing API Design

# Round 1: Long-term wisdom
llmswap chat --mentor guru "Design API for multi-tenant SaaS"
# Catches: scalability, technical debt, maintenance

# Round 2: Critical questions
llmswap chat --mentor socrates "Review this API design"
# Catches: assumptions, alternatives, edge cases

# Round 3: Practical execution
llmswap chat --mentor coach "What's the fastest path to v1?"
# Catches: over-engineering, paralysis by analysis

Same project context. Different perspectives. Complete understanding.

What Makes v5.1.0 Revolutionary:

🧠 Works with ANY provider - Transform Claude, GPT-4, or Gemini into your mentor
🎭 6 Teaching Personas - Guru, Coach, Friend, Socrates, Professor, Tutor
📊 Project Memory - Per-project context that persists across sessions ⭐ NEW
📚 Auto-Learning Journal - Automatically tracks what you learn ⭐ NEW
📖 Decision Tracking - Documents architecture decisions ⭐ NEW
🎓 Age-Appropriate - Explanations tailored to your level (--age 10, --age 25, etc.)
💰 Cost Optimized - Use cheaper providers for learning, premium for complex problems
🔄 Workspace Detection - Automatically loads project context ⭐ NEW

Traditional AI tools give you answers. LLMSwap v5.1.0 gives you a personalized learning journey that REMEMBERS.

🏆 Production-Validated with LMArena Top Models

Every model in LLMSwap's defaults comes from LMArena's top performers:

All 10 providers ship with carefully selected default models based on LMArena rankings and real-world production testing. We track arena performance and update defaults to ensure you're always using validated, battle-tested models.

Provider	Default Model	Arena Status	Why We Chose It
Anthropic	claude-sonnet-4-5	#1 Coding	Best coding model in the world (Sept 2025)
xAI	grok-4-0709	Top 5 Overall	Advanced reasoning, real-time data access
Gemini	gemini-2.0-flash-exp	Top 10	Lightning-fast, multimodal, cutting-edge
OpenAI	gpt-4o-mini	Cost Leader	Best price/performance ratio
Cohere	command-r-08-2024	Top RAG	Enterprise-grade retrieval-augmented generation
Perplexity	sonar	Web Search	Real-time web-connected AI with citations
Groq	llama-3.1-8b-instant	Speed King	840+ tokens/second ultra-fast inference
Sarvam	sarvam-m	Multilingual	24B params, best for 10 Indian languages
Watsonx	granite-3-8b-instruct	Enterprise	IBM's production-grade AI for business
Ollama	granite-code:8b	Local AI	Privacy-first, runs on your hardware

✅ Battle-tested with real API calls - Every provider validated in production, not simulated tests.

✅ Weekly model updates - We monitor LMArena rankings and deprecation notices to keep defaults current.

✅ Zero lock-in - Don't like our defaults? Override with any model: LLMClient(model="gpt-5") or llmswap config set provider.models.openai gpt-5

🔓 Use Any Model Your Provider Supports (Zero-Wait Model Support)

Here's something cool: LLMSwap doesn't restrict which models you can use. When GPT-5 or Gemini 2.5 Pro drops tomorrow, you can start using it immediately. No waiting for us to update anything.

How? We use pass-through architecture. Whatever model name you pass goes directly to your provider's API. We don't gatekeep.

CLI Examples:

# Use any OpenAI model (even ones that don't exist yet)
llmswap chat --provider openai --model gpt-5
llmswap chat --provider openai --model o3-mini

# Use any Anthropic model
llmswap chat --provider anthropic --model claude-opus-4
llmswap chat --provider anthropic --model claude-sonnet-4-5

# Use any Gemini model
llmswap chat --provider gemini --model gemini-2-5-pro
llmswap chat --provider gemini --model gemini-ultra-2

# Set as default so you don't have to type it every time
llmswap config set provider.models.openai gpt-5
llmswap config set provider.models.anthropic claude-opus-4

Python SDK:

from llmswap import LLMClient

# Use whatever model your provider offers
client = LLMClient(provider="openai", model="gpt-5")
client = LLMClient(provider="anthropic", model="claude-opus-4")
client = LLMClient(provider="gemini", model="gemini-2-5-pro")

# Model just released? Use it right now
client = LLMClient(provider="openai", model="gpt-6")  # works!

The point: You're not limited to what we've documented. If your provider supports it, llmswap supports it.

