attune-ai 2.0.0

AI collaboration framework with real LLM agent execution, AskUserQuestion tool integration, Socratic agent generation, progressive tier escalation (70-85% cost savings), meta-orchestration, dynamic agent composition (10 patterns including Anthropic-inspired), intelligent caching (85% hit rate), semantic workflow discovery, visual workflow editor, MCP integration for Claude Code, and multi-agent orchestration.

Empathy Framework

AI-powered developer workflows with cost optimization and pattern learning.

Run code review, debugging, testing, and release workflows from your terminal or Claude Code. Smart tier routing saves 34-86% on LLM costs.

pip install attune-ai[developer]

---

🎯 Transitioning to Claude-Native Architecture

Empathy Framework is evolving to focus exclusively on Anthropic/Claude to unlock features impossible with multi-provider abstraction:

• 📦 Prompt Caching: 90% cost reduction on repeated prompts
• 📖 Flexible Context: 200K via subscription for most tasks, up to 1M via API for large codebases
• 🧠 Extended Thinking: See Claude's internal reasoning process
• 🔧 Advanced Tool Use: Optimized for agentic workflows

Timeline:

• ✅ v4.8.0 (Jan 2026): Deprecation warnings for OpenAI/Google/Ollama providers
• ✅ v5.0.0 (Jan 26, 2026): Non-Anthropic providers removed (BREAKING - COMPLETE)
• ✅ v5.0.2 (Jan 28, 2026): Cost optimization suite with batch processing and caching monitoring

Migration Guide: docs/CLAUDE_NATIVE.md

---

What's New in v5.3.0

🎨 Dashboard Enhancements - Improved usability and clarity:

• Agent Display Names - Human-readable labels for agents in dashboard (e.g., "Code Analyzer" instead of UUID)
• Comprehensive Help Panel - 5-section accordion explaining dashboard features, use cases, and Redis setup
• UX Improvements - "Source Agent:" label clarity, "Redis Requires Enabling" status message
• Browser Cache Busting - Date-based versioning ensures updates appear immediately

📚 Documentation Improvements:

• Clarified flexible context strategy (200K subscription + 1M API routing)
• Added Redis requirement documentation for dashboard
• Root directory cleanup (8 archived files)

🧪 Test Infrastructure:

• Sequential test execution to fix import timing issues
• All agent tracking tests passing (19/19)

See Full Changelog

---

What's New in v5.1.0

🤖 Multi-Agent Orchestration - Full support for custom agents and Anthropic LLM agents:

• Agent Coordination Dashboard - Real-time monitoring with 6 coordination patterns:

  • Agent heartbeats and status tracking
  • Inter-agent coordination signals
  • Event streaming across agent workflows
  • Approval gates for human-in-the-loop
  • Quality feedback and performance metrics
  • Demo mode with test data generation

• Custom Agents - Build specialized agents for your workflow needs

• LLM Agents from Anthropic - Leverage Claude's advanced capabilities

• Dashboard accessible at http://localhost:8000 with python examples/dashboard_demo.py (Requires Redis)

🔐 Authentication Strategy System - Intelligent routing between Claude subscriptions and Anthropic API:

# Interactive setup
python -m attune.models.auth_cli setup

# View current configuration
python -m attune.models.auth_cli status

# Get recommendation for a file
python -m attune.models.auth_cli recommend src/module.py

💰 Automatic Cost Optimization - Workflows choose the best auth method:

• Small/medium modules (<2000 LOC) → Claude subscription (free)
• Large modules (>2000 LOC) → Anthropic API (pay for what you need)
• 7 workflows integrated: document-gen, test-gen, code-review, bug-predict, security-audit, perf-audit, release-prep
• Auth mode tracking in all workflow outputs for telemetry

🧪 Comprehensive Testing - 7 new integration tests for auth strategy:

• All workflows tested with auth enabled/disabled
• API and subscription mode verification
• Cost tracking validation

📖 Documentation - 950+ lines across 3 guides:

