Build Local AI Agent Workflows
On Any Standard CPU
Sub-billion parameter local agents engineered for speed, privacy, and zero network latency. Run offline multi-agent routing, RAG retrieval, SQL generation, and voice synthesis on Apple Silicon, Intel, or AMD CPUs.
ONNX Runtime GenAI
Highly-optimized cross-platform CPU execution. Leverage hardware acceleration on Apple Silicon (ARM64), Intel, and AMD standard setups.
Shared Model Cache
All framework agents share a single local cache folder (`~/.cache/slm_summarizer/`). Running all libraries requires only ~1.1 GB disk space.
Permissive & Open Source
Built for privacy-first enterprise workflows. 100% offline edge inference with zero cloud data transmission and complete local data privacy.
def is_prime(n):
if n < 2:
return False
for i in range(2, int(n**0.5) + 1):
if n % i == 0:
return False
return True
fib = [0, 1]
while len(fib) < 10:
fib.append(fib[-1] + fib[-2])
print("First 10 Fibonacci Numbers & Primality:")
for idx, num in enumerate(fib, 1):
status = "PRIME" if is_prime(num) else "Not prime"
print(f"#{idx:02d}: {num:3d} -> {status}")The Python script computed the first 10 Fibonacci numbers and evaluated their primality. The prime Fibonacci numbers identified are 2, 3, 5, and 13.
Open Source Local AI Frameworks for Python
A collection of modular libraries designed to perform specialized agentic tasks using small models.
SLM Orchestrator
A lightweight, local semantic router. Dynamically maps user queries to specialized sub-agents with JSON constraints and ReAct tool use.
SLM RAG
Local Retrieval-Augmented Generation library. Answers user questions offline using secure, CPU-optimized ONNX models and custom tools.
SLM Summarizer
Local CPU text summarizer agent. Supports Map-Reduce for large documents and an evaluator-corrector reflection loop.
SLM Text-to-SQL
Translate natural language questions into database queries locally. Includes QLoRA adapter training workflows.
SLM CLI Agent
Lightweight shell execution and CLI automation companion. Translates queries to platform-specific scripts with sandbox safety.
SLM Code Interpreter
Secure local Python execution sandbox. Iteratively runs code blocks on CPU with automated self-correction tracebacks.
SLM Git Repo Manager
Commit generation helper. Automatically reviews branch updates and structures beautiful conventional commit messages locally.
SLM Document Parser
Parse docx, txt, and PDF structures locally to extract semantic hierarchies and configurations matching strict JSON schemas.
SLM Vision Parser
Offline chart, diagram, and whiteboard reader. Converts scanned infographics and drawings to structured text layouts.
SLM Web Agent
Local browser automation and navigation companion. Crawls pages and maps visual DOM nodes to selectors autonomously.
SLM Web Scraper
High-speed local parsing engine that converts raw HTML documents into structured schema JSON layouts.
SLM Search Orchestrator
Off-grid query expansion and multi-step routing companion. Integrates with DuckDuckGo to aggregate search snippets locally.
Complete Multi-Agent Ecosystem
Explore high-performance, secure, and 100% offline local SLM agents running with zero cloud latency.
SLM Web Agent
Automate multi-step browser tasks offline. Plans actions, selects locators, and navigates forms using Playwright on local CPU.
SLM Web Scraper
High-speed local parsing engine that converts raw HTML documents into structured schema JSON layouts without cloud scrapers.
SLM Search Orchestrator
Off-grid query expansion and multi-step routing companion. Aggregates search snippets locally and distills clean answers.
SLM Orchestrator
Autonomous meta-agent that inspects user queries, dynamically routes to local sub-agents, and sequences multi-step execution graphs.
SLM Text-to-SQL
Translates natural language questions into valid SQL queries against PostgreSQL, SQLite, and MySQL database schemas offline.
SLM Code Interpreter
Sandboxed Python execution environment for performing data analytics, plotting math charts, and evaluating algorithm logic.
SLM Git Repo Manager
Automates git commit generation, code review diff analysis, and branch conflict resolution offline on standard desktop hardware.
SLM Database Migrator
Analyzes legacy database schemas and generates zero-downtime, CPU-optimized migrations and modern ORM model definitions offline.
SLM Summarizer
Condenses lengthy reports, legal documents, and research papers into concise bullet points offline using Map-Reduce chunking.
SLM RAG
Answer questions from your own documents locally, privately, with zero API costs using quantized ONNX vector search models.
SLM CLI Agent
Terminal terminal assistant that translates natural language intentions into safe bash/zsh command line invocations offline.
SLM Email Assistant
Securely processes your incoming inbox streams. Auto-drafts contexts, filters spam, and extracts urgent action items on standard CPUs.
SLM Meeting Summarizer
Offline transcription post-processor. Distills meeting transcripts into action trackers, schedules, and bulleted logs with strict formatting rules.
SLM Voice Agent
Fast offline conversational companion combining local speech-to-text, edge neural reasoning, and lightweight CPU audio synthesis.
SLM Memory Manager
Manages long-term personal state and preference graphs. Learns and adapts to user query patterns locally without cloud synchronization.
SLM Task Planner
Autonomous goal decomposition system. Breaks complex tasks into prioritized action items and assigns them to specialized local sub-agents.
SLM PDF Chat
Securely parses complex PDF documents. Assembles layouts, reads tables, and lets you chat with local legal contracts, research articles, or receipts.
SLM PKB Agent
Local knowledge management assistant. Builds, links, and tags markdown documents in Obsidian, Notion, or Logseq vaults offline.
SLM JSON Cleaner
Real-time RFC 8259 syntax repair engine that normalizes corrupted webhooks, converts single quotes, strips comments, and standardizes snake_case keys.
SLM Document Parser
Universal multi-format parser supporting PDF, DOCX, XLSX, and TXT files with layout preservation, semantic chunking, and token metadata.
SLM Vision Parser
Local multimodal visual reasoning engine for chart transcription, diagram analysis, and image table extraction.
SLM Data Analyst
Domain-agnostic multi-sheet Excel & CSV statistical profiling engine that detects trends, aggregates data, and delivers direct analytical insights.
SLM Translation Hub
Local neural multilingual translation hub supporting Indian & global languages while preserving Markdown formatting, error codes, and JSON keys.
SLM Math Agent
Specialized arithmetic reasoning model. Handles math formulations, algebraic simplifications, and steps through complex equations offline.
SLM Security Audit
Comprehensive application security auditor that scans code, SQL queries, and endpoints for CWE vulnerabilities, injection vectors, and PII leaks.
SLM Embeddings Server
Local CPU-optimized dense vector embedding engine providing normalized vector representations and live cosine similarity without document uploads.
Blogs
Real empirical ablations, CPU optimization breakdowns, and architectural guides from the SLMAgents team.
Context Distillation for CPU RAG: A Controlled Study on Sub-3B Models
Why small local models fall apart with bloated context, controlled ablation benchmarks on CPU hardware, evaluation methodology on 120 held-out queries, and how hybrid RRF + 350-token distillation delivers 95.8% grounded accuracy.