🆚 LLMSwap vs Single-Provider Tools

For Python Developers Building Apps:

Your Need	Single-Provider SDKs	LLMSwap SDK
Build chatbot/app	Import `openai` library (locked in)	Import `llmswap` (works with any provider)
Switch providers	Rewrite all API calls	Change 1 line: `provider="anthropic"`
Try different models	Sign up, new SDK, refactor code	Just change config, same code
Use new models	Wait for SDK update	Works immediately (pass-through)
Cost optimization	Manual implementation	Built-in caching (50-90% savings)
Use multiple providers	Maintain separate codebases	One codebase, switch dynamically

For Developers Using Terminal:

Your Need	Vendor CLIs	LLMSwap CLI
Have Claude subscription	Install Claude Code (Claude only)	Use llmswap (works with Claude)
Have OpenAI subscription	Build your own scripts	Use llmswap (works with OpenAI)
Have multiple subscriptions	Install 3+ different CLIs	One CLI for all subscriptions
New model launches	Wait for CLI update	Use it same day (pass-through)
Want AI to teach you	Not available	Built-in Eklavya mentorship
Switch providers mid-chat	Can't - locked in	`/switch anthropic` command

The Bottom Line:

Building an app? Use llmswap SDK - no vendor lock-in
Using terminal? Use llmswap CLI - works with your existing subscriptions
Both? Perfect - it's the same tool!

# 🆕 NEW v5.1.0: Workspace System - Project Memory That Persists
llmswap workspace init
# Creates .llmswap/ directory with:
#   - workspace.json (project metadata)
#   - context.md (editable project description)
#   - learnings.md (auto-tracked learning journal)
#   - decisions.md (architecture decision log)

llmswap workspace list                    # View all your workspaces
llmswap workspace info                    # Show current workspace statistics
llmswap workspace journal                 # View learning journal
llmswap workspace decisions               # View decision log
llmswap workspace context                 # Edit project context

# 🆕 NEW v5.1.0: Context-Aware Mentorship
# AI mentor automatically loads project context, past learnings, and decisions
llmswap chat
# Mentor knows: your tech stack, what you've learned, decisions made

# 🆕 NEW v5.0: Age-Appropriate AI Explanations
llmswap ask "What is Docker?" --age 10
# Output: "Docker is like a magic lunch box! 🥪 When your mom packs..."

llmswap ask "What is blockchain?" --audience "business owner"
# Output: "Think of blockchain like your business ledger system..."

# 🆕 NEW v5.0: Teaching Personas & Personalization  
llmswap ask "Explain Python classes" --teach --mentor developer --alias "Sarah"
# Output: "[Sarah - Senior Developer]: Here's how we handle classes in production..."

# 🆕 NEW v5.0: Conversational Chat with Provider Switching
llmswap chat --age 25 --mentor tutor
# In chat: /switch anthropic  # Switch mid-conversation
# In chat: /provider         # See current provider
# Commands: /help, /switch, /clear, /stats, /quit

# 🆕 NEW v5.0: Provider Management & Configuration
llmswap providers                    # View all providers and their status
llmswap config set provider.models.cohere command-r-plus-08-2024
llmswap config set provider.default anthropic
llmswap config show

# Code Generation (GitHub Copilot CLI Alternative)
llmswap generate "sort files by size in reverse order"
# Output: du -sh * | sort -hr

llmswap generate "Python function to read JSON with error handling" --language python
# Output: Complete Python function with try/catch blocks

# Advanced Log Analysis with AI
llmswap logs --analyze /var/log/app.log --since "2h ago"
llmswap logs --request-id REQ-12345 --correlate

# Code Review & Debugging
llmswap review app.py --focus security
llmswap debug --error "IndexError: list index out of range"

# ❌ Problem: Vendor Lock-in
import openai  # Locked to OpenAI forever
client = openai.Client(api_key="sk-...")
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello"}]
)
# To switch to Claude? Rewrite everything.

# ✅ Solution: llmswap SDK - Universal Interface
from llmswap import LLMClient

# Works with any provider you're subscribed to
client = LLMClient()  # Auto-detects from env vars
response = client.query("Hello")

# Want Claude instead? Just change provider:
client = LLMClient(provider="anthropic")  # That's it!

# Want to try Gemini? 
client = LLMClient(provider="gemini")  # Same code, different provider

# Built-in cost optimization:
# - Automatic response caching (50-90% savings)
# - Provider cost comparison
# - Smart provider selection based on query type

🆕 v5.1.0: Workspace System - Real-World Scenarios

🎯 Scenario 1: New Developer Learning Flask

Problem: Junior developer learning Flask keeps asking AI the same questions because AI forgets previous conversations.