• AUTH_STRATEGY_GUIDE.md - User guide for configuration
• AUTH_CLI_IMPLEMENTATION.md - CLI command reference
• AUTH_WORKFLOW_INTEGRATIONS.md - Integration patterns

See Full Changelog

---

What's New in v5.0.2

💰 50% Cost Savings with Batch API - Process non-urgent tasks asynchronously:

empathy batch submit batch_requests.json  # Submit batch job
empathy batch status msgbatch_abc123      # Check progress
empathy batch results msgbatch_abc123 output.json  # Download results

Perfect for: log analysis, report generation, bulk classification, test generation

📊 Precise Token Counting - >98% accurate cost tracking:

• Integrated Anthropic's count_tokens() API for billing-accurate measurements
• 3-tier fallback: API → tiktoken (local) → heuristic
• Cache-aware cost calculation (25% write markup, 90% read discount)

📈 Cache Performance Monitoring - Track your 20-30% caching savings:

empathy cache stats           # Show hit rates and cost savings
empathy cache stats --verbose # Detailed token metrics
empathy cache stats --format json  # Machine-readable output

🧭 Adaptive Routing Analytics - Intelligent tier recommendations:

empathy routing stats <workflow>    # Performance metrics
empathy routing check --all         # Tier upgrade recommendations
empathy routing models --provider anthropic  # Compare models

🔧 Dashboard Fixes - All 6 agent coordination patterns now operational:

• Agent heartbeats displaying correctly
• Event streaming functional
• Coordination signals working
• Approval gates operational

See Full Changelog | Batch API Guide | User API Docs

---

What's New in v4.9.0

⚡ 18x Faster Performance - Massive performance gains through Phase 2 optimizations:

• Redis Two-Tier Caching: 2x faster memory operations (37,000x for cached keys)
• Generator Expressions: 99.9% memory reduction across 27 optimizations
• Parallel Scanning: Multi-core processing enabled by default (2-4x faster)
• Incremental Scanning: Git diff-based updates (10x faster)

🧭 Natural Language Workflows - Use plain English instead of workflow names:

/workflows "find security vulnerabilities"  # → security-audit
/workflows "check code performance"         # → perf-audit
/workflows "predict bugs"                   # → bug-predict
/plan "review my code"                      # → code-review

📊 Real-World Performance:

• Combined workflow: 3.59s → 0.2s (18x faster)
• Full scan: 3,472 files in 0.98s (was 3.59s)
• Redis cached operations: 37ms → 0.001ms

🎯 Improved Navigation:

• Split /workflow into /workflows (automated analysis) and /plan (planning/review)
• Clearer hub organization with better categorization
• Natural language routing matches intent to workflow

See CHANGELOG.md | Performance Docs

---

What's New in v4.7.0

$0 Workflows via Skills - Multi-agent workflows run through Claude Code's Task tool instead of API calls. No additional cost with your Claude subscription.

Socratic Workflows - Interactive discovery through guided questions. Workflows ask what you need rather than requiring upfront configuration.

Security Hardened - Fixed critical vulnerabilities (path traversal, JWT, SSRF).

Hub-Based Commands - Organized workflows into intuitive command hubs.

---

Quick Start

1. Install

pip install attune-ai[developer]

2. Configure

# Auto-detect API keys
python -m attune.models.cli provider

# Or set explicitly
python -m attune.models.cli provider --set anthropic

3. Use

In Claude Code:

/dev           # Developer tools (debug, commit, PR, review)
/testing       # Run tests, coverage, benchmarks
/workflows     # Automated analysis (security, bugs, perf)
/plan          # Planning, TDD, code review
/docs          # Documentation generation
/release       # Release preparation

# Natural language support:
/workflows "find security issues"
/plan "review my code"

# Direct tool access via MCP (v5.1.1+):
# Claude Code automatically discovers Empathy tools through the MCP server
# Just describe what you need in natural language:
"Run a security audit on src/"          → Invokes security_audit tool
"Generate tests for config.py"          → Invokes test_generation tool
"Check my auth configuration"           → Invokes auth_status tool
"Analyze performance bottlenecks"       → Invokes performance_audit tool

MCP Server Integration (v5.1.1+):

Empathy Framework now includes a Model Context Protocol (MCP) server that exposes all workflows as native Claude Code tools:

• 10 Tools Available: security_audit, bug_predict, code_review, test_generation, performance_audit, release_prep, auth_status, auth_recommend, telemetry_stats, dashboard_status
• Automatic Discovery: No manual configuration needed - Claude Code finds tools via .claude/mcp.json
• Natural Language Access: Describe your need and Claude invokes the appropriate tool
• Verification Hooks: Automatic validation of Python/JSON files and workflow outputs

To verify MCP integration:

# Check server is running
echo '{"method":"tools/list","params":{}}' | PYTHONPATH=./src python -m attune.mcp.server

# Restart Claude Code to load the MCP server
# Tools will appear in Claude's tool list automatically

See .claude/MCP_TEST_RESULTS.md for full integration details.

CLI:

empathy workflow run security-audit --path ./src
empathy workflow run test-coverage --target 90
empathy telemetry show  # View cost savings

Python:

from attune import EmpathyOS

async with EmpathyOS() as empathy:
    result = await empathy.level_2_guided(
        "Review this code for security issues"
    )
    print(result["response"])

---

Command Hubs

Workflows are organized into hubs for easy discovery:

Hub	Command	Description
Developer	/dev	Debug, commit, PR, code review, quality
Testing	/testing	Run tests, coverage analysis, benchmarks
Documentation	/docs	Generate and manage documentation
Release	/release	Release prep, security scan, publishing
Workflows	/workflows	Automated analysis (security, bugs, perf)
Plan	/plan	Planning, TDD, code review, refactoring
Utilities	/utilities	Project init, dependencies, profiling
Learning	/learning	Pattern learning and session evaluation
Context	/context	State management and memory
Agent	/agent	Create and manage custom agents

Natural Language Support:

# Use plain English - intelligent routing matches your intent
/workflows "find security vulnerabilities"  # → security-audit
/workflows "check code performance"         # → perf-audit
/workflows "predict bugs"                   # → bug-predict
/plan "review my code"                      # → code-review
/plan "help me plan this feature"           # → planning

# Or use traditional workflow names
/workflows security-audit
/plan code-review

Interactive menus:

/dev                    # Show interactive menu
/dev "debug auth error" # Jump directly to debugging
/testing "run coverage" # Run coverage analysis
/release                # Start release preparation

---

Socratic Method

Workflows guide you through discovery instead of requiring upfront configuration:

You: /dev

Claude: What development task do you need?
  1. Debug issue
  2. Create commit
  3. PR workflow
  4. Quality check

You: 1

Claude: What error or unexpected behavior are you seeing?

How it works:

1. Discovery - Workflow asks targeted questions to understand your needs
2. Context gathering - Collects relevant code, errors, and constraints
3. Dynamic agent creation - Assembles the right team based on your answers
4. Execution - Runs with appropriate tier selection

Create custom agents with Socratic guidance:

/agent create    # Guided agent creation
/agent team      # Build multi-agent teams interactively

---

Cost Optimization

Skills = $0 (Claude Code)

When using Claude Code, workflows run as skills through the Task tool - no API costs:

/dev           # $0 - uses your Claude subscription
/testing       # $0
/release       # $0
/agent create  # $0

API Mode (CI/CD, Automation)

For programmatic use, smart tier routing saves 34-86%:

Tier	Model	Use Case	Cost
CHEAP	Haiku / GPT-4o-mini	Formatting, simple tasks	~$0.005
CAPABLE	Sonnet / GPT-4o	Bug fixes, code review	~$0.08
PREMIUM	Opus / o1	Architecture, complex design	~$0.45

# Track API usage and savings
empathy telemetry savings --days 30

---

Key Features

Multi-Agent Workflows

# 4 parallel agents check release readiness
empathy orchestrate release-prep

# Sequential coverage improvement
empathy orchestrate test-coverage --target 90

Response Caching

Up to 57% cache hit rate on similar prompts. Zero config needed.

from attune.workflows import SecurityAuditWorkflow

workflow = SecurityAuditWorkflow(enable_cache=True)
result = await workflow.execute(target_path="./src")
print(f"Cache hit rate: {result.cost_report.cache_hit_rate:.1f}%")

Pattern Learning

Workflows learn from outcomes and improve over time:

from attune.orchestration.config_store import ConfigurationStore

store = ConfigurationStore()
best = store.get_best_for_task("release_prep")
print(f"Success rate: {best.success_rate:.1%}")

Multi-Provider Support

from attune_llm.providers import (
    AnthropicProvider,  # Claude
    OpenAIProvider,     # GPT-4
    GeminiProvider,     # Gemini
    LocalProvider,      # Ollama, LM Studio
)

---

CLI Reference

# Provider configuration
python -m attune.models.cli provider
python -m attune.models.cli provider --set hybrid

# Workflows
empathy workflow list
empathy workflow run <workflow-name>

# Cost tracking
empathy telemetry show
empathy telemetry savings --days 30
empathy telemetry export --format csv

# Orchestration
empathy orchestrate release-prep
empathy orchestrate test-coverage --target 90

# Meta-workflows
empathy meta-workflow list
empathy meta-workflow run release-prep --real

---

Install Options

# Individual developers (recommended)
pip install attune-ai[developer]

# All LLM providers
pip install attune-ai[llm]

# With caching (semantic similarity)
pip install attune-ai[cache]

# Enterprise (auth, rate limiting)
pip install attune-ai[enterprise]

# Healthcare (HIPAA compliance)
pip install attune-ai[healthcare]

# Development
git clone https://github.com/Smart-AI-Memory/attune-ai.git
cd attune-ai && pip install -e .[dev]

---

Environment Setup

# At least one provider required
export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."
export GOOGLE_API_KEY="..."

# Optional (but required for Agent Dashboard): Redis for memory
export REDIS_URL="redis://localhost:6379"

---

VSCode Extension

Install the Empathy VSCode extension for:

• Dashboard - Health score, costs, patterns
• One-Click Workflows - Run from command palette
• Memory Panel - Manage Redis and patterns
• Cost Tracking - Real-time savings display

---

Documentation

• Quick Start Guide
• CLI Reference
• Testing Guide
• Keyboard Shortcuts
• Full Documentation

---

Security

• Path traversal protection on all file operations
• JWT authentication with rate limiting
• PII scrubbing in telemetry
• HIPAA/GDPR compliance options
• Automated security scanning with 82% accuracy (Phase 3 AST-based detection)

See SECURITY.md for vulnerability reporting.

Security Scanning

Automated security scanning in CI/CD - 82% accuracy, blocks critical issues:

# Run security audit locally
empathy workflow run security-audit

# Scan specific directory
empathy workflow run security-audit --input '{"path":"./src"}'

Documentation:

• Developer Workflow Guide - Quick reference for handling security findings (all developers)
• CI/CD Integration Guide - Complete setup and troubleshooting (DevOps, developers)
• Scanner Architecture - Technical implementation details (engineers, architects)
• Remediation Process - 3-phase methodology for improving scanners (security teams, leadership)
• API Reference - Complete API documentation (developers extending scanner)

Key achievements:

• 82.3% reduction in false positives (350 → 62 findings)
• 16x improvement in scanner accuracy
• <15 minute average fix time for critical issues
• Zero critical vulnerabilities in production code

---

Contributing

See CONTRIBUTING.md for guidelines.

---

License

Apache License 2.0 - Free and open source for everyone. Use it, modify it, build commercial products with it. Details →

---

Acknowledgements

This project stands on the shoulders of giants. We are deeply grateful to the open source community and all the amazing projects that make this framework possible.

View Full Acknowledgements →

Special thanks to:

• Anthropic - For Claude AI and the Model Context Protocol
• LangChain - Agent framework powering our meta-orchestration
• FastAPI - Modern Python web framework
• pytest - Testing framework making quality assurance effortless

And to all 50+ open source projects we depend on. See the complete list →

Want to contribute? See CONTRIBUTORS.md

---

Built by Smart AI Memory · Docs · Examples · Issues